4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
37 #include <linux/posix_acl.h>
38 #include <linux/kobject.h>
39 #include <linux/spinlock.h>
40 #include <linux/sysfs.h>
41 #include <linux/xarray.h>
43 #include <uapi/linux/lustre/lustre_idl.h>
44 #include <lustre_lib.h>
45 #include <libcfs/bitmap.h>
46 #ifdef HAVE_SERVER_SUPPORT
47 # include <lu_target.h>
48 # include <obd_target.h>
51 #include <lustre_export.h>
52 #include <lustre_fid.h>
53 #include <lustre_fld.h>
54 #include <lustre_handles.h>
55 #include <lustre_intent.h>
57 #include <lustre_quota.h>
59 #define MAX_OBD_DEVICES 8192
67 struct lov_oinfo { /* per-stripe data structure */
68 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
69 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
70 int loi_ost_gen; /* generation of this loi_ost_idx */
72 unsigned long loi_kms_valid:1;
73 __u64 loi_kms; /* known minimum size */
74 struct ost_lvb loi_lvb;
75 struct osc_async_rc loi_ar;
78 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
81 oinfo->loi_kms_valid = 1;
87 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
89 /* obd info for a particular level (lov, osc). */
91 /* OBD_STATFS_* flags */
93 struct obd_device *oi_obd;
94 struct lu_tgt_desc *oi_tgt;
95 /* statfs data specific for every OSC, if needed at all. */
96 struct obd_statfs *oi_osfs;
97 /* An update callback which is called to update some data on upper
98 * level. E.g. it is used for update lsm->lsm_oinfo at every received
99 * request in osc level for enqueue requests. It is also possible to
100 * update some caller data from LOV layer if needed. */
101 obd_enqueue_update_f oi_cb_up;
105 const struct obd_ops *typ_dt_ops;
106 const struct md_ops *typ_md_ops;
107 struct proc_dir_entry *typ_procroot;
108 struct dentry *typ_debugfs_entry;
109 #ifdef HAVE_SERVER_SUPPORT
113 struct lu_device_type *typ_lu;
114 struct kobject typ_kobj;
116 #define typ_name typ_kobj.name
117 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
124 /* used for encryption: difference with offset in clear text page */
126 /* used for encryption: difference with count in clear text page */
131 struct timeout_item {
132 enum timeout_event ti_event;
136 struct list_head ti_obd_list;
137 struct list_head ti_chain;
140 #define OBD_MAX_RIF_DEFAULT 8
141 #define OBD_MAX_RIF_MAX 512
142 #define OSC_MAX_RIF_MAX 256
143 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
144 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
145 #define OSC_DEFAULT_RESENDS 10
147 /* possible values for lut_sync_lock_cancel */
148 enum tgt_sync_lock_cancel {
149 SYNC_LOCK_CANCEL_NEVER = 0,
150 SYNC_LOCK_CANCEL_BLOCKING = 1,
151 SYNC_LOCK_CANCEL_ALWAYS = 2,
155 * Limit reply buffer size for striping data to one x86_64 page. This
156 * value is chosen to fit the striping data for common use cases while
157 * staying well below the limit at which the buffer must be backed by
158 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
161 #define OBD_MAX_DEFAULT_EA_SIZE 4096
164 * Lustre can handle larger xattrs internally, but we must respect the Linux
165 * VFS limitation or tools like tar cannot interact with Lustre volumes
168 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
171 enum obd_cl_sem_lock_class {
180 struct rw_semaphore cl_sem;
181 struct obd_uuid cl_target_uuid;
182 struct obd_import *cl_import; /* ptlrpc connection state */
183 size_t cl_conn_count;
185 /* Cache maximum and default values for easize. This is
186 * strictly a performance optimization to minimize calls to
187 * obd_size_diskmd(). The default values are used to calculate the
188 * initial size of a request buffer. The ptlrpc layer will resize the
189 * buffer as needed to accommodate a larger reply from the
190 * server. The default values should be small enough to avoid wasted
191 * memory and excessive use of vmalloc(), yet large enough to avoid
192 * reallocating the buffer in the common use case. */
194 /* Default EA size for striping attributes. It is initialized at
195 * mount-time based on the default stripe width of the filesystem,
196 * then it tracks the largest observed EA size advertised by
197 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
198 __u32 cl_default_mds_easize;
200 /* Maximum possible EA size computed at mount-time based on
201 * the number of OSTs in the filesystem. May be increased at
202 * run-time if a larger observed size is advertised by the MDT. */
203 __u32 cl_max_mds_easize;
205 /* Data-on-MDT specific value to set larger reply buffer for possible
206 * data read along with open/stat requests. By default it tries to use
207 * unused space in reply buffer.
