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/
36 #include <linux/posix_acl.h>
37 #include <linux/kobject.h>
38 #include <linux/spinlock.h>
39 #include <linux/sysfs.h>
40 #include <libcfs/linux/xarray.h>
42 #include <uapi/linux/lustre/lustre_idl.h>
43 #include <lustre_lib.h>
45 #include <lustre_export.h>
46 #include <lustre_fid.h>
47 #include <lustre_fld.h>
48 #include <lustre_handles.h>
49 #include <lustre_intent.h>
58 struct lov_oinfo { /* per-stripe data structure */
59 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
60 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
61 int loi_ost_gen; /* generation of this loi_ost_idx */
63 unsigned long loi_kms_valid:1;
64 __u64 loi_kms; /* known minimum size */
65 struct ost_lvb loi_lvb;
66 struct osc_async_rc loi_ar;
69 void lov_fix_ea_for_replay(void *lovea);
71 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
74 oinfo->loi_kms_valid = 1;
80 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
82 /* obd info for a particular level (lov, osc). */
84 /* OBD_STATFS_* flags */
86 struct obd_device *oi_obd;
87 struct lu_tgt_desc *oi_tgt;
88 /* statfs data specific for every OSC, if needed at all. */
89 struct obd_statfs *oi_osfs;
90 /* An update callback which is called to update some data on upper
91 * level. E.g. it is used for update lsm->lsm_oinfo at every received
92 * request in osc level for enqueue requests. It is also possible to
93 * update some caller data from LOV layer if needed. */
94 obd_enqueue_update_f oi_cb_up;
98 const struct obd_ops *typ_dt_ops;
99 const struct md_ops *typ_md_ops;
100 struct proc_dir_entry *typ_procroot;
101 struct dentry *typ_debugfs_entry;
102 #ifdef HAVE_SERVER_SUPPORT
106 struct lu_device_type *typ_lu;
107 struct kobject typ_kobj;
109 #define typ_name typ_kobj.name
110 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
114 struct page *bp_page;
117 /* used for encryption: difference with offset in clear text page */
119 /* used for encryption: difference with count in clear text page */
121 } __attribute__((packed));
123 struct timeout_item {
124 enum timeout_event ti_event;
128 struct list_head ti_obd_list;
129 struct list_head ti_chain;
132 #define OBD_MAX_RIF_DEFAULT 8
133 #define OBD_MAX_RIF_MAX 512
134 #define OSC_MAX_RIF_MAX 256
135 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
136 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
137 #define OSC_DEFAULT_RESENDS 10
139 /* possible values for lut_sync_lock_cancel */
140 enum tgt_sync_lock_cancel {
141 SYNC_LOCK_CANCEL_NEVER = 0,
142 SYNC_LOCK_CANCEL_BLOCKING = 1,
143 SYNC_LOCK_CANCEL_ALWAYS = 2,
147 * Limit reply buffer size for striping data to one x86_64 page. This
148 * value is chosen to fit the striping data for common use cases while
149 * staying well below the limit at which the buffer must be backed by
150 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
153 #define OBD_MAX_DEFAULT_EA_SIZE 4096
156 * Lustre can handle larger xattrs internally, but we must respect the Linux
157 * VFS limitation or tools like tar cannot interact with Lustre volumes
160 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
163 enum obd_cl_sem_lock_class {
171 struct rw_semaphore cl_sem;
172 struct obd_uuid cl_target_uuid;
173 struct obd_import *cl_import; /* ptlrpc connection state */
174 size_t cl_conn_count;
176 /* Cache maximum and default values for easize. This is
177 * strictly a performance optimization to minimize calls to
178 * obd_size_diskmd(). The default values are used to calculate the
179 * initial size of a request buffer. The ptlrpc layer will resize the
180 * buffer as needed to accommodate a larger reply from the
181 * server. The default values should be small enough to avoid wasted
182 * memory and excessive use of vmalloc(), yet large enough to avoid
183 * reallocating the buffer in the common use case. */
185 /* Default EA size for striping attributes. It is initialized at
186 * mount-time based on the default stripe width of the filesystem,
187 * then it tracks the largest observed EA size advertised by
188 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
189 __u32 cl_default_mds_easize;
191 /* Maximum possible EA size computed at mount-time based on
192 * the number of OSTs in the filesystem. May be increased at
193 * run-time if a larger observed size is advertised by the MDT. */
194 __u32 cl_max_mds_easize;
196 /* Data-on-MDT specific value to set larger reply buffer for possible
197 * data read along with open/stat requests. By default it tries to use
198 * unused space in reply buffer.
199 * This value is used to ensure that reply buffer has at least as
200 * much free space as value indicates. That free space is gained from
201 * LOV EA buffer which is small for DoM files and on big systems can
202 * provide up to 32KB of extra space in reply buffer.
203 * Default value is 8K now.
205 __u32 cl_dom_min_inline_repsize;
207 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
208 cl_checksum_dump:1, /* same */
209 cl_ocd_grant_param:1,
210 cl_lsom_update:1, /* send LSOM updates */
211 cl_root_squash:1, /* if root squash enabled*/
212 /* check prj quota for root */
214 enum lustre_sec_part cl_sp_me;
215 enum lustre_sec_part cl_sp_to;
216 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
218 /* the grant values are protected by loi_list_lock below */
219 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
220 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
221 unsigned long cl_avail_grant; /* bytes of credit for ost */
222 unsigned long cl_lost_grant; /* lost credits (trunc) */
223 /* grant consumed for dirty pages */
224 unsigned long cl_dirty_grant;
226 /* since we allocate grant by blocks, we don't know how many grant will
227 * be used to add a page into cache. As a solution, we reserve maximum
228 * grant before trying to dirty a page and unreserve the rest.
