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 <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>
52 #define MAX_OBD_DEVICES 8192
60 struct lov_oinfo { /* per-stripe data structure */
61 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
62 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
63 int loi_ost_gen; /* generation of this loi_ost_idx */
65 unsigned long loi_kms_valid:1;
66 __u64 loi_kms; /* known minimum size */
67 struct ost_lvb loi_lvb;
68 struct osc_async_rc loi_ar;
71 void lov_fix_ea_for_replay(void *lovea);
73 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
76 oinfo->loi_kms_valid = 1;
82 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
84 /* obd info for a particular level (lov, osc). */
86 /* OBD_STATFS_* flags */
88 struct obd_device *oi_obd;
89 struct lu_tgt_desc *oi_tgt;
90 /* statfs data specific for every OSC, if needed at all. */
91 struct obd_statfs *oi_osfs;
92 /* An update callback which is called to update some data on upper
93 * level. E.g. it is used for update lsm->lsm_oinfo at every received
94 * request in osc level for enqueue requests. It is also possible to
95 * update some caller data from LOV layer if needed. */
96 obd_enqueue_update_f oi_cb_up;
100 const struct obd_ops *typ_dt_ops;
101 const struct md_ops *typ_md_ops;
102 struct proc_dir_entry *typ_procroot;
103 struct dentry *typ_debugfs_entry;
104 #ifdef HAVE_SERVER_SUPPORT
108 struct lu_device_type *typ_lu;
109 struct kobject typ_kobj;
111 #define typ_name typ_kobj.name
112 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
119 /* used for encryption: difference with offset in clear text page */
121 /* used for encryption: difference with count in clear text page */
123 } __attribute__((packed));
125 struct timeout_item {
126 enum timeout_event ti_event;
130 struct list_head ti_obd_list;
131 struct list_head ti_chain;
134 #define OBD_MAX_RIF_DEFAULT 8
135 #define OBD_MAX_RIF_MAX 512
136 #define OSC_MAX_RIF_MAX 256
137 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
138 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
139 #define OSC_DEFAULT_RESENDS 10
141 /* possible values for lut_sync_lock_cancel */
142 enum tgt_sync_lock_cancel {
143 SYNC_LOCK_CANCEL_NEVER = 0,
144 SYNC_LOCK_CANCEL_BLOCKING = 1,
145 SYNC_LOCK_CANCEL_ALWAYS = 2,
149 * Limit reply buffer size for striping data to one x86_64 page. This
150 * value is chosen to fit the striping data for common use cases while
151 * staying well below the limit at which the buffer must be backed by
152 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
155 #define OBD_MAX_DEFAULT_EA_SIZE 4096
158 * Lustre can handle larger xattrs internally, but we must respect the Linux
159 * VFS limitation or tools like tar cannot interact with Lustre volumes
162 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
165 enum obd_cl_sem_lock_class {
173 struct rw_semaphore cl_sem;
174 struct obd_uuid cl_target_uuid;
175 struct obd_import *cl_import; /* ptlrpc connection state */
176 size_t cl_conn_count;
178 /* Cache maximum and default values for easize. This is
179 * strictly a performance optimization to minimize calls to
180 * obd_size_diskmd(). The default values are used to calculate the
181 * initial size of a request buffer. The ptlrpc layer will resize the
182 * buffer as needed to accommodate a larger reply from the
183 * server. The default values should be small enough to avoid wasted
184 * memory and excessive use of vmalloc(), yet large enough to avoid
185 * reallocating the buffer in the common use case. */
187 /* Default EA size for striping attributes. It is initialized at
188 * mount-time based on the default stripe width of the filesystem,
189 * then it tracks the largest observed EA size advertised by
190 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
191 __u32 cl_default_mds_easize;
193 /* Maximum possible EA size computed at mount-time based on
194 * the number of OSTs in the filesystem. May be increased at
195 * run-time if a larger observed size is advertised by the MDT. */
196 __u32 cl_max_mds_easize;
198 /* Data-on-MDT specific value to set larger reply buffer for possible
199 * data read along with open/stat requests. By default it tries to use
200 * unused space in reply buffer.
201 * This value is used to ensure that reply buffer has at least as
202 * much free space as value indicates. That free space is gained from
203 * LOV EA buffer which is small for DoM files and on big systems can
204 * provide up to 32KB of extra space in reply buffer.
205 * Default value is 8K now.