208 * This value is used to ensure that reply buffer has at least as
209 * much free space as value indicates. That free space is gained from
210 * LOV EA buffer which is small for DoM files and on big systems can
211 * provide up to 32KB of extra space in reply buffer.
212 * Default value is 8K now.
214 __u32 cl_dom_min_inline_repsize;
216 enum lustre_sec_part cl_sp_me;
217 enum lustre_sec_part cl_sp_to;
218 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
220 /* the grant values are protected by loi_list_lock below */
221 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
222 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
223 unsigned long cl_avail_grant; /* bytes of credit for ost */
224 unsigned long cl_lost_grant; /* lost credits (trunc) */
225 /* grant consumed for dirty pages */
226 unsigned long cl_dirty_grant;
228 /* since we allocate grant by blocks, we don't know how many grant will
229 * be used to add a page into cache. As a solution, we reserve maximum
230 * grant before trying to dirty a page and unreserve the rest.
231 * See osc_{reserve|unreserve}_grant for details. */
232 long cl_reserved_grant;
233 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
234 time64_t cl_next_shrink_grant; /* seconds */
235 struct list_head cl_grant_chain;
236 time64_t cl_grant_shrink_interval; /* seconds */
238 /* A chunk is an optimal size used by osc_extent to determine
239 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
241 /* extent insertion metadata overhead to be accounted in grant,
243 unsigned int cl_grant_extent_tax;
244 /* maximum extent size, in number of pages */
245 unsigned int cl_max_extent_pages;
247 /* keep track of objects that have lois that contain pages which
248 * have been queued for async brw. this lock also protects the
249 * lists of osc_client_pages that hang off of the loi */
251 * ->cl_loi_list_lock protects consistency of
252 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
253 * ->ap_completion() call-backs are executed under this lock. As we
254 * cannot guarantee that these call-backs never block on all platforms
255 * (as a matter of fact they do block on Mac OS X), type of
256 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
257 * and blocking mutex on Mac OS X. (Alternative is to make this lock
258 * blocking everywhere, but we don't want to slow down fast-path of
259 * our main platform.)
261 * NB by Jinshan: though field names are still _loi_, but actually
262 * osc_object{}s are in the list.
264 spinlock_t cl_loi_list_lock;
265 struct list_head cl_loi_ready_list;
266 struct list_head cl_loi_hp_ready_list;
267 struct list_head cl_loi_write_list;
268 struct list_head cl_loi_read_list;
269 __u32 cl_r_in_flight;
270 __u32 cl_w_in_flight;
271 /* just a sum of the loi/lop pending numbers to be exported by /proc */
272 atomic_t cl_pending_w_pages;
273 atomic_t cl_pending_r_pages;
274 u32 cl_max_pages_per_rpc;
275 u32 cl_max_rpcs_in_flight;
276 u32 cl_max_short_io_bytes;
277 struct obd_histogram cl_read_rpc_hist;
278 struct obd_histogram cl_write_rpc_hist;
279 struct obd_histogram cl_read_page_hist;
280 struct obd_histogram cl_write_page_hist;
281 struct obd_histogram cl_read_offset_hist;
282 struct obd_histogram cl_write_offset_hist;
284 /** LRU for osc caching pages */
285 struct cl_client_cache *cl_cache;
286 /** member of cl_cache->ccc_lru */
287 struct list_head cl_lru_osc;
288 /** # of available LRU slots left in the per-OSC cache.
289 * Available LRU slots are shared by all OSCs of the same file system,
290 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
291 atomic_long_t *cl_lru_left;
292 /** # of busy LRU pages. A page is considered busy if it's in writeback
293 * queue, or in transfer. Busy pages can't be discarded so they are not
295 atomic_long_t cl_lru_busy;
296 /** # of LRU pages in the cache for this client_obd */
297 atomic_long_t cl_lru_in_list;
298 /** # of threads are shrinking LRU cache. To avoid contention, it's not
299 * allowed to have multiple threads shrinking LRU cache. */
300 atomic_t cl_lru_shrinkers;
301 /** The time when this LRU cache was last used. */
302 time64_t cl_lru_last_used;
303 /** stats: how many reclaims have happened for this client_obd.
304 * reclaim and shrink - shrink is async, voluntarily rebalancing;
305 * reclaim is sync, initiated by IO thread when the LRU slots are
307 __u64 cl_lru_reclaim;
308 /** List of LRU pages for this client_obd */
309 struct list_head cl_lru_list;
310 /** Lock for LRU page list */
311 spinlock_t cl_lru_list_lock;
312 /** # of unstable pages in this client_obd.