229 * See osc_{reserve|unreserve}_grant for details. */
230 long cl_reserved_grant;
231 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
232 time64_t cl_next_shrink_grant; /* seconds */
233 struct list_head cl_grant_chain;
234 time64_t cl_grant_shrink_interval; /* seconds */
236 /* A chunk is an optimal size used by osc_extent to determine
237 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
239 /* extent insertion metadata overhead to be accounted in grant,
241 unsigned int cl_grant_extent_tax;
242 /* maximum extent size, in number of pages */
243 unsigned int cl_max_extent_pages;
245 /* keep track of objects that have lois that contain pages which
246 * have been queued for async brw. this lock also protects the
247 * lists of osc_client_pages that hang off of the loi */
249 * ->cl_loi_list_lock protects consistency of
250 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
251 * ->ap_completion() call-backs are executed under this lock. As we
252 * cannot guarantee that these call-backs never block on all platforms
253 * (as a matter of fact they do block on Mac OS X), type of
254 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
255 * and blocking mutex on Mac OS X. (Alternative is to make this lock
256 * blocking everywhere, but we don't want to slow down fast-path of
257 * our main platform.)
259 * NB by Jinshan: though field names are still _loi_, but actually
260 * osc_object{}s are in the list.
262 spinlock_t cl_loi_list_lock;
263 struct list_head cl_loi_ready_list;
264 struct list_head cl_loi_hp_ready_list;
265 struct list_head cl_loi_write_list;
266 struct list_head cl_loi_read_list;
267 __u32 cl_r_in_flight;
268 __u32 cl_w_in_flight;
269 /* just a sum of the loi/lop pending numbers to be exported by /proc */
270 atomic_t cl_pending_w_pages;
271 atomic_t cl_pending_r_pages;
272 u32 cl_max_pages_per_rpc;
273 u32 cl_max_rpcs_in_flight;
274 u32 cl_max_short_io_bytes;
275 ktime_t cl_stats_init;
276 struct obd_histogram cl_read_rpc_hist;
277 struct obd_histogram cl_write_rpc_hist;
278 struct obd_histogram cl_read_page_hist;
279 struct obd_histogram cl_write_page_hist;
280 struct obd_histogram cl_read_offset_hist;
281 struct obd_histogram cl_write_offset_hist;
282 ktime_t cl_batch_stats_init;
283 struct obd_histogram cl_batch_rpc_hist;
285 /** LRU for osc caching pages */
286 struct cl_client_cache *cl_cache;
287 /** member of cl_cache->ccc_lru */
288 struct list_head cl_lru_osc;
289 /** # of available LRU slots left in the per-OSC cache.
290 * Available LRU slots are shared by all OSCs of the same file system,
291 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
292 atomic_long_t *cl_lru_left;
293 /** # of busy LRU pages. A page is considered busy if it's in writeback
294 * queue, or in transfer. Busy pages can't be discarded so they are not
296 atomic_long_t cl_lru_busy;
297 /** # of LRU pages in the cache for this client_obd */
298 atomic_long_t cl_lru_in_list;
299 /** # of threads are shrinking LRU cache. To avoid contention, it's not
300 * allowed to have multiple threads shrinking LRU cache. */
301 atomic_t cl_lru_shrinkers;
302 /** The time when this LRU cache was last used. */
303 time64_t cl_lru_last_used;
304 /** stats: how many reclaims have happened for this client_obd.
305 * reclaim and shrink - shrink is async, voluntarily rebalancing;
306 * reclaim is sync, initiated by IO thread when the LRU slots are
308 __u64 cl_lru_reclaim;
309 /** List of LRU pages for this client_obd */
310 struct list_head cl_lru_list;
311 /** Lock for LRU page list */
312 spinlock_t cl_lru_list_lock;
313 /** # of unstable pages in this client_obd.
