2 * pNFS functions to call and manage layout drivers.
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
8 * Dean Hildebrand <dhildebz@umich.edu>
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_page.h>
32 #include <linux/module.h>
33 #include <linux/sort.h>
37 #include "nfs4trace.h"
38 #include "delegation.h"
42 #define NFSDBG_FACILITY NFSDBG_PNFS
43 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
48 * protects pnfs_modules_tbl.
50 static DEFINE_SPINLOCK(pnfs_spinlock
);
53 * pnfs_modules_tbl holds all pnfs modules
55 static LIST_HEAD(pnfs_modules_tbl
);
57 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
);
58 static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
59 struct list_head
*free_me
,
60 const struct pnfs_layout_range
*range
,
62 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
63 struct list_head
*tmp_list
);
65 /* Return the registered pnfs layout driver module matching given id */
66 static struct pnfs_layoutdriver_type
*
67 find_pnfs_driver_locked(u32 id
)
69 struct pnfs_layoutdriver_type
*local
;
71 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
76 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
80 static struct pnfs_layoutdriver_type
*
81 find_pnfs_driver(u32 id
)
83 struct pnfs_layoutdriver_type
*local
;
85 spin_lock(&pnfs_spinlock
);
86 local
= find_pnfs_driver_locked(id
);
87 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
88 dprintk("%s: Could not grab reference on module\n", __func__
);
91 spin_unlock(&pnfs_spinlock
);
95 const struct pnfs_layoutdriver_type
*pnfs_find_layoutdriver(u32 id
)
97 return find_pnfs_driver(id
);
100 void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type
*ld
)
103 module_put(ld
->owner
);
107 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
109 if (nfss
->pnfs_curr_ld
) {
110 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
111 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
112 /* Decrement the MDS count. Purge the deviceid cache if zero */
113 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
114 nfs4_deviceid_purge_client(nfss
->nfs_client
);
115 module_put(nfss
->pnfs_curr_ld
->owner
);
117 nfss
->pnfs_curr_ld
= NULL
;
121 * When the server sends a list of layout types, we choose one in the order
122 * given in the list below.
124 * FIXME: should this list be configurable in some fashion? module param?
125 * mount option? something else?
127 static const u32 ld_prefs
[] = {
132 LAYOUT_NFSV4_1_FILES
,
137 ld_cmp(const void *e1
, const void *e2
)
139 u32 ld1
= *((u32
*)e1
);
140 u32 ld2
= *((u32
*)e2
);
143 for (i
= 0; ld_prefs
[i
] != 0; i
++) {
144 if (ld1
== ld_prefs
[i
])
147 if (ld2
== ld_prefs
[i
])
154 * Try to set the server's pnfs module to the pnfs layout type specified by id.
155 * Currently only one pNFS layout driver per filesystem is supported.
157 * @ids array of layout types supported by MDS.
160 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
161 struct nfs_fsinfo
*fsinfo
)
163 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
167 if (fsinfo
->nlayouttypes
== 0)
169 if (!(server
->nfs_client
->cl_exchange_flags
&
170 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
171 printk(KERN_ERR
"NFS: %s: cl_exchange_flags 0x%x\n",
172 __func__
, server
->nfs_client
->cl_exchange_flags
);
176 sort(fsinfo
->layouttype
, fsinfo
->nlayouttypes
,
177 sizeof(*fsinfo
->layouttype
), ld_cmp
, NULL
);
179 for (i
= 0; i
< fsinfo
->nlayouttypes
; i
++) {
180 id
= fsinfo
->layouttype
[i
];
181 ld_type
= find_pnfs_driver(id
);
183 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
,
185 ld_type
= find_pnfs_driver(id
);
192 dprintk("%s: No pNFS module found!\n", __func__
);
196 server
->pnfs_curr_ld
= ld_type
;
197 if (ld_type
->set_layoutdriver
198 && ld_type
->set_layoutdriver(server
, mntfh
)) {
199 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
200 "driver %u.\n", __func__
, id
);
201 module_put(ld_type
->owner
);
204 /* Bump the MDS count */
205 atomic_inc(&server
->nfs_client
->cl_mds_count
);
207 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
211 dprintk("%s: Using NFSv4 I/O\n", __func__
);
212 server
->pnfs_curr_ld
= NULL
;
216 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
218 int status
= -EINVAL
;
219 struct pnfs_layoutdriver_type
*tmp
;
221 if (ld_type
->id
== 0) {
222 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
225 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
226 printk(KERN_ERR
"NFS: %s Layout driver must provide "
227 "alloc_lseg and free_lseg.\n", __func__
);
231 spin_lock(&pnfs_spinlock
);
232 tmp
= find_pnfs_driver_locked(ld_type
->id
);
234 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
236 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
239 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
240 __func__
, ld_type
->id
);
242 spin_unlock(&pnfs_spinlock
);
246 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
249 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
251 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
252 spin_lock(&pnfs_spinlock
);
253 list_del(&ld_type
->pnfs_tblid
);
254 spin_unlock(&pnfs_spinlock
);