313 * An unstable page is a page state that WRITE RPC has finished but
314 * the transaction has NOT yet committed. */
315 atomic_long_t cl_unstable_count;
316 /** Link to osc_shrinker_list */
317 struct list_head cl_shrink_list;
319 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
320 atomic_t cl_destroy_in_flight;
321 wait_queue_head_t cl_destroy_waitq;
323 /* modify rpcs in flight
324 * currently used for metadata only */
325 spinlock_t cl_mod_rpcs_lock;
326 __u16 cl_max_mod_rpcs_in_flight;
327 __u16 cl_mod_rpcs_in_flight;
328 __u16 cl_close_rpcs_in_flight;
329 wait_queue_head_t cl_mod_rpcs_waitq;
330 unsigned long *cl_mod_tag_bitmap;
331 struct obd_histogram cl_mod_rpcs_hist;
334 struct mutex cl_mgc_mutex;
335 struct local_oid_storage *cl_mgc_los;
336 struct dt_object *cl_mgc_configs_dir;
337 struct obd_export *cl_mgc_mgsexp;
338 atomic_t cl_mgc_refcount;
339 /* in-flight control list and total RPCs counter */
340 struct list_head cl_flight_waiters;
341 __u32 cl_rpcs_in_flight;
343 /* checksumming for data sent over the network */
344 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
345 cl_checksum_dump:1; /* same */
346 /* supported checksum types that are worked out at connect time */
347 __u32 cl_supp_cksum_types;
348 /* checksum algorithm to be used */
349 enum cksum_types cl_cksum_type;
350 /* preferred checksum algorithm to be used */
351 enum cksum_types cl_preferred_cksum_type;
353 /* also protected by the poorly named _loi_list_lock lock above */
354 struct osc_async_rc cl_ar;
356 /* sequence manager */
357 struct lu_client_seq *cl_seq;
358 struct rw_semaphore cl_seq_rwsem;
360 atomic_t cl_resends; /* resend count */
362 /* ptlrpc work for writeback in ptlrpcd context */
363 void *cl_writeback_work;
365 struct mutex cl_quota_mutex;
366 /* hash tables for osc_quota_info */
367 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
368 /* the xid of the request updating the hash tables */
369 __u64 cl_quota_last_xid;
370 /* Links to the global list of registered changelog devices */
371 struct list_head cl_chg_dev_linkage;
373 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
380 struct echo_client_obd {
381 struct obd_export *ec_exp; /* the local connection to osc/lov */
383 struct list_head ec_objects;
384 struct list_head ec_locks;
388 /* allow statfs data caching for 1 second */
389 #define OBD_STATFS_CACHE_SECONDS 1
390 /* arbitrary maximum. larger would be useless, allows catching bogus input */
391 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
393 #define lov_tgt_desc lu_tgt_desc
395 struct lov_md_tgt_desc {
396 struct obd_device *lmtd_mdc;
401 struct lov_desc desc;
402 struct lov_tgt_desc **lov_tgts; /* sparse array */
403 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
405 struct mutex lov_lock;
406 struct obd_connect_data lov_ocd;
407 struct proc_dir_entry *targets_proc_entry;
408 atomic_t lov_refcount;
409 __u32 lov_death_row; /* tgts scheduled to be deleted */
410 __u32 lov_tgt_size; /* size of tgts array */
413 struct rhashtable lov_pools_hash_body; /* used for key access */
414 struct list_head lov_pool_list; /* used for sequential access */
415 struct proc_dir_entry *lov_pool_proc_entry;
416 enum lustre_sec_part lov_sp_me;
418 /* Cached LRU and unstable data from upper layer */
419 struct cl_client_cache *lov_cache;
421 struct rw_semaphore lov_notify_lock;
422 /* Data-on-MDT: MDC array */
423 struct lov_md_tgt_desc *lov_mdc_tgts;
425 struct kobject *lov_tgts_kobj;
428 #define lmv_tgt_desc lu_tgt_desc
431 struct lu_client_fld lmv_fld;
437 u32 lmv_statfs_start;
439 struct lu_tgt_descs lmv_mdt_descs;
441 struct obd_connect_data conn_data;
442 struct kobject *lmv_tgts_kobj;
445 __u32 lmv_qos_rr_index;
448 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
449 #define lmv_qos lmv_mdt_descs.ltd_qos
451 /* Minimum sector size is 512 */
452 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
454 struct niobuf_local {
455 __u64 lnb_file_offset;