314 * An unstable page is a page state that WRITE RPC has finished but
315 * the transaction has NOT yet committed. */
316 atomic_long_t cl_unstable_count;
317 /** Link to osc_shrinker_list */
318 struct list_head cl_shrink_list;
320 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
321 atomic_t cl_destroy_in_flight;
322 wait_queue_head_t cl_destroy_waitq;
324 /* modify rpcs in flight
325 * currently used for metadata only */
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 ktime_t cl_mod_rpcs_init;
332 struct obd_histogram cl_mod_rpcs_hist;
335 struct mutex cl_mgc_mutex;
336 struct local_oid_storage *cl_mgc_los;
337 struct dt_object *cl_mgc_configs_dir;
338 struct obd_export *cl_mgc_mgsexp;
339 atomic_t cl_mgc_refcount;
340 /* in-flight control list and total RPCs counter */
341 struct list_head cl_flight_waiters;
342 __u32 cl_rpcs_in_flight;
344 /* supported checksum types that are worked out at connect time */
345 __u32 cl_supp_cksum_types;
346 /* checksum algorithm to be used */
347 enum cksum_types cl_cksum_type;
348 /* preferred checksum algorithm to be used */
349 enum cksum_types cl_preferred_cksum_type;
351 /* also protected by the poorly named _loi_list_lock lock above */
352 struct osc_async_rc cl_ar;
354 /* sequence manager */
355 struct lu_client_seq *cl_seq;
356 struct rw_semaphore cl_seq_rwsem;
358 atomic_t cl_resends; /* resend count */
360 /* ptlrpc work for writeback in ptlrpcd context */
361 void *cl_writeback_work;
363 struct mutex cl_quota_mutex;
364 /* quota IDs/types that have exceeded quota */
365 struct xarray cl_quota_exceeded_ids;
366 /* the xid of the request updating the hash tables */
367 __u64 cl_quota_last_xid;
368 /* Links to the global list of registered changelog devices */
369 struct list_head cl_chg_dev_linkage;
371 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
378 struct echo_client_obd {
379 struct obd_export *ec_exp; /* the local connection to osc/lov */
381 struct list_head ec_objects;
382 struct list_head ec_locks;
386 /* allow statfs data caching for 1 second */
387 #define OBD_STATFS_CACHE_SECONDS 1
388 /* arbitrary maximum. larger would be useless, allows catching bogus input */
389 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
391 #define lov_tgt_desc lu_tgt_desc
393 struct lov_md_tgt_desc {
394 struct obd_device *lmtd_mdc;
399 struct lov_desc desc;
400 struct lov_tgt_desc **lov_tgts; /* sparse array */
401 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
403 struct mutex lov_lock;
404 struct obd_connect_data lov_ocd;
405 atomic_t lov_refcount;
406 __u32 lov_death_row; /* tgts scheduled to be deleted */
407 __u32 lov_tgt_size; /* size of tgts array */
410 struct rhashtable lov_pools_hash_body; /* used for key access */
411 struct list_head lov_pool_list; /* used for sequential access */
412 struct proc_dir_entry *lov_pool_proc_entry;
413 enum lustre_sec_part lov_sp_me;
415 /* Cached LRU and unstable data from upper layer */
416 struct cl_client_cache *lov_cache;
418 struct rw_semaphore lov_notify_lock;
419 /* Data-on-MDT: MDC array */
420 struct lov_md_tgt_desc *lov_mdc_tgts;
422 struct kobject *lov_tgts_kobj;
425 #define lmv_tgt_desc lu_tgt_desc
428 struct lu_client_fld lmv_fld;
434 u32 lmv_statfs_start;
436 struct lu_tgt_descs lmv_mdt_descs;
438 struct obd_connect_data conn_data;
439 struct kobject *lmv_tgts_kobj;
442 __u32 lmv_qos_rr_index; /* next round-robin MDT idx */
445 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
446 #define lmv_qos lmv_mdt_descs.ltd_qos
448 /* Minimum sector size is 512 */
449 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
451 struct niobuf_local {
452 __u64 lnb_file_offset;
453 __u32 lnb_page_offset;
457 struct page *lnb_page;
459 __be16 lnb_guards[MAX_GUARD_NUMBER];
460 __u16 lnb_guard_rpc:1;
461 __u16 lnb_guard_disk:1;
462 /* separate unlock for read path to allow shared access */
466 struct tgt_thread_big_cache {
467 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
470 #define LUSTRE_FLD_NAME "fld"
471 #define LUSTRE_SEQ_NAME "seq"
473 #define LUSTRE_MDD_NAME "mdd"
474 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
475 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
476 #define LUSTRE_VVP_NAME "vvp"
477 #define LUSTRE_LMV_NAME "lmv"
478 #define LUSTRE_SLP_NAME "slp"
479 #define LUSTRE_LOD_NAME "lod"
480 #define LUSTRE_OSP_NAME "osp"
481 #define LUSTRE_LWP_NAME "lwp"
483 /* obd device type names */
484 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
485 #define LUSTRE_MDS_NAME "mds"
486 #define LUSTRE_MDT_NAME "mdt"
487 #define LUSTRE_MDC_NAME "mdc"
488 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
489 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
490 #define LUSTRE_OSC_NAME "osc"
491 #define LUSTRE_LOV_NAME "lov"
492 #define LUSTRE_MGS_NAME "mgs"
493 #define LUSTRE_MGC_NAME "mgc"
495 #define LUSTRE_ECHO_NAME "obdecho"
496 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
497 #define LUSTRE_QMT_NAME "qmt"
499 /* Constant obd names (post-rename) */
500 #define LUSTRE_MDS_OBDNAME "MDS"
501 #define LUSTRE_OSS_OBDNAME "OSS"
502 #define LUSTRE_MGS_OBDNAME "MGS"
503 #define LUSTRE_MGC_OBDNAME "MGC"
505 static inline int is_lwp_on_mdt(char *name)
509 ptr = strrchr(name, '-');
511 CERROR("%s is not a obdname\n", name);
515 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
517 if (strncmp(ptr + 1, "MDT", 3) != 0)
520 while (*(--ptr) != '-' && ptr != name);
525 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
531 static inline int is_lwp_on_ost(char *name)
535 ptr = strrchr(name, '-');
537 CERROR("%s is not a obdname\n", name);
541 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
543 if (strncmp(ptr + 1, "OST", 3) != 0)
546 while (*(--ptr) != '-' && ptr != name);
551 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
558 * Events signalled through obd_notify() upcall-chain.