207 __u32 cl_dom_min_inline_repsize;
209 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
210 cl_checksum_dump:1, /* same */
211 cl_ocd_grant_param:1,
212 cl_lsom_update:1, /* send LSOM updates */
213 cl_root_squash:1, /* if root squash enabled*/
214 /* check prj quota for root */
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 ktime_t cl_stats_init;
278 struct obd_histogram cl_read_rpc_hist;
279 struct obd_histogram cl_write_rpc_hist;
280 struct obd_histogram cl_read_page_hist;
281 struct obd_histogram cl_write_page_hist;
282 struct obd_histogram cl_read_offset_hist;
283 struct obd_histogram cl_write_offset_hist;
284 ktime_t cl_batch_stats_init;
285 struct obd_histogram cl_batch_rpc_hist;
287 /** LRU for osc caching pages */
288 struct cl_client_cache *cl_cache;
289 /** member of cl_cache->ccc_lru */
290 struct list_head cl_lru_osc;
291 /** # of available LRU slots left in the per-OSC cache.
292 * Available LRU slots are shared by all OSCs of the same file system,
293 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
294 atomic_long_t *cl_lru_left;
295 /** # of busy LRU pages. A page is considered busy if it's in writeback
296 * queue, or in transfer. Busy pages can't be discarded so they are not
298 atomic_long_t cl_lru_busy;
299 /** # of LRU pages in the cache for this client_obd */
300 atomic_long_t cl_lru_in_list;
301 /** # of threads are shrinking LRU cache. To avoid contention, it's not
302 * allowed to have multiple threads shrinking LRU cache. */
303 atomic_t cl_lru_shrinkers;
304 /** The time when this LRU cache was last used. */
305 time64_t cl_lru_last_used;
306 /** stats: how many reclaims have happened for this client_obd.
307 * reclaim and shrink - shrink is async, voluntarily rebalancing;
308 * reclaim is sync, initiated by IO thread when the LRU slots are
310 __u64 cl_lru_reclaim;
311 /** List of LRU pages for this client_obd */
312 struct list_head cl_lru_list;
313 /** Lock for LRU page list */
314 spinlock_t cl_lru_list_lock;
315 /** # of unstable pages in this client_obd.
316 * An unstable page is a page state that WRITE RPC has finished but
317 * the transaction has NOT yet committed. */
318 atomic_long_t cl_unstable_count;
319 /** Link to osc_shrinker_list */
320 struct list_head cl_shrink_list;
322 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
323 atomic_t cl_destroy_in_flight;
324 wait_queue_head_t cl_destroy_waitq;
326 /* modify rpcs in flight
327 * currently used for metadata only */
328 __u16 cl_max_mod_rpcs_in_flight;
329 __u16 cl_mod_rpcs_in_flight;
330 __u16 cl_close_rpcs_in_flight;
331 wait_queue_head_t cl_mod_rpcs_waitq;
332 unsigned long *cl_mod_tag_bitmap;
333 ktime_t cl_mod_rpcs_init;
334 struct obd_histogram cl_mod_rpcs_hist;
337 struct mutex cl_mgc_mutex;
338 struct local_oid_storage *cl_mgc_los;
339 struct dt_object *cl_mgc_configs_dir;
340 struct obd_export *cl_mgc_mgsexp;
341 atomic_t cl_mgc_refcount;
342 /* in-flight control list and total RPCs counter */
343 struct list_head cl_flight_waiters;
344 __u32 cl_rpcs_in_flight;
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 atomic_t lov_refcount;
408 __u32 lov_death_row; /* tgts scheduled to be deleted */
409 __u32 lov_tgt_size; /* size of tgts array */
412 struct rhashtable lov_pools_hash_body; /* used for key access */
413 struct list_head lov_pool_list; /* used for sequential access */
414 struct proc_dir_entry *lov_pool_proc_entry;
415 enum lustre_sec_part lov_sp_me;
417 /* Cached LRU and unstable data from upper layer */
418 struct cl_client_cache *lov_cache;
420 struct rw_semaphore lov_notify_lock;
421 /* Data-on-MDT: MDC array */
422 struct lov_md_tgt_desc *lov_mdc_tgts;
424 struct kobject *lov_tgts_kobj;
427 #define lmv_tgt_desc lu_tgt_desc
430 struct lu_client_fld lmv_fld;
436 u32 lmv_statfs_start;
438 struct lu_tgt_descs lmv_mdt_descs;
440 struct obd_connect_data conn_data;