456 __u32 lnb_page_offset;
460 struct page *lnb_page;
462 __u16 lnb_guards[MAX_GUARD_NUMBER];
463 __u16 lnb_guard_rpc:1;
464 __u16 lnb_guard_disk:1;
465 /* separate unlock for read path to allow shared access */
469 struct tgt_thread_big_cache {
470 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
473 #define LUSTRE_FLD_NAME "fld"
474 #define LUSTRE_SEQ_NAME "seq"
476 #define LUSTRE_MDD_NAME "mdd"
477 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
478 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
479 #define LUSTRE_VVP_NAME "vvp"
480 #define LUSTRE_LMV_NAME "lmv"
481 #define LUSTRE_SLP_NAME "slp"
482 #define LUSTRE_LOD_NAME "lod"
483 #define LUSTRE_OSP_NAME "osp"
484 #define LUSTRE_LWP_NAME "lwp"
486 /* obd device type names */
487 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
488 #define LUSTRE_MDS_NAME "mds"
489 #define LUSTRE_MDT_NAME "mdt"
490 #define LUSTRE_MDC_NAME "mdc"
491 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
492 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
493 #define LUSTRE_OSC_NAME "osc"
494 #define LUSTRE_LOV_NAME "lov"
495 #define LUSTRE_MGS_NAME "mgs"
496 #define LUSTRE_MGC_NAME "mgc"
498 #define LUSTRE_ECHO_NAME "obdecho"
499 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
500 #define LUSTRE_QMT_NAME "qmt"
502 /* Constant obd names (post-rename) */
503 #define LUSTRE_MDS_OBDNAME "MDS"
504 #define LUSTRE_OSS_OBDNAME "OSS"
505 #define LUSTRE_MGS_OBDNAME "MGS"
506 #define LUSTRE_MGC_OBDNAME "MGC"
508 static inline int is_lwp_on_mdt(char *name)
512 ptr = strrchr(name, '-');
514 CERROR("%s is not a obdname\n", name);
518 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
520 if (strncmp(ptr + 1, "MDT", 3) != 0)
523 while (*(--ptr) != '-' && ptr != name);
528 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
534 static inline int is_lwp_on_ost(char *name)
538 ptr = strrchr(name, '-');
540 CERROR("%s is not a obdname\n", name);
544 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
546 if (strncmp(ptr + 1, "OST", 3) != 0)
549 while (*(--ptr) != '-' && ptr != name);
554 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
561 * Events signalled through obd_notify() upcall-chain.
563 enum obd_notify_event {
564 /* Device connect start */
566 /* Device activated */
568 /* Device deactivated */
570 /* Connect data for import were changed */
572 /* Administratively deactivate/activate event */
573 OBD_NOTIFY_DEACTIVATE,
578 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
579 * being main example).
581 struct obd_notify_upcall {
582 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
583 enum obd_notify_event ev, void *owner);
584 /* Opaque datum supplied by upper layer listener */
588 struct target_recovery_data {
589 svc_handler_t trd_recovery_handler;
590 pid_t trd_processing_task;
591 struct completion trd_starting;
592 struct completion trd_finishing;
595 struct obd_llog_group {
596 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
597 wait_queue_head_t olg_waitq;
601 /* corresponds to one of the obd's */
602 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
605 struct obd_type *obd_type;
606 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
607 int obd_minor; /* device number: lctl dl */
608 struct lu_device *obd_lu_dev;
610 /* common and UUID name of this device */
611 struct obd_uuid obd_uuid;
612 char obd_name[MAX_OBD_NAME];
614 /* bitfield modification is protected by obd_dev_lock */
616 obd_attached:1, /* finished attach */
617 obd_set_up:1, /* finished setup */
618 obd_recovering:1, /* there are recoverable clients */
619 obd_abort_recovery:1, /* recovery expired */
620 obd_version_recov:1, /* obd uses version checking */
621 obd_replayable:1, /* recovery enabled; inform clients */
622 obd_no_recov:1, /* fail instead of retry messages */
623 obd_stopping:1, /* started cleanup */
624 obd_starting:1, /* started setup */
625 obd_force:1, /* cleanup with > 0 obd refcount */