560 enum obd_notify_event {
561 /* Device connect start */
563 /* Device activated */
565 /* Device deactivated */
567 /* Connect data for import were changed */
569 /* Administratively deactivate/activate event */
570 OBD_NOTIFY_DEACTIVATE,
575 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
576 * being main example).
578 struct obd_notify_upcall {
579 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
580 enum obd_notify_event ev, void *owner);
581 /* Opaque datum supplied by upper layer listener */
585 struct target_recovery_data {
586 svc_handler_t trd_recovery_handler;
587 pid_t trd_processing_task;
588 struct completion trd_starting;
589 struct completion trd_finishing;
592 struct obd_llog_group {
593 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
594 wait_queue_head_t olg_waitq;
598 /* corresponds to one of the obd's */
599 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
602 struct obd_type *obd_type;
603 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
604 int obd_minor; /* device number: lctl dl */
605 struct lu_device *obd_lu_dev;
607 /* common and UUID name of this device */
608 struct obd_uuid obd_uuid;
609 char obd_name[MAX_OBD_NAME];
611 /* bitfield modification is protected by obd_dev_lock */
613 obd_attached:1, /* finished attach */
614 obd_set_up:1, /* finished setup */
615 obd_recovering:1, /* there are recoverable clients */
616 obd_abort_recovery:1, /* abort client and MDT recovery */
617 obd_abort_mdt_recovery:1, /* abort recovery between MDTs */
618 obd_version_recov:1, /* obd uses version checking */
619 obd_replayable:1, /* recovery enabled; inform clients */
620 obd_no_recov:1, /* fail instead of retry messages */
621 obd_stopping:1, /* started cleanup */
622 obd_starting:1, /* started setup */
623 obd_force:1, /* cleanup with > 0 obd refcount */
624 obd_fail:1, /* cleanup with failover */
625 obd_no_conn:1, /* deny new connections */
626 obd_inactive:1, /* device active/inactive
627 * (for /proc/status only!!) */
628 obd_no_ir:1, /* no imperative recovery. */
629 obd_process_conf:1, /* device is processing mgs config */
630 obd_checksum_dump:1, /* dump pages upon cksum error */
631 obd_dynamic_nids:1, /* Allow dynamic NIDs on device */
632 obd_read_only:1; /* device is read-only */
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 struct kref obd_refcount;
656 int obd_grant_check_threshold;
657 spinlock_t obd_nid_lock;
658 struct ldlm_namespace *obd_namespace;
659 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
660 /* a spinlock is OK for what we do now, may need a semaphore later */
661 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
662 spinlock_t obd_osfs_lock;
663 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
664 time64_t obd_osfs_age;
665 __u64 obd_last_committed;
666 struct mutex obd_dev_mutex;
667 struct lvfs_run_ctxt obd_lvfs_ctxt;
668 struct obd_llog_group obd_olg; /* default llog group */
669 struct obd_device *obd_observer;
670 struct rw_semaphore obd_observer_link_sem;
671 struct obd_notify_upcall obd_upcall;
672 struct obd_export *obd_self_export;
673 struct obd_export *obd_lwp_export;
674 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
675 struct list_head obd_exports_timed;
676 time64_t obd_eviction_timer; /* for ping evictor */
678 atomic_t obd_max_recoverable_clients;
679 atomic_t obd_connected_clients;
680 int obd_stale_clients;
681 /* this lock protects all recovery list_heads, timer and
682 * obd_next_recovery_transno value */
683 spinlock_t obd_recovery_task_lock;
684 __u64 obd_next_recovery_transno;
685 int obd_replayed_requests;
686 int obd_requests_queued_for_recovery;
687 wait_queue_head_t obd_next_transno_waitq;
688 /* protected by obd_recovery_task_lock */
689 struct hrtimer obd_recovery_timer;
691 time64_t obd_recovery_start;
692 /* seconds, for lprocfs_status */
693 time64_t obd_recovery_end;
694 /* To tell timeouts from time stamps Lustre uses timeout_t
695 * instead of time64_t.
697 timeout_t obd_recovery_time_hard;
698 timeout_t obd_recovery_timeout;
699 int obd_recovery_ir_factor;
701 /* new recovery stuff from CMD2 */
702 int obd_replayed_locks;
703 atomic_t obd_req_replay_clients;
704 atomic_t obd_lock_replay_clients;
705 struct target_recovery_data obd_recovery_data;
707 /* all lists are protected by obd_recovery_task_lock */
708 struct list_head obd_req_replay_queue;
709 struct list_head obd_lock_replay_queue;
710 struct list_head obd_final_req_queue;
713 struct client_obd cli;
714 struct echo_client_obd echo_client;
719 /* Fields used by LProcFS */
720 struct lprocfs_stats *obd_stats;
722 struct lprocfs_stats *obd_md_stats;
724 struct dentry *obd_debugfs_entry;
725 struct proc_dir_entry *obd_proc_entry;
726 struct proc_dir_entry *obd_proc_exports_entry;
727 struct dentry *obd_svc_debugfs_entry;
728 struct lprocfs_stats *obd_svc_stats;
729 const struct attribute **obd_attrs;
730 struct lprocfs_vars *obd_vars;
731 struct ldebugfs_vars *obd_debugfs_vars;
732 struct list_head obd_evict_list; /* protected with pet_lock */
733 atomic_t obd_eviction_count;
736 * LDLM pool part. Save last calculated SLV and Limit.