441 struct kobject *lmv_tgts_kobj;
444 __u32 lmv_qos_rr_index;
447 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
448 #define lmv_qos lmv_mdt_descs.ltd_qos
450 /* Minimum sector size is 512 */
451 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
453 struct niobuf_local {
454 __u64 lnb_file_offset;
455 __u32 lnb_page_offset;
459 struct page *lnb_page;
461 __be16 lnb_guards[MAX_GUARD_NUMBER];
462 __u16 lnb_guard_rpc:1;
463 __u16 lnb_guard_disk:1;
464 /* separate unlock for read path to allow shared access */
468 struct tgt_thread_big_cache {
469 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
472 #define LUSTRE_FLD_NAME "fld"
473 #define LUSTRE_SEQ_NAME "seq"
475 #define LUSTRE_MDD_NAME "mdd"
476 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
477 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
478 #define LUSTRE_VVP_NAME "vvp"
479 #define LUSTRE_LMV_NAME "lmv"
480 #define LUSTRE_SLP_NAME "slp"
481 #define LUSTRE_LOD_NAME "lod"
482 #define LUSTRE_OSP_NAME "osp"
483 #define LUSTRE_LWP_NAME "lwp"
485 /* obd device type names */
486 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
487 #define LUSTRE_MDS_NAME "mds"
488 #define LUSTRE_MDT_NAME "mdt"
489 #define LUSTRE_MDC_NAME "mdc"
490 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
491 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
492 #define LUSTRE_OSC_NAME "osc"
493 #define LUSTRE_LOV_NAME "lov"
494 #define LUSTRE_MGS_NAME "mgs"
495 #define LUSTRE_MGC_NAME "mgc"
497 #define LUSTRE_ECHO_NAME "obdecho"
498 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
499 #define LUSTRE_QMT_NAME "qmt"
501 /* Constant obd names (post-rename) */
502 #define LUSTRE_MDS_OBDNAME "MDS"
503 #define LUSTRE_OSS_OBDNAME "OSS"
504 #define LUSTRE_MGS_OBDNAME "MGS"
505 #define LUSTRE_MGC_OBDNAME "MGC"
507 static inline int is_lwp_on_mdt(char *name)
511 ptr = strrchr(name, '-');
513 CERROR("%s is not a obdname\n", name);
517 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
519 if (strncmp(ptr + 1, "MDT", 3) != 0)
522 while (*(--ptr) != '-' && ptr != name);
527 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
533 static inline int is_lwp_on_ost(char *name)
537 ptr = strrchr(name, '-');
539 CERROR("%s is not a obdname\n", name);
543 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
545 if (strncmp(ptr + 1, "OST", 3) != 0)
548 while (*(--ptr) != '-' && ptr != name);
553 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
560 * Events signalled through obd_notify() upcall-chain.
562 enum obd_notify_event {
563 /* Device connect start */
565 /* Device activated */
567 /* Device deactivated */
569 /* Connect data for import were changed */
571 /* Administratively deactivate/activate event */
572 OBD_NOTIFY_DEACTIVATE,
577 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
578 * being main example).
580 struct obd_notify_upcall {
581 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
582 enum obd_notify_event ev, void *owner);
583 /* Opaque datum supplied by upper layer listener */
587 struct target_recovery_data {
588 svc_handler_t trd_recovery_handler;
589 pid_t trd_processing_task;
590 struct completion trd_starting;
591 struct completion trd_finishing;
594 struct obd_llog_group {
595 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
596 wait_queue_head_t olg_waitq;
600 /* corresponds to one of the obd's */
601 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
604 struct obd_type *obd_type;
605 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
606 int obd_minor; /* device number: lctl dl */
607 struct lu_device *obd_lu_dev;
609 /* common and UUID name of this device */
610 struct obd_uuid obd_uuid;
611 char obd_name[MAX_OBD_NAME];
613 /* bitfield modification is protected by obd_dev_lock */
615 obd_attached:1, /* finished attach */
616 obd_set_up:1, /* finished setup */
617 obd_recovering:1, /* there are recoverable clients */