256 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
259 * pNFS client layout cache
262 /* Need to hold i_lock if caller does not already hold reference */
264 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
266 refcount_inc(&lo
->plh_refcount
);
269 static struct pnfs_layout_hdr
*
270 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
272 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
273 return ld
->alloc_layout_hdr(ino
, gfp_flags
);
277 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
279 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
280 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
282 if (test_and_clear_bit(NFS_LAYOUT_HASHED
, &lo
->plh_flags
)) {
283 struct nfs_client
*clp
= server
->nfs_client
;
285 spin_lock(&clp
->cl_lock
);
286 list_del_rcu(&lo
->plh_layouts
);
287 spin_unlock(&clp
->cl_lock
);
289 put_cred(lo
->plh_lc_cred
);
290 return ld
->free_layout_hdr(lo
);
294 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
296 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
297 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
299 /* Reset MDS Threshold I/O counters */
305 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
308 unsigned long i_state
;
312 inode
= lo
->plh_inode
;
313 pnfs_layoutreturn_before_put_layout_hdr(lo
);
315 if (refcount_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
316 if (!list_empty(&lo
->plh_segs
))
317 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
318 pnfs_detach_layout_hdr(lo
);
319 i_state
= inode
->i_state
;
320 spin_unlock(&inode
->i_lock
);
321 pnfs_free_layout_hdr(lo
);
322 /* Notify pnfs_destroy_layout_final() that we're done */
323 if (i_state
& (I_FREEING
| I_CLEAR
))
328 static struct inode
*
329 pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr
*lo
)
331 struct inode
*inode
= igrab(lo
->plh_inode
);
334 set_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
);
339 * Compare 2 layout stateid sequence ids, to see which is newer,
340 * taking into account wraparound issues.
342 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
344 return (s32
)(s1
- s2
) > 0;
347 static void pnfs_barrier_update(struct pnfs_layout_hdr
*lo
, u32 newseq
)
349 if (pnfs_seqid_is_newer(newseq
, lo
->plh_barrier
) || !lo
->plh_barrier
)
350 lo
->plh_barrier
= newseq
;
354 pnfs_set_plh_return_info(struct pnfs_layout_hdr
*lo
, enum pnfs_iomode iomode
,
357 if (lo
->plh_return_iomode
!= 0 && lo
->plh_return_iomode
!= iomode
)
359 lo
->plh_return_iomode
= iomode
;
360 set_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
362 * We must set lo->plh_return_seq to avoid livelocks with
363 * pnfs_layout_need_return()
366 seq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
367 if (!lo
->plh_return_seq
|| pnfs_seqid_is_newer(seq
, lo
->plh_return_seq
))
368 lo
->plh_return_seq
= seq
;
369 pnfs_barrier_update(lo
, seq
);
373 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr
*lo
)
375 struct pnfs_layout_segment
*lseg
;
376 lo
->plh_return_iomode
= 0;
377 lo
->plh_return_seq
= 0;
378 clear_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
379 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
380 if (!test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
382 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
386 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr
*lo
)
388 clear_bit_unlock(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
389 clear_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
);
390 smp_mb__after_atomic();
391 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
);
392 rpc_wake_up(&NFS_SERVER(lo
->plh_inode
)->roc_rpcwaitq
);
396 pnfs_clear_lseg_state(struct pnfs_layout_segment
*lseg
,
397 struct list_head
*free_me
)
399 clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
400 clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
401 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
402 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
403 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
404 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
408 * Update the seqid of a layout stateid after receiving
409 * NFS4ERR_OLD_STATEID
411 bool nfs4_layout_refresh_old_stateid(nfs4_stateid
*dst
,
412 struct pnfs_layout_range
*dst_range
,
415 struct pnfs_layout_hdr
*lo
;
416 struct pnfs_layout_range range
= {
417 .iomode
= IOMODE_ANY
,
419 .length
= NFS4_MAX_UINT64
,
425 spin_lock(&inode
->i_lock
);
426 lo
= NFS_I(inode
)->layout
;
427 if (lo
&& pnfs_layout_is_valid(lo
) &&
428 nfs4_stateid_match_other(dst
, &lo
->plh_stateid
)) {
429 /* Is our call using the most recent seqid? If so, bump it */
430 if (!nfs4_stateid_is_newer(&lo
->plh_stateid
, dst
)) {
431 nfs4_stateid_seqid_inc(dst
);
435 /* Try to update the seqid to the most recent */
436 err
= pnfs_mark_matching_lsegs_return(lo
, &head
, &range
, 0);
438 dst
->seqid
= lo
->plh_stateid
.seqid
;
444 spin_unlock(&inode
->i_lock
);
445 pnfs_free_lseg_list(&head
);
450 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
452 * In order to continue using the pnfs_layout_hdr, a full recovery
454 * Note that caller must hold inode->i_lock.