738 rwlock_t obd_pool_lock;
742 int obd_conn_inprogress;
745 * List of outstanding class_incref()'s fo this OBD. For debugging. */
746 struct lu_ref obd_reference;
748 struct kset obd_kset; /* sysfs object collection */
749 struct kobj_type obd_ktype;
750 struct completion obd_kobj_unregister;
752 /* adaptive timeout parameters */
753 unsigned int obd_at_min;
754 unsigned int obd_at_max;
755 unsigned int obd_at_history;
756 unsigned int obd_ldlm_enqueue_min;
759 #define obd_get_at_min(obd) ({ \
760 struct obd_device *_obd = obd; \
761 _obd && _obd->obd_at_min ? _obd->obd_at_min : at_min; \
763 #define obd_get_at_max(obd) ({\
764 struct obd_device *_obd = obd; \
765 _obd && _obd->obd_at_max ? _obd->obd_at_max : at_max; \
767 #define obd_get_at_history(obd) ({ \
768 struct obd_device *_obd = obd; \
769 _obd && _obd->obd_at_history ? _obd->obd_at_history : at_history; \
771 extern unsigned int ldlm_enqueue_min;
772 #define obd_get_ldlm_enqueue_min(obd) ({ \
773 struct obd_device *_obd = obd; \
774 _obd && _obd->obd_ldlm_enqueue_min ? _obd->obd_ldlm_enqueue_min : \
777 #define obd_at_off(obd) (obd_get_at_max(obd) == 0)
779 #define obd_at_get(obd, at) ({ \
780 timeout_t t1 = obd_get_at_min(obd); \
781 max_t(timeout_t, (at)->at_current_timeout, t1); \
784 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
785 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
786 #ifdef HAVE_SERVER_SUPPORT
787 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
788 struct obd_uuid *uuid);
790 int obd_nid_export_for_each(struct obd_device *obd, struct lnet_nid *nid,
791 int cb(struct obd_export *exp, void *data),
793 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
794 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
796 /* both client and MDT recovery are aborted, or MDT is stopping */
797 static inline bool obd_recovery_abort(struct obd_device *obd)
799 return obd->obd_stopping || obd->obd_abort_recovery;
802 /* MDT recovery is aborted, or MDT is stopping */
803 static inline bool obd_mdt_recovery_abort(struct obd_device *obd)
805 return obd->obd_stopping || obd->obd_abort_recovery ||
806 obd->obd_abort_mdt_recovery;
810 /* get/set_info keys */
811 #define KEY_ASYNC "async"
812 #define KEY_CHANGELOG_CLEAR "changelog_clear"
813 #define KEY_FID2PATH "fid2path"
814 #define KEY_CHECKSUM "checksum"
815 #define KEY_CLEAR_FS "clear_fs"
816 #define KEY_CONN_DATA "conn_data"
817 #define KEY_EVICT_BY_NID "evict_by_nid"
818 #define KEY_FIEMAP "fiemap"
819 #define KEY_FLUSH_CTX "flush_ctx"
820 #define KEY_GRANT_SHRINK "grant_shrink"
821 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
822 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
823 #define KEY_INTERMDS "inter_mds"
824 #define KEY_LAST_ID "last_id"
825 #define KEY_LAST_FID "last_fid"
826 #define KEY_MAX_EASIZE "max_easize"
827 #define KEY_DEFAULT_EASIZE "default_easize"
828 #define KEY_MGSSEC "mgssec"
829 #define KEY_READ_ONLY "read-only"
830 #define KEY_REGISTER_TARGET "register_target"
831 #define KEY_SET_FS "set_fs"
832 #define KEY_TGT_COUNT "tgt_count"
833 /* KEY_SET_INFO in lustre_idl.h */
834 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
836 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
837 #define KEY_OSP_CONNECTED "osp_connected"
839 /* Flags for op_xvalid */
841 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
842 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
843 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
844 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
845 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
846 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
847 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
852 static inline int it_to_lock_mode(struct lookup_intent *it)
854 /* CREAT needs to be tested before open (both could be set) */
855 if (it->it_op & IT_CREAT)
857 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
859 else if (it->it_op & IT_LAYOUT)
860 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
861 else if (it->it_op & IT_READDIR)
863 else if (it->it_op & IT_GETXATTR)
866 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
871 MF_MDC_CANCEL_FID1 = BIT(0),
872 MF_MDC_CANCEL_FID2 = BIT(1),
873 MF_MDC_CANCEL_FID3 = BIT(2),
874 MF_MDC_CANCEL_FID4 = BIT(3),
875 MF_GET_MDT_IDX = BIT(4),
876 MF_GETATTR_BY_FID = BIT(5),
877 MF_QOS_MKDIR = BIT(6),
878 MF_RR_MKDIR = BIT(7),
879 MF_OPNAME_KMALLOCED = BIT(8),
883 CLI_SET_MEA = BIT(0),
884 CLI_RM_ENTRY = BIT(1),
887 CLI_MIGRATE = BIT(4),
888 CLI_DIRTY_DATA = BIT(5),
889 CLI_NO_SLOT = BIT(6),
893 LUSTRE_OPC_MKDIR = 1,
903 * GETXATTR is not included as only a couple of fields in the reply body
904 * is filled, but not FID which is needed for common intent handling in
905 * mdc_finish_intent_lock()
907 static inline bool it_has_reply_body(const struct lookup_intent *it)
909 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
913 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
914 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
915 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
916 struct lu_fid op_fid4; /* to the operation locks. */