626 obd_fail:1, /* cleanup with failover */
627 obd_no_conn:1, /* deny new connections */
628 obd_inactive:1, /* device active/inactive
629 * (for /proc/status only!!) */
630 obd_no_ir:1, /* no imperative recovery. */
631 obd_process_conf:1, /* device is processing mgs config */
632 obd_checksum_dump:1; /* dump pages upon cksum error */
633 #ifdef HAVE_SERVER_SUPPORT
634 /* no committed-transno notification */
635 unsigned long obd_no_transno:1;
638 /* use separate field as it is set in interrupt to don't mess with
639 * protection of other bits using _bh lock */
640 unsigned long obd_recovery_expired:1;
641 /* uuid-export hash body */
642 struct rhashtable obd_uuid_hash;
643 /* nid-export hash body */
644 struct rhltable obd_nid_hash;
646 struct cfs_hash *obd_nid_stats_hash;
647 /* client_generation-export hash body */
648 struct cfs_hash *obd_gen_hash;
649 struct list_head obd_nid_stats;
650 struct list_head obd_exports;
651 struct list_head obd_unlinked_exports;
652 struct list_head obd_delayed_exports;
653 struct list_head obd_lwp_list;
654 atomic_t obd_refcount;
656 spinlock_t obd_nid_lock;
657 struct ldlm_namespace *obd_namespace;
658 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
659 /* a spinlock is OK for what we do now, may need a semaphore later */
660 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
661 spinlock_t obd_osfs_lock;
662 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
663 time64_t obd_osfs_age;
664 __u64 obd_last_committed;
665 struct mutex obd_dev_mutex;
666 struct lvfs_run_ctxt obd_lvfs_ctxt;
667 struct obd_llog_group obd_olg; /* default llog group */
668 struct obd_device *obd_observer;
669 struct rw_semaphore obd_observer_link_sem;
670 struct obd_notify_upcall obd_upcall;
671 struct obd_export *obd_self_export;
672 struct obd_export *obd_lwp_export;
673 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
674 struct list_head obd_exports_timed;
675 time64_t obd_eviction_timer; /* for ping evictor */
677 atomic_t obd_max_recoverable_clients;
678 atomic_t obd_connected_clients;
679 int obd_stale_clients;
680 /* this lock protects all recovery list_heads, timer and
681 * obd_next_recovery_transno value */
682 spinlock_t obd_recovery_task_lock;
683 __u64 obd_next_recovery_transno;
684 int obd_replayed_requests;
685 int obd_requests_queued_for_recovery;
686 wait_queue_head_t obd_next_transno_waitq;
687 /* protected by obd_recovery_task_lock */
688 struct hrtimer obd_recovery_timer;
690 time64_t obd_recovery_start;
691 /* seconds, for lprocfs_status */
692 time64_t obd_recovery_end;
693 /* To tell timeouts from time stamps Lustre uses timeout_t
694 * instead of time64_t.
696 timeout_t obd_recovery_time_hard;
697 timeout_t obd_recovery_timeout;
698 int obd_recovery_ir_factor;
700 /* new recovery stuff from CMD2 */
701 int obd_replayed_locks;
702 atomic_t obd_req_replay_clients;
703 atomic_t obd_lock_replay_clients;
704 struct target_recovery_data obd_recovery_data;
706 /* all lists are protected by obd_recovery_task_lock */
707 struct list_head obd_req_replay_queue;
708 struct list_head obd_lock_replay_queue;
709 struct list_head obd_final_req_queue;
712 #ifdef HAVE_SERVER_SUPPORT
713 struct obd_device_target obt;
714 struct filter_obd filter;
716 struct echo_obd echo;
718 struct client_obd cli;
719 struct echo_client_obd echo_client;
724 /* Fields used by LProcFS */
725 struct lprocfs_stats *obd_stats;
727 struct lprocfs_stats *obd_md_stats;
729 struct dentry *obd_debugfs_entry;
730 struct proc_dir_entry *obd_proc_entry;
731 struct proc_dir_entry *obd_proc_exports_entry;
732 struct dentry *obd_svc_debugfs_entry;
733 struct lprocfs_stats *obd_svc_stats;
734 const struct attribute **obd_attrs;
735 struct lprocfs_vars *obd_vars;
736 struct ldebugfs_vars *obd_debugfs_vars;
737 atomic_t obd_evict_inprogress;
738 wait_queue_head_t obd_evict_inprogress_waitq;
739 struct list_head obd_evict_list; /* protected with pet_lock */
742 * LDLM pool part. Save last calculated SLV and Limit.