457 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr
*lo
,
458 struct list_head
*lseg_list
)
460 struct pnfs_layout_range range
= {
461 .iomode
= IOMODE_ANY
,
463 .length
= NFS4_MAX_UINT64
,
465 struct pnfs_layout_segment
*lseg
, *next
;
467 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
468 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
469 pnfs_clear_lseg_state(lseg
, lseg_list
);
470 pnfs_clear_layoutreturn_info(lo
);
471 pnfs_free_returned_lsegs(lo
, lseg_list
, &range
, 0);
472 set_bit(NFS_LAYOUT_DRAIN
, &lo
->plh_flags
);
473 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
) &&
474 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
475 pnfs_clear_layoutreturn_waitbit(lo
);
476 return !list_empty(&lo
->plh_segs
);
480 pnfs_iomode_to_fail_bit(u32 iomode
)
482 return iomode
== IOMODE_RW
?
483 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
487 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
489 lo
->plh_retry_timestamp
= jiffies
;
490 if (!test_and_set_bit(fail_bit
, &lo
->plh_flags
))
491 refcount_inc(&lo
->plh_refcount
);
495 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
497 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
498 refcount_dec(&lo
->plh_refcount
);
502 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
504 struct inode
*inode
= lo
->plh_inode
;
505 struct pnfs_layout_range range
= {
508 .length
= NFS4_MAX_UINT64
,
512 spin_lock(&inode
->i_lock
);
513 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
514 pnfs_mark_matching_lsegs_return(lo
, &head
, &range
, 0);
515 spin_unlock(&inode
->i_lock
);
516 pnfs_free_lseg_list(&head
);
517 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
518 iomode
== IOMODE_RW
? "RW" : "READ");
522 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
524 unsigned long start
, end
;
525 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
527 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
530 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
531 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
532 /* It is time to retry the failed layoutgets */
533 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
540 pnfs_init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
,
541 const struct pnfs_layout_range
*range
,
542 const nfs4_stateid
*stateid
)
544 INIT_LIST_HEAD(&lseg
->pls_list
);
545 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
546 INIT_LIST_HEAD(&lseg
->pls_commits
);
547 refcount_set(&lseg
->pls_refcount
, 1);
548 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
549 lseg
->pls_layout
= lo
;
550 lseg
->pls_range
= *range
;
551 lseg
->pls_seq
= be32_to_cpu(stateid
->seqid
);
554 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
557 struct inode
*inode
= lseg
->pls_layout
->plh_inode
;
558 NFS_SERVER(inode
)->pnfs_curr_ld
->free_lseg(lseg
);
563 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
564 struct pnfs_layout_segment
*lseg
)
566 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
567 list_del_init(&lseg
->pls_list
);
568 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
569 refcount_dec(&lo
->plh_refcount
);
570 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
572 if (list_empty(&lo
->plh_segs
) &&
573 !test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
574 !test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
575 if (atomic_read(&lo
->plh_outstanding
) == 0)
576 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
577 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
582 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr
*lo
,
583 struct pnfs_layout_segment
*lseg
)
585 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
586 pnfs_layout_is_valid(lo
)) {
587 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
588 list_move_tail(&lseg
->pls_list
, &lo
->plh_return_segs
);
595 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
597 struct pnfs_layout_hdr
*lo
;
603 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
604 refcount_read(&lseg
->pls_refcount
),
605 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
607 lo
= lseg
->pls_layout
;
608 inode
= lo
->plh_inode
;
610 if (refcount_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
611 pnfs_get_layout_hdr(lo
);
612 pnfs_layout_remove_lseg(lo
, lseg
);
613 if (pnfs_cache_lseg_for_layoutreturn(lo
, lseg
))
615 spin_unlock(&inode
->i_lock
);
616 pnfs_free_lseg(lseg
);
617 pnfs_put_layout_hdr(lo
);
620 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
623 * is l2 fully contained in l1?