917 u32 op_mds; /* what mds server open will go to */
919 enum md_op_code op_code;
920 struct lustre_handle op_open_handle;
924 struct lmv_stripe_object *op_lso1;
925 struct lmv_stripe_object *op_lso2;
926 struct lmv_stripe_object *op_default_lso1; /* default LMV */
927 __u32 op_suppgids[2];
934 /* iattr fields and blocks. */
935 struct iattr op_attr;
936 enum op_xvalid op_xvalid; /* eXtra validity flags */
937 loff_t op_attr_blocks;
938 u64 op_valid; /* OBD_MD_* */
939 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
941 enum md_op_flags op_flags;
943 /* Various operation flags. */
944 enum mds_op_bias op_bias;
946 /* used to transfer info between the stacks of MD client
947 * see enum op_cli_flags */
948 enum md_cli_flags op_cli_flags;
950 /* File object data version for HSM release, on client */
951 __u64 op_data_version;
952 struct lustre_handle op_lease_handle;
954 /* File security context, for creates/metadata ops */
955 const char *op_file_secctx_name;
956 __u32 op_file_secctx_name_size;
957 void *op_file_secctx;
958 __u32 op_file_secctx_size;
959 int op_file_secctx_slot;
961 /* File encryption context, for creates/metadata ops */
962 void *op_file_encctx;
963 __u32 op_file_encctx_size;
968 /* Used by readdir */
969 unsigned int op_max_pages;
971 unsigned short op_dir_depth;
977 * used to access dir that is changing layout: if it's set, access
978 * dir by new layout, otherwise old layout.
979 * By default it's not set, because new files are created under new
980 * layout, if we can't find file with name under both old and new
981 * layout, we are sure file with name doesn't exist, but in reverse
982 * order there may be a race with creation by others.
985 /* used to access dir with bash hash */
986 __u32 op_stripe_index;
987 /* Archive ID for PCC attach */
991 struct md_readdir_info {
992 int (*mr_blocking_ast)(struct ldlm_lock *lock,
993 struct ldlm_lock_desc *desc,
994 void *data, int flag);
995 /* if striped directory is partially read, the result is stored here */
996 int mr_partial_readdir_rc;
1000 typedef int (*md_op_item_cb_t)(struct md_op_item *item, int rc);
1002 enum md_item_opcode {
1009 enum md_item_opcode mop_opc;
1010 struct md_op_data mop_data;
1011 struct lookup_intent mop_it;
1012 struct lustre_handle mop_lockh;
1013 struct ldlm_enqueue_info mop_einfo;
1014 md_op_item_cb_t mop_cb;
1016 struct inode *mop_dir;
1017 struct req_capsule *mop_pill;
1018 struct work_struct mop_work;
1019 __u64 mop_lock_flags;
1020 unsigned int mop_subpill_allocated:1;
1023 enum lu_batch_flags {
1024 BATCH_FL_NONE = 0x0,
1025 /* All requests in a batch are read-only. */
1026 BATCH_FL_RDONLY = 0x1,
1027 /* Will create PTLRPC request set for the batch. */
1028 BATCH_FL_RQSET = 0x2,
1029 /* Whether need sync commit. */
1030 BATCH_FL_SYNC = 0x4,
1034 struct ptlrpc_request_set *lbt_rqset;
1037 /* Max batched SUB requests count in a batch. */
1038 __u32 lbt_max_count;
1041 struct batch_update_head {
1042 struct obd_export *buh_exp;
1043 struct lu_batch *buh_batch;
1046 __u32 buh_update_count;
1047 __u32 buh_buf_count;
1051 struct list_head buh_buf_list;
1052 struct list_head buh_cb_list;
1055 struct object_update_callback;
1056 typedef int (*object_update_interpret_t)(struct ptlrpc_request *req,
1057 struct lustre_msg *repmsg,
1058 struct object_update_callback *ouc,
1061 struct object_update_callback {
1062 struct list_head ouc_item;
1063 object_update_interpret_t ouc_interpret;
1064 struct batch_update_head *ouc_head;
1068 typedef int (*md_update_pack_t)(struct batch_update_head *head,
1069 struct lustre_msg *reqmsg,
1070 size_t *max_pack_size,
1071 struct md_op_item *item);
1074 struct lu_batch cbh_super;
1075 struct batch_update_head *cbh_head;
1078 struct lu_batch *cli_batch_create(struct obd_export *exp,
1079 enum lu_batch_flags flags, __u32 max_count);
1080 int cli_batch_stop(struct obd_export *exp, struct lu_batch *bh);
1081 int cli_batch_flush(struct obd_export *exp, struct lu_batch *bh, bool wait);
1082 int cli_batch_add(struct obd_export *exp, struct lu_batch *bh,
1083 struct md_op_item *item, md_update_pack_t packer,
1084 object_update_interpret_t interpreter);
1087 struct module *o_owner;
1088 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1089 void *karg, void __user *uarg);
1090 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1091 __u32 keylen, void *key, __u32 *vallen, void *val);
1092 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1093 __u32 keylen, void *key,
1094 __u32 vallen, void *val,
1095 struct ptlrpc_request_set *set);
1096 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
1097 int (*o_precleanup)(struct obd_device *obd);
1098 int (*o_cleanup)(struct obd_device *obd);
1099 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
1100 int (*o_postrecov)(struct obd_device *obd);
1101 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1103 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1104 /* connect to the target device with given connection
1105 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1106 * granted by the target, which are guaranteed to be a subset of flags
1107 * asked for. If @ocd == NULL, use default parameters.