744 rwlock_t obd_pool_lock;
748 int obd_conn_inprogress;
751 * List of outstanding class_incref()'s fo this OBD. For debugging. */
752 struct lu_ref obd_reference;
754 struct kset obd_kset; /* sysfs object collection */
755 struct kobj_type obd_ktype;
756 struct completion obd_kobj_unregister;
759 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
760 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
761 #ifdef HAVE_SERVER_SUPPORT
762 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
763 struct obd_uuid *uuid);
765 int obd_nid_export_for_each(struct obd_device *obd, lnet_nid_t nid,
766 int cb(struct obd_export *exp, void *data),
768 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
769 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
772 /* get/set_info keys */
773 #define KEY_ASYNC "async"
774 #define KEY_CHANGELOG_CLEAR "changelog_clear"
775 #define KEY_FID2PATH "fid2path"
776 #define KEY_CHECKSUM "checksum"
777 #define KEY_CLEAR_FS "clear_fs"
778 #define KEY_CONN_DATA "conn_data"
779 #define KEY_EVICT_BY_NID "evict_by_nid"
780 #define KEY_FIEMAP "fiemap"
781 #define KEY_FLUSH_CTX "flush_ctx"
782 #define KEY_GRANT_SHRINK "grant_shrink"
783 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
784 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
785 #define KEY_INTERMDS "inter_mds"
786 #define KEY_LAST_ID "last_id"
787 #define KEY_LAST_FID "last_fid"
788 #define KEY_MAX_EASIZE "max_easize"
789 #define KEY_DEFAULT_EASIZE "default_easize"
790 #define KEY_MGSSEC "mgssec"
791 #define KEY_READ_ONLY "read-only"
792 #define KEY_REGISTER_TARGET "register_target"
793 #define KEY_SET_FS "set_fs"
794 #define KEY_TGT_COUNT "tgt_count"
795 /* KEY_SET_INFO in lustre_idl.h */
796 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
798 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
799 #define KEY_OSP_CONNECTED "osp_connected"
801 /* Flags for op_xvalid */
803 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
804 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
805 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
806 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
807 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
808 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
809 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
814 static inline int it_to_lock_mode(struct lookup_intent *it)
816 /* CREAT needs to be tested before open (both could be set) */
817 if (it->it_op & IT_CREAT)
819 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
821 else if (it->it_op & IT_LAYOUT)
822 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
823 else if (it->it_op & IT_READDIR)
825 else if (it->it_op & IT_GETXATTR)
828 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
833 MF_MDC_CANCEL_FID1 = BIT(0),
834 MF_MDC_CANCEL_FID2 = BIT(1),
835 MF_MDC_CANCEL_FID3 = BIT(2),
836 MF_MDC_CANCEL_FID4 = BIT(3),
837 MF_GET_MDT_IDX = BIT(4),
841 CLI_SET_MEA = BIT(0),
842 CLI_RM_ENTRY = BIT(1),
845 CLI_MIGRATE = BIT(4),
846 CLI_DIRTY_DATA = BIT(5),
850 LUSTRE_OPC_MKDIR = 0,
851 LUSTRE_OPC_SYMLINK = 1,
852 LUSTRE_OPC_MKNOD = 2,
853 LUSTRE_OPC_CREATE = 3,
858 * GETXATTR is not included as only a couple of fields in the reply body
859 * is filled, but not FID which is needed for common intent handling in
860 * mdc_finish_intent_lock()
862 static inline bool it_has_reply_body(const struct lookup_intent *it)
864 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
868 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
869 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
870 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
871 struct lu_fid op_fid4; /* to the operation locks. */
872 u32 op_mds; /* what mds server open will go to */
874 enum md_op_code op_code;
875 struct lustre_handle op_open_handle;
879 struct rw_semaphore *op_mea1_sem;
880 struct rw_semaphore *op_mea2_sem;
881 struct lmv_stripe_md *op_mea1;
882 struct lmv_stripe_md *op_mea2;
883 struct lmv_stripe_md *op_default_mea1; /* default LMV */
884 __u32 op_suppgids[2];
891 /* iattr fields and blocks. */
892 struct iattr op_attr;
893 enum op_xvalid op_xvalid; /* eXtra validity flags */
894 loff_t op_attr_blocks;
895 u64 op_valid; /* OBD_MD_* */
896 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
898 enum md_op_flags op_flags;
900 /* Various operation flags. */
901 enum mds_op_bias op_bias;
903 /* used to transfer info between the stacks of MD client
904 * see enum op_cli_flags */
905 enum md_cli_flags op_cli_flags;
907 /* File object data version for HSM release, on client */
908 __u64 op_data_version;
909 struct lustre_handle op_lease_handle;
911 /* File security context, for creates/metadata ops */
912 const char *op_file_secctx_name;
913 __u32 op_file_secctx_name_size;
914 void *op_file_secctx;
915 __u32 op_file_secctx_size;
917 /* File encryption context, for creates/metadata ops */
918 void *op_file_encctx;
919 __u32 op_file_encctx_size;
923 /* Used by readdir */
924 unsigned int op_max_pages;
929 * used to access dir that is changing layout: if it's set, access
930 * dir by new layout, otherwise old layout.
931 * By default it's not set, because new files are created under new
932 * layout, if we can't find file with name under both old and new
933 * layout, we are sure file with name doesn't exist, but in reverse
934 * order there may be a race with creation by others.