618 obd_abort_recovery:1, /* abort client and MDT recovery */
619 obd_abort_mdt_recovery:1, /* abort recovery between MDTs */
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 obd_dynamic_nids:1, /* Allow dynamic NIDs on device */
634 obd_read_only:1; /* device is read-only */
635 #ifdef HAVE_SERVER_SUPPORT
636 /* no committed-transno notification */
637 unsigned long obd_no_transno:1;
640 /* use separate field as it is set in interrupt to don't mess with
641 * protection of other bits using _bh lock */
642 unsigned long obd_recovery_expired:1;
643 /* uuid-export hash body */
644 struct rhashtable obd_uuid_hash;
645 /* nid-export hash body */
646 struct rhltable obd_nid_hash;
648 struct cfs_hash *obd_nid_stats_hash;
649 /* client_generation-export hash body */
650 struct cfs_hash *obd_gen_hash;
651 struct list_head obd_nid_stats;
652 struct list_head obd_exports;
653 struct list_head obd_unlinked_exports;
654 struct list_head obd_delayed_exports;
655 struct list_head obd_lwp_list;
656 atomic_t obd_refcount;
658 int obd_grant_check_threshold;
659 spinlock_t obd_nid_lock;
660 struct ldlm_namespace *obd_namespace;
661 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
662 /* a spinlock is OK for what we do now, may need a semaphore later */
663 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
664 spinlock_t obd_osfs_lock;
665 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
666 time64_t obd_osfs_age;
667 __u64 obd_last_committed;
668 struct mutex obd_dev_mutex;
669 struct lvfs_run_ctxt obd_lvfs_ctxt;
670 struct obd_llog_group obd_olg; /* default llog group */
671 struct obd_device *obd_observer;
672 struct rw_semaphore obd_observer_link_sem;
673 struct obd_notify_upcall obd_upcall;
674 struct obd_export *obd_self_export;
675 struct obd_export *obd_lwp_export;
676 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
677 struct list_head obd_exports_timed;
678 time64_t obd_eviction_timer; /* for ping evictor */
680 atomic_t obd_max_recoverable_clients;
681 atomic_t obd_connected_clients;
682 int obd_stale_clients;
683 /* this lock protects all recovery list_heads, timer and
684 * obd_next_recovery_transno value */
685 spinlock_t obd_recovery_task_lock;
686 __u64 obd_next_recovery_transno;
687 int obd_replayed_requests;
688 int obd_requests_queued_for_recovery;
689 wait_queue_head_t obd_next_transno_waitq;
690 /* protected by obd_recovery_task_lock */
691 struct hrtimer obd_recovery_timer;
693 time64_t obd_recovery_start;
694 /* seconds, for lprocfs_status */
695 time64_t obd_recovery_end;
696 /* To tell timeouts from time stamps Lustre uses timeout_t
697 * instead of time64_t.
699 timeout_t obd_recovery_time_hard;
700 timeout_t obd_recovery_timeout;
701 int obd_recovery_ir_factor;
703 /* new recovery stuff from CMD2 */
704 int obd_replayed_locks;
705 atomic_t obd_req_replay_clients;
706 atomic_t obd_lock_replay_clients;
707 struct target_recovery_data obd_recovery_data;
709 /* all lists are protected by obd_recovery_task_lock */
710 struct list_head obd_req_replay_queue;
711 struct list_head obd_lock_replay_queue;
712 struct list_head obd_final_req_queue;
715 struct client_obd cli;
716 struct echo_client_obd echo_client;
721 /* Fields used by LProcFS */
722 struct lprocfs_stats *obd_stats;
724 struct lprocfs_stats *obd_md_stats;
726 struct dentry *obd_debugfs_entry;
727 struct proc_dir_entry *obd_proc_entry;
728 struct proc_dir_entry *obd_proc_exports_entry;
729 struct dentry *obd_svc_debugfs_entry;
730 struct lprocfs_stats *obd_svc_stats;
731 const struct attribute **obd_attrs;
732 struct lprocfs_vars *obd_vars;
733 struct ldebugfs_vars *obd_debugfs_vars;
734 atomic_t obd_evict_inprogress;
735 wait_queue_head_t obd_evict_inprogress_waitq;
736 struct list_head obd_evict_list; /* protected with pet_lock */
737 atomic_t obd_eviction_count;
740 * LDLM pool part. Save last calculated SLV and Limit.