625 * [----------------------------------)
630 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
631 const struct pnfs_layout_range
*l2
)
633 u64 start1
= l1
->offset
;
634 u64 end1
= pnfs_end_offset(start1
, l1
->length
);
635 u64 start2
= l2
->offset
;
636 u64 end2
= pnfs_end_offset(start2
, l2
->length
);
638 return (start1
<= start2
) && (end1
>= end2
);
641 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
642 struct list_head
*tmp_list
)
644 if (!refcount_dec_and_test(&lseg
->pls_refcount
))
646 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
647 list_add(&lseg
->pls_list
, tmp_list
);
651 /* Returns 1 if lseg is removed from list, 0 otherwise */
652 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
653 struct list_head
*tmp_list
)
657 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
658 /* Remove the reference keeping the lseg in the
659 * list. It will now be removed when all
660 * outstanding io is finished.
662 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
663 refcount_read(&lseg
->pls_refcount
));
664 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
671 pnfs_should_free_range(const struct pnfs_layout_range
*lseg_range
,
672 const struct pnfs_layout_range
*recall_range
)
674 return (recall_range
->iomode
== IOMODE_ANY
||
675 lseg_range
->iomode
== recall_range
->iomode
) &&
676 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
680 pnfs_match_lseg_recall(const struct pnfs_layout_segment
*lseg
,
681 const struct pnfs_layout_range
*recall_range
,
684 if (seq
!= 0 && pnfs_seqid_is_newer(lseg
->pls_seq
, seq
))
686 if (recall_range
== NULL
)
688 return pnfs_should_free_range(&lseg
->pls_range
, recall_range
);
692 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
693 * @lo: layout header containing the lsegs
694 * @tmp_list: list head where doomed lsegs should go
695 * @recall_range: optional recall range argument to match (may be NULL)
696 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
698 * Walk the list of lsegs in the layout header, and tear down any that should
699 * be destroyed. If "recall_range" is specified then the segment must match
700 * that range. If "seq" is non-zero, then only match segments that were handed
701 * out at or before that sequence.
703 * Returns number of matching invalid lsegs remaining in list after scanning
704 * it and purging them.
707 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
708 struct list_head
*tmp_list
,
709 const struct pnfs_layout_range
*recall_range
,
712 struct pnfs_layout_segment
*lseg
, *next
;
713 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
716 dprintk("%s:Begin lo %p\n", __func__
, lo
);
718 if (list_empty(&lo
->plh_segs
))
720 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
721 if (pnfs_match_lseg_recall(lseg
, recall_range
, seq
)) {
722 dprintk("%s: freeing lseg %p iomode %d seq %u "
723 "offset %llu length %llu\n", __func__
,
724 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_seq
,
725 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
726 if (mark_lseg_invalid(lseg
, tmp_list
))
729 pnfs_lseg_cancel_io(server
, lseg
);
731 dprintk("%s:Return %i\n", __func__
, remaining
);
736 pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
737 struct list_head
*free_me
,
738 const struct pnfs_layout_range
*range
,
741 struct pnfs_layout_segment
*lseg
, *next
;
743 list_for_each_entry_safe(lseg
, next
, &lo
->plh_return_segs
, pls_list
) {
744 if (pnfs_match_lseg_recall(lseg
, range
, seq
))
745 list_move_tail(&lseg
->pls_list
, free_me
);
749 /* note free_me must contain lsegs from a single layout_hdr */
751 pnfs_free_lseg_list(struct list_head
*free_me
)
753 struct pnfs_layout_segment
*lseg
, *tmp
;
755 if (list_empty(free_me
))
758 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
759 list_del(&lseg
->pls_list
);
760 pnfs_free_lseg(lseg
);
764 static struct pnfs_layout_hdr
*__pnfs_destroy_layout(struct nfs_inode
*nfsi
)
766 struct pnfs_layout_hdr
*lo
;
769 spin_lock(&nfsi
->vfs_inode
.i_lock
);
772 pnfs_get_layout_hdr(lo
);
773 pnfs_mark_layout_stateid_invalid(lo
, &tmp_list
);
774 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
775 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
776 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
777 pnfs_free_lseg_list(&tmp_list
);
778 nfs_commit_inode(&nfsi
->vfs_inode
, 0);
779 pnfs_put_layout_hdr(lo
);
781 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
785 void pnfs_destroy_layout(struct nfs_inode
*nfsi
)
787 __pnfs_destroy_layout(nfsi
);
789 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
791 static bool pnfs_layout_removed(struct nfs_inode
*nfsi
,