937 /* used to access dir with bash hash */
938 __u32 op_stripe_index;
939 /* Archive ID for PCC attach */
944 int (*md_blocking_ast)(struct ldlm_lock *lock,
945 struct ldlm_lock_desc *desc,
946 void *data, int flag);
949 struct md_enqueue_info;
950 /* metadata stat-ahead */
951 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
952 struct md_enqueue_info *minfo,
955 struct md_enqueue_info {
956 struct md_op_data mi_data;
957 struct lookup_intent mi_it;
958 struct lustre_handle mi_lockh;
959 struct inode *mi_dir;
960 struct ldlm_enqueue_info mi_einfo;
961 md_enqueue_cb_t mi_cb;
966 struct module *o_owner;
967 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
968 void *karg, void __user *uarg);
969 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
970 __u32 keylen, void *key, __u32 *vallen, void *val);
971 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
972 __u32 keylen, void *key,
973 __u32 vallen, void *val,
974 struct ptlrpc_request_set *set);
975 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
976 int (*o_precleanup)(struct obd_device *obd);
977 int (*o_cleanup)(struct obd_device *obd);
978 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
979 int (*o_postrecov)(struct obd_device *obd);
980 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
982 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
983 /* connect to the target device with given connection
984 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
985 * granted by the target, which are guaranteed to be a subset of flags
986 * asked for. If @ocd == NULL, use default parameters. */
987 int (*o_connect)(const struct lu_env *env,
988 struct obd_export **exp, struct obd_device *src,
989 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
991 int (*o_reconnect)(const struct lu_env *env,
992 struct obd_export *exp, struct obd_device *src,
993 struct obd_uuid *cluuid,
994 struct obd_connect_data *ocd,
996 int (*o_disconnect)(struct obd_export *exp);
998 /* Initialize/finalize fids infrastructure. */
999 int (*o_fid_init)(struct obd_device *obd,
1000 struct obd_export *exp, enum lu_cli_type type);
1001 int (*o_fid_fini)(struct obd_device *obd);
1003 /* Allocate new fid according to passed @hint. */
1004 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1005 struct lu_fid *fid, struct md_op_data *op_data);
1008 * Object with @fid is getting deleted, we may want to do something
1011 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1012 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1013 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1014 time64_t max_age, struct ptlrpc_request_set *set);
1015 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1017 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1019 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1021 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1023 int (*o_preprw)(const struct lu_env *env, int cmd,
1024 struct obd_export *exp, struct obdo *oa, int objcount,
1025 struct obd_ioobj *obj, struct niobuf_remote *remote,
1026 int *nr_pages, struct niobuf_local *local);
1027 int (*o_commitrw)(const struct lu_env *env, int cmd,
1028 struct obd_export *exp, struct obdo *oa,
1029 int objcount, struct obd_ioobj *obj,
1030 struct niobuf_remote *remote, int pages,
1031 struct niobuf_local *local, int rc);
1032 int (*o_init_export)(struct obd_export *exp);
1033 int (*o_destroy_export)(struct obd_export *exp);
1035 int (*o_import_event)(struct obd_device *, struct obd_import *,
1036 enum obd_import_event);
1038 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1039 enum obd_notify_event ev);
1041 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1042 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1045 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1046 struct obd_quotactl *);
1049 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1050 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1051 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1053 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1057 /* lmv structures */
1059 struct mdt_body *body;
1060 struct lu_buf layout;
1062 struct lmv_stripe_md *lmv;
1063 struct lmv_foreign_md *lfm;
1065 struct lmv_stripe_md *default_lmv;
1066 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1067 struct posix_acl *posix_acl;
1071 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1072 static inline void lmd_clear_acl(struct lustre_md *md)
1074 if (md->posix_acl) {
1075 posix_acl_release(md->posix_acl);
1076 md->posix_acl = NULL;
1080 #define OBD_CONNECT_ACL_FLAGS \
1081 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1083 static inline void lmd_clear_acl(struct lustre_md *md)
1087 #define OBD_CONNECT_ACL_FLAGS (0)
1090 struct md_open_data {
1091 struct obd_client_handle *mod_och;
1092 struct ptlrpc_request *mod_open_req;
1093 struct ptlrpc_request *mod_close_req;
1094 atomic_t mod_refcount;
1098 struct obd_client_handle {
1099 struct lustre_handle och_open_handle;
1100 struct lu_fid och_fid;
1101 struct md_open_data *och_mod;
1102 struct lustre_handle och_lease_handle; /* open lock for lease */
1107 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1109 struct lookup_intent;
1113 int (*m_close)(struct obd_export *, struct md_op_data *,
1114 struct md_open_data *, struct ptlrpc_request **);
1116 int (*m_create)(struct obd_export *, struct md_op_data *,
1117 const void *, size_t, umode_t, uid_t, gid_t,
1118 cfs_cap_t, __u64, struct ptlrpc_request **);
1120 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1121 const union ldlm_policy_data *, struct md_op_data *,
1122 struct lustre_handle *, __u64);
1124 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1125 struct ptlrpc_request **);
1127 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1128 struct lookup_intent *,
1129 struct ptlrpc_request **,
1130 ldlm_blocking_callback, __u64);
1132 int (*m_link)(struct obd_export *, struct md_op_data *,
1133 struct ptlrpc_request **);
1135 int (*m_rename)(struct obd_export *, struct md_op_data *,