742 rwlock_t obd_pool_lock;
746 int obd_conn_inprogress;
749 * List of outstanding class_incref()'s fo this OBD. For debugging. */
750 struct lu_ref obd_reference;
752 struct kset obd_kset; /* sysfs object collection */
753 struct kobj_type obd_ktype;
754 struct completion obd_kobj_unregister;
757 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
758 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
759 #ifdef HAVE_SERVER_SUPPORT
760 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
761 struct obd_uuid *uuid);
763 int obd_nid_export_for_each(struct obd_device *obd, struct lnet_nid *nid,
764 int cb(struct obd_export *exp, void *data),
766 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
767 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
769 /* both client and MDT recovery are aborted, or MDT is stopping */
770 static inline bool obd_recovery_abort(struct obd_device *obd)
772 return obd->obd_stopping || obd->obd_abort_recovery;
775 /* MDT recovery is aborted, or MDT is stopping */
776 static inline bool obd_mdt_recovery_abort(struct obd_device *obd)
778 return obd->obd_stopping || obd->obd_abort_recovery ||
779 obd->obd_abort_mdt_recovery;
783 /* get/set_info keys */
784 #define KEY_ASYNC "async"
785 #define KEY_CHANGELOG_CLEAR "changelog_clear"
786 #define KEY_FID2PATH "fid2path"
787 #define KEY_CHECKSUM "checksum"
788 #define KEY_CLEAR_FS "clear_fs"
789 #define KEY_CONN_DATA "conn_data"
790 #define KEY_EVICT_BY_NID "evict_by_nid"
791 #define KEY_FIEMAP "fiemap"
792 #define KEY_FLUSH_CTX "flush_ctx"
793 #define KEY_GRANT_SHRINK "grant_shrink"
794 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
795 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
796 #define KEY_INTERMDS "inter_mds"
797 #define KEY_LAST_ID "last_id"
798 #define KEY_LAST_FID "last_fid"
799 #define KEY_MAX_EASIZE "max_easize"
800 #define KEY_DEFAULT_EASIZE "default_easize"
801 #define KEY_MGSSEC "mgssec"
802 #define KEY_READ_ONLY "read-only"
803 #define KEY_REGISTER_TARGET "register_target"
804 #define KEY_SET_FS "set_fs"
805 #define KEY_TGT_COUNT "tgt_count"
806 /* KEY_SET_INFO in lustre_idl.h */
807 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
809 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
810 #define KEY_OSP_CONNECTED "osp_connected"
812 /* Flags for op_xvalid */
814 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
815 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
816 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
817 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
818 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
819 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
820 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
825 static inline int it_to_lock_mode(struct lookup_intent *it)
827 /* CREAT needs to be tested before open (both could be set) */
828 if (it->it_op & IT_CREAT)
830 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
832 else if (it->it_op & IT_LAYOUT)
833 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
834 else if (it->it_op & IT_READDIR)
836 else if (it->it_op & IT_GETXATTR)
839 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
844 MF_MDC_CANCEL_FID1 = BIT(0),
845 MF_MDC_CANCEL_FID2 = BIT(1),
846 MF_MDC_CANCEL_FID3 = BIT(2),
847 MF_MDC_CANCEL_FID4 = BIT(3),
848 MF_GET_MDT_IDX = BIT(4),
849 MF_GETATTR_BY_FID = BIT(5),
850 MF_QOS_MKDIR = BIT(6),
851 MF_RR_MKDIR = BIT(7),
852 MF_OPNAME_KMALLOCED = BIT(8),
856 CLI_SET_MEA = BIT(0),
857 CLI_RM_ENTRY = BIT(1),
860 CLI_MIGRATE = BIT(4),
861 CLI_DIRTY_DATA = BIT(5),
862 CLI_NO_SLOT = BIT(6),
866 LUSTRE_OPC_MKDIR = 1,
876 * GETXATTR is not included as only a couple of fields in the reply body
877 * is filled, but not FID which is needed for common intent handling in
878 * mdc_finish_intent_lock()
880 static inline bool it_has_reply_body(const struct lookup_intent *it)
882 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
886 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
887 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
888 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
889 struct lu_fid op_fid4; /* to the operation locks. */
890 u32 op_mds; /* what mds server open will go to */
892 enum md_op_code op_code;
893 struct lustre_handle op_open_handle;
897 struct rw_semaphore *op_mea1_sem;
898 struct rw_semaphore *op_mea2_sem;
899 struct lmv_stripe_md *op_mea1;
900 struct lmv_stripe_md *op_mea2;
901 struct lmv_stripe_md *op_default_mea1; /* default LMV */
902 __u32 op_suppgids[2];
909 /* iattr fields and blocks. */
910 struct iattr op_attr;
911 enum op_xvalid op_xvalid; /* eXtra validity flags */
912 loff_t op_attr_blocks;
913 u64 op_valid; /* OBD_MD_* */
914 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
916 enum md_op_flags op_flags;
918 /* Various operation flags. */
919 enum mds_op_bias op_bias;
921 /* used to transfer info between the stacks of MD client
922 * see enum op_cli_flags */
923 enum md_cli_flags op_cli_flags;
925 /* File object data version for HSM release, on client */
926 __u64 op_data_version;
927 struct lustre_handle op_lease_handle;
929 /* File security context, for creates/metadata ops */
930 const char *op_file_secctx_name;
931 __u32 op_file_secctx_name_size;
932 void *op_file_secctx;
933 __u32 op_file_secctx_size;
934 int op_file_secctx_slot;
936 /* File encryption context, for creates/metadata ops */
937 void *op_file_encctx;
938 __u32 op_file_encctx_size;
943 /* Used by readdir */
944 unsigned int op_max_pages;
946 unsigned short op_dir_depth;
952 * used to access dir that is changing layout: if it's set, access
953 * dir by new layout, otherwise old layout.