792 struct pnfs_layout_hdr
*lo
)
796 spin_lock(&nfsi
->vfs_inode
.i_lock
);
797 ret
= nfsi
->layout
!= lo
;
798 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
802 void pnfs_destroy_layout_final(struct nfs_inode
*nfsi
)
804 struct pnfs_layout_hdr
*lo
= __pnfs_destroy_layout(nfsi
);
807 wait_var_event(lo
, pnfs_layout_removed(nfsi
, lo
));
811 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
812 struct list_head
*layout_list
)
814 struct pnfs_layout_hdr
*lo
;
817 spin_lock(&inode
->i_lock
);
818 lo
= NFS_I(inode
)->layout
;
819 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
820 pnfs_get_layout_hdr(lo
);
821 list_add(&lo
->plh_bulk_destroy
, layout_list
);
824 spin_unlock(&inode
->i_lock
);
828 /* Caller must hold rcu_read_lock and clp->cl_lock */
830 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
831 struct nfs_server
*server
,
832 struct list_head
*layout_list
)
833 __must_hold(&clp
->cl_lock
)
836 struct pnfs_layout_hdr
*lo
, *next
;
839 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
840 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
) ||
841 test_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
) ||
842 !list_empty(&lo
->plh_bulk_destroy
))
844 /* If the sb is being destroyed, just bail */
845 if (!nfs_sb_active(server
->super
))
847 inode
= pnfs_grab_inode_layout_hdr(lo
);
849 if (test_and_clear_bit(NFS_LAYOUT_HASHED
, &lo
->plh_flags
))
850 list_del_rcu(&lo
->plh_layouts
);
851 if (pnfs_layout_add_bulk_destroy_list(inode
,
855 spin_unlock(&clp
->cl_lock
);
859 spin_unlock(&clp
->cl_lock
);
861 nfs_sb_deactive(server
->super
);
862 spin_lock(&clp
->cl_lock
);
870 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
873 struct pnfs_layout_hdr
*lo
;
875 LIST_HEAD(lseg_list
);
878 while (!list_empty(layout_list
)) {
879 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
881 dprintk("%s freeing layout for inode %lu\n", __func__
,
882 lo
->plh_inode
->i_ino
);
883 inode
= lo
->plh_inode
;
885 pnfs_layoutcommit_inode(inode
, false);
887 spin_lock(&inode
->i_lock
);
888 list_del_init(&lo
->plh_bulk_destroy
);
889 if (pnfs_mark_layout_stateid_invalid(lo
, &lseg_list
)) {
891 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
894 spin_unlock(&inode
->i_lock
);
895 pnfs_free_lseg_list(&lseg_list
);
896 /* Free all lsegs that are attached to commit buckets */
897 nfs_commit_inode(inode
, 0);
898 pnfs_put_layout_hdr(lo
);
899 nfs_iput_and_deactive(inode
);
905 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
906 struct nfs_fsid
*fsid
,
909 struct nfs_server
*server
;
910 LIST_HEAD(layout_list
);
912 spin_lock(&clp
->cl_lock
);
915 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
916 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
918 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
924 spin_unlock(&clp
->cl_lock
);
926 if (list_empty(&layout_list
))
928 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
932 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
935 struct nfs_server
*server
;
936 LIST_HEAD(layout_list
);
938 spin_lock(&clp
->cl_lock
);
941 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
942 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
948 spin_unlock(&clp
->cl_lock
);
950 if (list_empty(&layout_list
))
952 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
956 * Called by the state manager to remove all layouts established under an
960 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
962 nfs4_deviceid_mark_client_invalid(clp
);
963 nfs4_deviceid_purge_client(clp
);
965 pnfs_destroy_layouts_byclid(clp
, false);
969 pnfs_set_layout_cred(struct pnfs_layout_hdr
*lo
, const struct cred
*cred
)
971 const struct cred
*old
;
973 if (cred
&& cred_fscmp(lo
->plh_lc_cred
, cred
) != 0) {
974 old
= xchg(&lo
->plh_lc_cred
, get_cred(cred
));
979 /* update lo->plh_stateid with new if is more recent */
981 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
982 const struct cred
*cred
, bool update_barrier
)
984 u32 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
985 u32 newseq
= be32_to_cpu(new->seqid
);
987 if (!pnfs_layout_is_valid(lo
)) {
988 pnfs_set_layout_cred(lo
, cred
);
989 nfs4_stateid_copy(&lo
->plh_stateid
, new);
990 lo
->plh_barrier
= newseq
;
991 pnfs_clear_layoutreturn_info(lo
);
992 clear_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
996 if (pnfs_seqid_is_newer(newseq
, oldseq
))
997 nfs4_stateid_copy(&lo
->plh_stateid
, new);
999 if (update_barrier
) {
1000 pnfs_barrier_update(lo
, newseq
);
1004 * Because of wraparound, we want to keep the barrier
1005 * "close" to the current seqids. We really only want to
1006 * get here from a layoutget call.