1108 * On the client, localdata is a struct cl_client_cache *.
1109 * On the server, localdata is a struct lnet_nid *.
1111 int (*o_connect)(const struct lu_env *env,
1112 struct obd_export **exp, struct obd_device *src,
1113 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1115 int (*o_reconnect)(const struct lu_env *env,
1116 struct obd_export *exp, struct obd_device *src,
1117 struct obd_uuid *cluuid,
1118 struct obd_connect_data *ocd,
1120 int (*o_disconnect)(struct obd_export *exp);
1122 /* Initialize/finalize fids infrastructure. */
1123 int (*o_fid_init)(struct obd_device *obd,
1124 struct obd_export *exp, enum lu_cli_type type);
1125 int (*o_fid_fini)(struct obd_device *obd);
1127 /* Allocate new fid according to passed @hint. */
1128 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1129 struct lu_fid *fid, struct md_op_data *op_data);
1132 * Object with @fid is getting deleted, we may want to do something
1135 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1136 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1137 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1138 time64_t max_age, struct ptlrpc_request_set *set);
1139 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1141 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1143 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1145 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1147 int (*o_preprw)(const struct lu_env *env, int cmd,
1148 struct obd_export *exp, struct obdo *oa, int objcount,
1149 struct obd_ioobj *obj, struct niobuf_remote *remote,
1150 int *nr_pages, struct niobuf_local *local);
1151 int (*o_commitrw)(const struct lu_env *env, int cmd,
1152 struct obd_export *exp, struct obdo *oa,
1153 int objcount, struct obd_ioobj *obj,
1154 struct niobuf_remote *remote, int pages,
1155 struct niobuf_local *local, int rc, int nob,
1157 int (*o_init_export)(struct obd_export *exp);
1158 int (*o_destroy_export)(struct obd_export *exp);
1160 int (*o_import_event)(struct obd_device *, struct obd_import *,
1161 enum obd_import_event);
1163 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1164 enum obd_notify_event ev);
1166 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1167 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1170 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1171 struct obd_quotactl *);
1174 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1175 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1176 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1178 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1182 /* lmv structures */
1184 struct mdt_body *body;
1185 struct lu_buf layout;
1186 struct lmv_stripe_object *lsm_obj;
1187 struct lmv_stripe_object *def_lsm_obj;
1188 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1189 struct posix_acl *posix_acl;
1193 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1194 static inline void lmd_clear_acl(struct lustre_md *md)
1196 if (md->posix_acl) {
1197 posix_acl_release(md->posix_acl);
1198 md->posix_acl = NULL;
1202 #define OBD_CONNECT_ACL_FLAGS \
1203 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1205 static inline void lmd_clear_acl(struct lustre_md *md)
1209 #define OBD_CONNECT_ACL_FLAGS (0)
1212 struct md_open_data {
1213 struct obd_client_handle *mod_och;
1214 struct ptlrpc_request *mod_open_req;
1215 struct ptlrpc_request *mod_close_req;
1216 atomic_t mod_refcount;
1220 struct obd_client_handle {
1221 struct lustre_handle och_open_handle;
1222 struct lu_fid och_fid;
1223 struct md_open_data *och_mod;
1224 struct lustre_handle och_lease_handle; /* open lock for lease */
1229 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1231 struct lookup_intent;
1235 int (*m_close)(struct obd_export *, struct md_op_data *,
1236 struct md_open_data *, struct ptlrpc_request **);
1238 int (*m_create)(struct obd_export *, struct md_op_data *,
1239 const void *, size_t, umode_t, uid_t, gid_t,
1240 kernel_cap_t, __u64, struct ptlrpc_request **);
1242 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1243 const union ldlm_policy_data *, struct md_op_data *,
1244 struct lustre_handle *, __u64);
1246 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1247 struct ptlrpc_request **);
1249 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1250 struct lookup_intent *,
1251 struct ptlrpc_request **,
1252 ldlm_blocking_callback, __u64);
1254 int (*m_link)(struct obd_export *, struct md_op_data *,
1255 struct ptlrpc_request **);
1257 int (*m_rename)(struct obd_export *, struct md_op_data *,
1258 const char *, size_t, const char *, size_t,
1259 struct ptlrpc_request **);
1261 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1262 size_t , struct ptlrpc_request **);
1264 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1265 struct ptlrpc_request **);
1267 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1268 struct md_readdir_info *mrinfo, __u64 hash_offset,
1269 struct page **ppage);
1271 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1272 struct ptlrpc_request **);
1274 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1275 u64, const char *, const void *, size_t, unsigned int,
1276 u32, struct ptlrpc_request **);
1278 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1279 u64, const char *, size_t, struct ptlrpc_request **);
1281 int (*m_intent_getattr_async)(struct obd_export *,
1282 struct md_op_item *);
1284 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1285 struct lu_fid *, __u64 *bits);