954 * By default it's not set, because new files are created under new
955 * layout, if we can't find file with name under both old and new
956 * layout, we are sure file with name doesn't exist, but in reverse
957 * order there may be a race with creation by others.
960 /* used to access dir with bash hash */
961 __u32 op_stripe_index;
962 /* Archive ID for PCC attach */
966 struct md_readdir_info {
967 int (*mr_blocking_ast)(struct ldlm_lock *lock,
968 struct ldlm_lock_desc *desc,
969 void *data, int flag);
970 /* if striped directory is partially read, the result is stored here */
971 int mr_partial_readdir_rc;
975 typedef int (*md_op_item_cb_t)(struct md_op_item *item, int rc);
977 enum md_item_opcode {
984 enum md_item_opcode mop_opc;
985 struct md_op_data mop_data;
986 struct lookup_intent mop_it;
987 struct lustre_handle mop_lockh;
988 struct ldlm_enqueue_info mop_einfo;
989 md_op_item_cb_t mop_cb;
991 struct inode *mop_dir;
992 struct req_capsule *mop_pill;
993 struct work_struct mop_work;
994 __u64 mop_lock_flags;
995 unsigned int mop_subpill_allocated:1;
998 enum lu_batch_flags {
1000 /* All requests in a batch are read-only. */
1001 BATCH_FL_RDONLY = 0x1,
1002 /* Will create PTLRPC request set for the batch. */
1003 BATCH_FL_RQSET = 0x2,
1004 /* Whether need sync commit. */
1005 BATCH_FL_SYNC = 0x4,
1009 struct ptlrpc_request_set *lbt_rqset;
1012 /* Max batched SUB requests count in a batch. */
1013 __u32 lbt_max_count;
1016 struct batch_update_head {
1017 struct obd_export *buh_exp;
1018 struct lu_batch *buh_batch;
1021 __u32 buh_update_count;
1022 __u32 buh_buf_count;
1026 struct list_head buh_buf_list;
1027 struct list_head buh_cb_list;
1030 struct object_update_callback;
1031 typedef int (*object_update_interpret_t)(struct ptlrpc_request *req,
1032 struct lustre_msg *repmsg,
1033 struct object_update_callback *ouc,
1036 struct object_update_callback {
1037 struct list_head ouc_item;
1038 object_update_interpret_t ouc_interpret;
1039 struct batch_update_head *ouc_head;
1043 typedef int (*md_update_pack_t)(struct batch_update_head *head,
1044 struct lustre_msg *reqmsg,
1045 size_t *max_pack_size,
1046 struct md_op_item *item);
1049 struct lu_batch cbh_super;
1050 struct batch_update_head *cbh_head;
1053 struct lu_batch *cli_batch_create(struct obd_export *exp,
1054 enum lu_batch_flags flags, __u32 max_count);
1055 int cli_batch_stop(struct obd_export *exp, struct lu_batch *bh);
1056 int cli_batch_flush(struct obd_export *exp, struct lu_batch *bh, bool wait);
1057 int cli_batch_add(struct obd_export *exp, struct lu_batch *bh,
1058 struct md_op_item *item, md_update_pack_t packer,
1059 object_update_interpret_t interpreter);
1062 struct module *o_owner;
1063 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1064 void *karg, void __user *uarg);
1065 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1066 __u32 keylen, void *key, __u32 *vallen, void *val);
1067 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1068 __u32 keylen, void *key,
1069 __u32 vallen, void *val,
1070 struct ptlrpc_request_set *set);
1071 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
1072 int (*o_precleanup)(struct obd_device *obd);
1073 int (*o_cleanup)(struct obd_device *obd);
1074 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
1075 int (*o_postrecov)(struct obd_device *obd);
1076 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1078 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1079 /* connect to the target device with given connection
1080 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1081 * granted by the target, which are guaranteed to be a subset of flags
1082 * asked for. If @ocd == NULL, use default parameters.
1083 * On the client, localdata is a struct cl_client_cache *.
1084 * On the server, localdata is a struct lnet_nid *.