1008 if (atomic_read(&lo
->plh_outstanding
) == 1)
1009 pnfs_barrier_update(lo
, be32_to_cpu(lo
->plh_stateid
.seqid
));
1013 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
1014 const nfs4_stateid
*stateid
)
1016 u32 seqid
= be32_to_cpu(stateid
->seqid
);
1018 return lo
->plh_barrier
&& pnfs_seqid_is_newer(lo
->plh_barrier
, seqid
);
1021 /* lget is set to 1 if called from inside send_layoutget call chain */
1023 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
)
1025 return lo
->plh_block_lgets
||
1026 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
1029 static struct nfs_server
*
1030 pnfs_find_server(struct inode
*inode
, struct nfs_open_context
*ctx
)
1032 struct nfs_server
*server
;
1035 server
= NFS_SERVER(inode
);
1037 struct dentry
*parent_dir
= dget_parent(ctx
->dentry
);
1038 server
= NFS_SERVER(parent_dir
->d_inode
);
1044 static void nfs4_free_pages(struct page
**pages
, size_t size
)
1051 for (i
= 0; i
< size
; i
++) {
1054 __free_page(pages
[i
]);
1059 static struct page
**nfs4_alloc_pages(size_t size
, gfp_t gfp_flags
)
1061 struct page
**pages
;
1064 pages
= kmalloc_array(size
, sizeof(struct page
*), gfp_flags
);
1066 dprintk("%s: can't alloc array of %zu pages\n", __func__
, size
);
1070 for (i
= 0; i
< size
; i
++) {
1071 pages
[i
] = alloc_page(gfp_flags
);
1073 dprintk("%s: failed to allocate page\n", __func__
);
1074 nfs4_free_pages(pages
, i
);
1082 static struct nfs4_layoutget
*
1083 pnfs_alloc_init_layoutget_args(struct inode
*ino
,
1084 struct nfs_open_context
*ctx
,
1085 const nfs4_stateid
*stateid
,
1086 const struct pnfs_layout_range
*range
,
1089 struct nfs_server
*server
= pnfs_find_server(ino
, ctx
);
1090 size_t max_reply_sz
= server
->pnfs_curr_ld
->max_layoutget_response
;
1091 size_t max_pages
= max_response_pages(server
);
1092 struct nfs4_layoutget
*lgp
;
1094 dprintk("--> %s\n", __func__
);
1096 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
1101 size_t npages
= (max_reply_sz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1102 if (npages
< max_pages
)
1106 lgp
->args
.layout
.pages
= nfs4_alloc_pages(max_pages
, gfp_flags
);
1107 if (!lgp
->args
.layout
.pages
) {
1111 lgp
->args
.layout
.pglen
= max_pages
* PAGE_SIZE
;
1112 lgp
->res
.layoutp
= &lgp
->args
.layout
;
1114 /* Don't confuse uninitialised result and success */
1115 lgp
->res
.status
= -NFS4ERR_DELAY
;
1117 lgp
->args
.minlength
= PAGE_SIZE
;
1118 if (lgp
->args
.minlength
> range
->length
)
1119 lgp
->args
.minlength
= range
->length
;
1121 loff_t i_size
= i_size_read(ino
);
1123 if (range
->iomode
== IOMODE_READ
) {
1124 if (range
->offset
>= i_size
)
1125 lgp
->args
.minlength
= 0;
1126 else if (i_size
- range
->offset
< lgp
->args
.minlength
)
1127 lgp
->args
.minlength
= i_size
- range
->offset
;
1130 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
1131 pnfs_copy_range(&lgp
->args
.range
, range
);
1132 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
1133 lgp
->args
.inode
= ino
;
1134 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
1135 nfs4_stateid_copy(&lgp
->args
.stateid
, stateid
);
1136 lgp
->gfp_flags
= gfp_flags
;
1137 lgp
->cred
= ctx
->cred
;
1141 void pnfs_layoutget_free(struct nfs4_layoutget
*lgp
)
1143 size_t max_pages
= lgp
->args
.layout
.pglen
/ PAGE_SIZE
;
1145 nfs4_free_pages(lgp
->args
.layout
.pages
, max_pages
);
1146 pnfs_put_layout_hdr(lgp
->lo
);
1147 put_nfs_open_context(lgp
->args
.ctx
);
1151 static void pnfs_clear_layoutcommit(struct inode
*inode
,
1152 struct list_head
*head
)
1154 struct nfs_inode