1287 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1289 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1290 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1292 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1293 struct ptlrpc_request **);
1295 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1297 int (*m_get_lustre_md)(struct obd_export *, struct req_capsule *,
1298 struct obd_export *, struct obd_export *,
1299 struct lustre_md *);
1301 int (*m_put_lustre_md)(struct obd_export *, struct lustre_md *);
1303 int (*m_merge_attr)(struct obd_export *,
1304 const struct lmv_stripe_object *,
1305 struct cl_attr *attr, ldlm_blocking_callback);
1307 int (*m_set_open_replay_data)(struct obd_export *,
1308 struct obd_client_handle *,
1309 struct lookup_intent *);
1311 int (*m_clear_open_replay_data)(struct obd_export *,
1312 struct obd_client_handle *);
1314 int (*m_set_lock_data)(struct obd_export *,
1315 const struct lustre_handle *, void *, __u64 *);
1317 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1318 const struct lu_fid *, enum ldlm_type,
1319 union ldlm_policy_data *, enum ldlm_mode,
1320 struct lustre_handle *);
1322 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1323 union ldlm_policy_data *, enum ldlm_mode,
1324 enum ldlm_cancel_flags flags, void *opaque);
1326 int (*m_get_fid_from_lsm)(struct obd_export *,
1327 const struct lmv_stripe_object *,
1328 const char *name, int namelen,
1329 struct lu_fid *fid);
1330 int (*m_stripe_object_create)(struct obd_export *exp,
1331 struct lmv_stripe_object **plso,
1332 const union lmv_mds_md *lmv,
1334 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1335 struct ptlrpc_request_set *set);
1336 struct lu_batch *(*m_batch_create)(struct obd_export *exp,
1337 enum lu_batch_flags flags,
1339 int (*m_batch_stop)(struct obd_export *exp, struct lu_batch *bh);
1340 int (*m_batch_flush)(struct obd_export *exp, struct lu_batch *bh,
1342 int (*m_batch_add)(struct obd_export *exp, struct lu_batch *bh,
1343 struct md_op_item *item);
1346 static inline struct md_open_data *obd_mod_alloc(void)
1348 struct md_open_data *mod;
1352 atomic_set(&mod->mod_refcount, 1);
1356 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1357 #define obd_mod_put(mod) \
1359 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1360 if ((mod)->mod_open_req) \
1361 ptlrpc_req_finished((mod)->mod_open_req); \
1362 if ((mod)->mod_close_req) \
1363 ptlrpc_req_finished((mod)->mod_close_req);\
1364 OBD_FREE_PTR(mod); \
1368 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1369 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1370 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1372 /* return 1 if client should be resend request */
1373 static inline int client_should_resend(int resend, struct client_obd *cli)
1375 return atomic_read(&cli->cl_resends) ?
1376 atomic_read(&cli->cl_resends) > resend : 1;
1380 * Return device name for this device
1382 * XXX: lu_device is declared before obd_device, while a pointer pointing
1383 * back to obd_device in lu_device, so this helper function defines here
1384 * instead of in lu_object.h
1386 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1388 return lu_dev->ld_obd->obd_name;
1391 static inline bool filename_is_volatile(const char *name, size_t namelen,
1397 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1400 /* caller does not care of idx */
1404 /* volatile file, the MDT can be set from name */
1405 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1406 /* if no MDT is specified, use std way */
1407 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1409 /* test for no MDT idx case */
1410 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1411 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1415 /* we have an idx, read it */
1416 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1417 rc = sscanf(start, "%4x:%4x:fd=%2d", idx, &rnd, &fd);
1418 /* error cases: no digit or negative value
1419 * rc will never be larger then 3
1421 if (rc <= 0 || *idx < 0)
1426 /* bad format of mdt idx, we cannot return an error
1427 * to caller so we use hash algo */
1428 CERROR("Bad volatile file name format: %s\n",
1429 name + LUSTRE_VOLATILE_HDR_LEN);
1433 static inline int cli_brw_size(struct obd_device *obd)
1435 LASSERT(obd != NULL);
1436 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1440 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1441 * of the client should be increased accordingly to avoid sending fragmented
1442 * RPCs over the network when the client runs out of the maximum dirty space
1443 * when so many RPCs are being generated.
1445 static inline void client_adjust_max_dirty(struct client_obd *cli)
1448 if (cli->cl_dirty_max_pages <= 0) {
1449 cli->cl_dirty_max_pages =
1450 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1452 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1453 cli->cl_max_pages_per_rpc;
1455 if (dirty_max > cli->cl_dirty_max_pages)
1456 cli->cl_dirty_max_pages = dirty_max;
1459 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1460 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1462 /* This value is exported to userspace through the max_dirty_mb
1463 * parameter. So we round up the number of pages to make it a round
1465 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1466 1 << (20 - PAGE_SHIFT));
1469 /* Must be used for page cache pages only,
1470 * not safe otherwise (e.g. direct IO pages)
1472 static inline struct inode *page2inode(struct page *page)
1474 if (page->mapping) {
1478 return page->mapping->host;
1484 #endif /* __OBD_H */