1086 int (*o_connect)(const struct lu_env *env,
1087 struct obd_export **exp, struct obd_device *src,
1088 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1090 int (*o_reconnect)(const struct lu_env *env,
1091 struct obd_export *exp, struct obd_device *src,
1092 struct obd_uuid *cluuid,
1093 struct obd_connect_data *ocd,
1095 int (*o_disconnect)(struct obd_export *exp);
1097 /* Initialize/finalize fids infrastructure. */
1098 int (*o_fid_init)(struct obd_device *obd,
1099 struct obd_export *exp, enum lu_cli_type type);
1100 int (*o_fid_fini)(struct obd_device *obd);
1102 /* Allocate new fid according to passed @hint. */
1103 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1104 struct lu_fid *fid, struct md_op_data *op_data);
1107 * Object with @fid is getting deleted, we may want to do something
1110 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1111 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1112 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1113 time64_t max_age, struct ptlrpc_request_set *set);
1114 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1116 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1118 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1120 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1122 int (*o_preprw)(const struct lu_env *env, int cmd,
1123 struct obd_export *exp, struct obdo *oa, int objcount,
1124 struct obd_ioobj *obj, struct niobuf_remote *remote,
1125 int *nr_pages, struct niobuf_local *local);
1126 int (*o_commitrw)(const struct lu_env *env, int cmd,
1127 struct obd_export *exp, struct obdo *oa,
1128 int objcount, struct obd_ioobj *obj,
1129 struct niobuf_remote *remote, int pages,
1130 struct niobuf_local *local, int rc, int nob,
1132 int (*o_init_export)(struct obd_export *exp);
1133 int (*o_destroy_export)(struct obd_export *exp);
1135 int (*o_import_event)(struct obd_device *, struct obd_import *,
1136 enum obd_import_event);
1138 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1139 enum obd_notify_event ev);
1141 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1142 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1145 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1146 struct obd_quotactl *);
1149 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1150 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1151 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1153 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1157 /* lmv structures */
1159 struct mdt_body *body;
1160 struct lu_buf layout;
1162 struct lmv_stripe_md *lmv;
1163 struct lmv_foreign_md *lfm;
1165 struct lmv_stripe_md *default_lmv;
1166 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1167 struct posix_acl *posix_acl;
1171 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1172 static inline void lmd_clear_acl(struct lustre_md *md)
1174 if (md->posix_acl) {
1175 posix_acl_release(md->posix_acl);
1176 md->posix_acl = NULL;
1180 #define OBD_CONNECT_ACL_FLAGS \
1181 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1183 static inline void lmd_clear_acl(struct lustre_md *md)
1187 #define OBD_CONNECT_ACL_FLAGS (0)
1190 struct md_open_data {
1191 struct obd_client_handle *mod_och;
1192 struct ptlrpc_request *mod_open_req;
1193 struct ptlrpc_request *mod_close_req;
1194 atomic_t mod_refcount;
1198 struct obd_client_handle {
1199 struct lustre_handle och_open_handle;
1200 struct lu_fid och_fid;
1201 struct md_open_data *och_mod;
1202 struct lustre_handle och_lease_handle; /* open lock for lease */
1207 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1209 struct lookup_intent;
1213 int (*m_close)(struct obd_export *, struct md_op_data *,
1214 struct md_open_data *, struct ptlrpc_request **);
1216 int (*m_create)(struct obd_export *, struct md_op_data *,
1217 const void *, size_t, umode_t, uid_t, gid_t,
1218 kernel_cap_t, __u64, struct ptlrpc_request **);
1220 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1221 const union ldlm_policy_data *, struct md_op_data *,
1222 struct lustre_handle *, __u64);
1224 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1225 struct ptlrpc_request **);
1227 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1228 struct lookup_intent *,
1229 struct ptlrpc_request **,
1230 ldlm_blocking_callback, __u64);
1232 int (*m_link)(struct obd_export *, struct md_op_data *,
1233 struct ptlrpc_request **);
1235 int (*m_rename)(struct obd_export *, struct md_op_data *,
1236 const char *, size_t, const char *, size_t,
1237 struct ptlrpc_request **);
1239 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1240 size_t , struct ptlrpc_request **);
1242 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1243 struct ptlrpc_request **);
1245 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1246 struct md_readdir_info *mrinfo, __u64 hash_offset,
1247 struct page **ppage);
1249 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1250 struct ptlrpc_request **);
1252 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1253 u64, const char *, const void *, size_t, unsigned int,
1254 u32, struct ptlrpc_request **);
1256 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1257 u64, const char *, size_t, struct ptlrpc_request **);
1259 int (*m_intent_getattr_async)(struct obd_export *,
1260 struct md_op_item *);
1262 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1263 struct lu_fid *, __u64 *bits);