*nfsi
= NFS_I(inode
);
1155 struct pnfs_layout_segment
*lseg
, *tmp
;
1157 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1159 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
1160 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1162 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
1166 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr
*lo
,
1167 const nfs4_stateid
*arg_stateid
,
1168 const struct pnfs_layout_range
*range
,
1169 const nfs4_stateid
*stateid
)
1171 struct inode
*inode
= lo
->plh_inode
;
1174 spin_lock(&inode
->i_lock
);
1175 if (!pnfs_layout_is_valid(lo
) ||
1176 !nfs4_stateid_match_other(&lo
->plh_stateid
, arg_stateid
))
1179 u32 seq
= be32_to_cpu(arg_stateid
->seqid
);
1181 pnfs_mark_matching_lsegs_invalid(lo
, &freeme
, range
, seq
);
1182 pnfs_free_returned_lsegs(lo
, &freeme
, range
, seq
);
1183 pnfs_set_layout_stateid(lo
, stateid
, NULL
, true);
1185 pnfs_mark_layout_stateid_invalid(lo
, &freeme
);
1187 pnfs_clear_layoutreturn_waitbit(lo
);
1188 spin_unlock(&inode
->i_lock
);
1189 pnfs_free_lseg_list(&freeme
);
1194 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr
*lo
,
1195 nfs4_stateid
*stateid
,
1196 const struct cred
**cred
,
1197 enum pnfs_iomode
*iomode
)
1199 /* Serialise LAYOUTGET/LAYOUTRETURN */
1200 if (atomic_read(&lo
->plh_outstanding
) != 0)
1202 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
1204 set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
1205 pnfs_get_layout_hdr(lo
);
1206 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
1207 *cred
= get_cred(lo
->plh_lc_cred
);
1208 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
)) {
1209 if (lo
->plh_return_seq
!= 0)
1210 stateid
->seqid
= cpu_to_be32(lo
->plh_return_seq
);
1212 *iomode
= lo
->plh_return_iomode
;
1213 pnfs_clear_layoutreturn_info(lo
);
1214 } else if (iomode
!= NULL
)
1215 *iomode
= IOMODE_ANY
;
1216 pnfs_barrier_update(lo
, be32_to_cpu(stateid
->seqid
));
1221 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args
*args
,
1222 struct pnfs_layout_hdr
*lo
,
1223 const nfs4_stateid
*stateid
,
1224 enum pnfs_iomode iomode
)
1226 struct inode
*inode
= lo
->plh_inode
;
1228 args
->layout_type
= NFS_SERVER(inode
)->pnfs_curr_ld
->id
;
1229 args
->inode
= inode
;
1230 args
->range
.iomode
= iomode
;
1231 args
->range
.offset
= 0;
1232 args
->range
.length
= NFS4_MAX_UINT64
;
1234 nfs4_stateid_copy(&args
->stateid
, stateid
);
1238 pnfs_send_layoutreturn(struct pnfs_layout_hdr
*lo
,
1239 const nfs4_stateid
*stateid
,
1240 const struct cred
**pcred
,
1241 enum pnfs_iomode iomode
,
1244 struct inode
*ino
= lo
->plh_inode
;
1245 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1246 struct nfs4_layoutreturn
*lrp
;
1247 const struct cred
*cred
= *pcred
;
1251 lrp
= kzalloc(sizeof(*lrp
), nfs_io_gfp_mask());
1252 if (unlikely(lrp
== NULL
)) {
1254 spin_lock(&ino
->i_lock
);
1255 pnfs_clear_layoutreturn_waitbit(lo
);
1256 spin_unlock(&ino
->i_lock
);
1258 pnfs_put_layout_hdr(lo
);
1262 pnfs_init_layoutreturn_args(&lrp
->args
, lo
, stateid
, iomode
);
1263 lrp
->args
.ld_private
= &lrp
->ld_private
;
1264 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
1266 if (ld
->prepare_layoutreturn
)
1267 ld
->prepare_layoutreturn(&lrp
->args
);
1269 status
= nfs4_proc_layoutreturn(lrp
, sync
);
1271 dprintk("<-- %s status: %d\n", __func__
, status
);
1276 pnfs_layout_segments_returnable(struct pnfs_layout_hdr
*lo
,
1277 enum pnfs_iomode iomode
,
1280 struct pnfs_layout_range recall_range
= {
1281 .length
= NFS4_MAX_UINT64
,