1136 const char *, size_t, const char *, size_t,
1137 struct ptlrpc_request **);
1139 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1140 size_t , struct ptlrpc_request **);
1142 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1143 struct ptlrpc_request **);
1145 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1146 struct md_callback *cb_op, __u64 hash_offset,
1147 struct page **ppage);
1149 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1150 struct ptlrpc_request **);
1152 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1153 u64, const char *, const void *, size_t, unsigned int,
1154 u32, struct ptlrpc_request **);
1156 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1157 u64, const char *, size_t, struct ptlrpc_request **);
1159 int (*m_intent_getattr_async)(struct obd_export *,
1160 struct md_enqueue_info *);
1162 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1163 struct lu_fid *, __u64 *bits);
1165 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1167 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1168 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1170 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1171 struct ptlrpc_request **);
1173 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1175 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1176 struct obd_export *, struct obd_export *,
1177 struct lustre_md *);
1179 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1181 int (*m_merge_attr)(struct obd_export *, const struct lu_fid *fid,
1182 const struct lmv_stripe_md *lsm,
1183 struct cl_attr *attr, ldlm_blocking_callback);
1185 int (*m_set_open_replay_data)(struct obd_export *,
1186 struct obd_client_handle *,
1187 struct lookup_intent *);
1189 int (*m_clear_open_replay_data)(struct obd_export *,
1190 struct obd_client_handle *);
1192 int (*m_set_lock_data)(struct obd_export *,
1193 const struct lustre_handle *, void *, __u64 *);
1195 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1196 const struct lu_fid *, enum ldlm_type,
1197 union ldlm_policy_data *, enum ldlm_mode,
1198 struct lustre_handle *);
1200 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1201 union ldlm_policy_data *, enum ldlm_mode,
1202 enum ldlm_cancel_flags flags, void *opaque);
1204 int (*m_get_fid_from_lsm)(struct obd_export *,
1205 const struct lmv_stripe_md *,
1206 const char *name, int namelen,
1207 struct lu_fid *fid);
1208 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1209 const union lmv_mds_md *lmv, size_t lmv_size);
1210 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1211 struct ptlrpc_request_set *set);
1214 static inline struct md_open_data *obd_mod_alloc(void)
1216 struct md_open_data *mod;
1220 atomic_set(&mod->mod_refcount, 1);
1224 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1225 #define obd_mod_put(mod) \
1227 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1228 if ((mod)->mod_open_req) \
1229 ptlrpc_req_finished((mod)->mod_open_req); \
1230 OBD_FREE_PTR(mod); \
1234 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1235 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1236 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1238 /* return 1 if client should be resend request */
1239 static inline int client_should_resend(int resend, struct client_obd *cli)
1241 return atomic_read(&cli->cl_resends) ?
1242 atomic_read(&cli->cl_resends) > resend : 1;
1246 * Return device name for this device
1248 * XXX: lu_device is declared before obd_device, while a pointer pointing
1249 * back to obd_device in lu_device, so this helper function defines here
1250 * instead of in lu_object.h
1252 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1254 return lu_dev->ld_obd->obd_name;
1257 static inline bool filename_is_volatile(const char *name, size_t namelen,
1263 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1266 /* caller does not care of idx */
1270 /* volatile file, the MDT can be set from name */
1271 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1272 /* if no MDT is specified, use std way */
1273 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1275 /* test for no MDT idx case */
1276 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1277 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1281 /* we have an idx, read it */
1282 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1283 *idx = simple_strtoul(start, &end, 16);
1285 * no digit, no trailing :, negative value
1287 if (((*idx == 0) && (end == start)) ||
1288 (*end != ':') || (*idx < 0))
1293 /* bad format of mdt idx, we cannot return an error
1294 * to caller so we use hash algo */
1295 CERROR("Bad volatile file name format: %s\n",
1296 name + LUSTRE_VOLATILE_HDR_LEN);
1300 static inline int cli_brw_size(struct obd_device *obd)
1302 LASSERT(obd != NULL);
1303 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1307 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1308 * of the client should be increased accordingly to avoid sending fragmented
1309 * RPCs over the network when the client runs out of the maximum dirty space
1310 * when so many RPCs are being generated.
1312 static inline void client_adjust_max_dirty(struct client_obd *cli)
1315 if (cli->cl_dirty_max_pages <= 0) {
1316 cli->cl_dirty_max_pages =
1317 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1319 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1320 cli->cl_max_pages_per_rpc;
1322 if (dirty_max > cli->cl_dirty_max_pages)
1323 cli->cl_dirty_max_pages = dirty_max;
1326 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1327 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1329 /* This value is exported to userspace through the max_dirty_mb
1330 * parameter. So we round up the number of pages to make it a round
1332 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1333 1 << (20 - PAGE_SHIFT));
1336 static inline struct inode *page2inode(struct page *page)
1338 if (page->mapping) {
1342 return page->mapping->host;
1348 #endif /* __OBD_H */