1265 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1267 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1268 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1270 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1271 struct ptlrpc_request **);
1273 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1275 int (*m_get_lustre_md)(struct obd_export *, struct req_capsule *,
1276 struct obd_export *, struct obd_export *,
1277 struct lustre_md *);
1279 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1281 int (*m_merge_attr)(struct obd_export *,
1282 const struct lmv_stripe_md *lsm,
1283 struct cl_attr *attr, ldlm_blocking_callback);
1285 int (*m_set_open_replay_data)(struct obd_export *,
1286 struct obd_client_handle *,
1287 struct lookup_intent *);
1289 int (*m_clear_open_replay_data)(struct obd_export *,
1290 struct obd_client_handle *);
1292 int (*m_set_lock_data)(struct obd_export *,
1293 const struct lustre_handle *, void *, __u64 *);
1295 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1296 const struct lu_fid *, enum ldlm_type,
1297 union ldlm_policy_data *, enum ldlm_mode,
1298 struct lustre_handle *);
1300 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1301 union ldlm_policy_data *, enum ldlm_mode,
1302 enum ldlm_cancel_flags flags, void *opaque);
1304 int (*m_get_fid_from_lsm)(struct obd_export *,
1305 const struct lmv_stripe_md *,
1306 const char *name, int namelen,
1307 struct lu_fid *fid);
1308 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1309 const union lmv_mds_md *lmv, size_t lmv_size);
1310 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1311 struct ptlrpc_request_set *set);
1312 struct lu_batch *(*m_batch_create)(struct obd_export *exp,
1313 enum lu_batch_flags flags,
1315 int (*m_batch_stop)(struct obd_export *exp, struct lu_batch *bh);
1316 int (*m_batch_flush)(struct obd_export *exp, struct lu_batch *bh,
1318 int (*m_batch_add)(struct obd_export *exp, struct lu_batch *bh,
1319 struct md_op_item *item);
1322 static inline struct md_open_data *obd_mod_alloc(void)
1324 struct md_open_data *mod;
1328 atomic_set(&mod->mod_refcount, 1);
1332 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1333 #define obd_mod_put(mod) \
1335 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1336 if ((mod)->mod_open_req) \
1337 ptlrpc_req_finished((mod)->mod_open_req); \
1338 OBD_FREE_PTR(mod); \
1342 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1343 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1344 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1346 /* return 1 if client should be resend request */
1347 static inline int client_should_resend(int resend, struct client_obd *cli)
1349 return atomic_read(&cli->cl_resends) ?
1350 atomic_read(&cli->cl_resends) > resend : 1;
1354 * Return device name for this device
1356 * XXX: lu_device is declared before obd_device, while a pointer pointing
1357 * back to obd_device in lu_device, so this helper function defines here
1358 * instead of in lu_object.h
1360 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1362 return lu_dev->ld_obd->obd_name;
1365 static inline bool filename_is_volatile(const char *name, size_t namelen,
1371 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1374 /* caller does not care of idx */
1378 /* volatile file, the MDT can be set from name */
1379 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1380 /* if no MDT is specified, use std way */
1381 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1383 /* test for no MDT idx case */
1384 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1385 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1389 /* we have an idx, read it */
1390 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1391 rc = sscanf(start, "%4x:%4x:fd=%2d", idx, &rnd, &fd);
1392 /* error cases: no digit or negative value
1393 * rc will never be larger then 3
1395 if (rc <= 0 || *idx < 0)
1400 /* bad format of mdt idx, we cannot return an error
1401 * to caller so we use hash algo */
1402 CERROR("Bad volatile file name format: %s\n",
1403 name + LUSTRE_VOLATILE_HDR_LEN);
1407 static inline int cli_brw_size(struct obd_device *obd)
1409 LASSERT(obd != NULL);
1410 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1414 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1415 * of the client should be increased accordingly to avoid sending fragmented
1416 * RPCs over the network when the client runs out of the maximum dirty space
1417 * when so many RPCs are being generated.
1419 static inline void client_adjust_max_dirty(struct client_obd *cli)
1422 if (cli->cl_dirty_max_pages <= 0) {
1423 cli->cl_dirty_max_pages =
1424 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1426 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1427 cli->cl_max_pages_per_rpc;
1429 if (dirty_max > cli->cl_dirty_max_pages)
1430 cli->cl_dirty_max_pages = dirty_max;
1433 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1434 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1436 /* This value is exported to userspace through the max_dirty_mb
1437 * parameter. So we round up the number of pages to make it a round
1439 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1440 1 << (20 - PAGE_SHIFT));
1443 /* Must be used for page cache pages only,
1444 * not safe otherwise (e.g. direct IO pages)
1446 static inline struct inode *page2inode(struct page *page)
1448 if (page->mapping) {
1452 return page->mapping->host;
1458 #endif /* __OBD_H */