1284 return pnfs_mark_matching_lsegs_return(lo
, &lo
->plh_return_segs
,
1285 &recall_range
, seq
) != -EBUSY
;
1288 /* Return true if layoutreturn is needed */
1290 pnfs_layout_need_return(struct pnfs_layout_hdr
*lo
)
1292 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1294 return pnfs_layout_segments_returnable(lo
, lo
->plh_return_iomode
,
1295 lo
->plh_return_seq
);
1298 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
1300 struct inode
*inode
= lo
->plh_inode
;
1302 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1304 spin_lock(&inode
->i_lock
);
1305 if (pnfs_layout_need_return(lo
)) {
1306 const struct cred
*cred
;
1307 nfs4_stateid stateid
;
1308 enum pnfs_iomode iomode
;
1311 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, &cred
, &iomode
);
1312 spin_unlock(&inode
->i_lock
);
1314 /* Send an async layoutreturn so we dont deadlock */
1315 pnfs_send_layoutreturn(lo
, &stateid
, &cred
, iomode
, false);
1318 spin_unlock(&inode
->i_lock
);
1322 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1323 * when the layout segment list is empty.
1325 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1326 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1327 * deviceid is marked invalid.
1330 _pnfs_return_layout(struct inode
*ino
)
1332 struct pnfs_layout_hdr
*lo
= NULL
;
1333 struct nfs_inode
*nfsi
= NFS_I(ino
);
1334 struct pnfs_layout_range range
= {
1335 .iomode
= IOMODE_ANY
,
1337 .length
= NFS4_MAX_UINT64
,
1339 LIST_HEAD(tmp_list
);
1340 const struct cred
*cred
;
1341 nfs4_stateid stateid
;
1343 bool send
, valid_layout
;
1345 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
1347 spin_lock(&ino
->i_lock
);
1350 spin_unlock(&ino
->i_lock
);
1351 dprintk("NFS: %s no layout to return\n", __func__
);
1354 /* Reference matched in nfs4_layoutreturn_release */
1355 pnfs_get_layout_hdr(lo
);
1356 /* Is there an outstanding layoutreturn ? */
1357 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1358 spin_unlock(&ino
->i_lock
);
1359 if (wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1360 TASK_UNINTERRUPTIBLE
))
1361 goto out_put_layout_hdr
;
1362 spin_lock(&ino
->i_lock
);
1364 valid_layout
= pnfs_layout_is_valid(lo
);
1365 pnfs_clear_layoutcommit(ino
, &tmp_list
);
1366 pnfs_mark_matching_lsegs_return(lo
, &tmp_list
, &range
, 0);
1368 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
)
1369 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
, &range
);
1371 /* Don't send a LAYOUTRETURN if list was initially empty */
1372 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) ||
1374 spin_unlock(&ino
->i_lock
);
1375 dprintk("NFS: %s no layout segments to return\n", __func__
);
1376 goto out_wait_layoutreturn
;
1379 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, &cred
, NULL
);
1380 spin_unlock(&ino
->i_lock
);
1382 status
= pnfs_send_layoutreturn(lo
, &stateid
, &cred
, IOMODE_ANY
, true);
1383 out_wait_layoutreturn
:
1384 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
, TASK_UNINTERRUPTIBLE
);
1386 pnfs_free_lseg_list(&tmp_list
);
1387 pnfs_put_layout_hdr(lo
);
1389 dprintk("<-- %s status: %d\n", __func__
, status
);
1394 pnfs_commit_and_return_layout(struct inode
*inode
)
1396 struct pnfs_layout_hdr
*lo
;
1399 spin_lock(&inode
->i_lock
);
1400 lo
= NFS_I(inode
)->layout
;
1402 spin_unlock(&inode
->i_lock
);
1405 pnfs_get_layout_hdr(lo
);
1406 /* Block new layoutgets and read/write to ds */
1407 lo
->plh_block_lgets
++;
1408 spin_unlock(&inode
->i_lock
);
1409 filemap_fdatawait(inode
->i_mapping
);
1410 ret
= pnfs_layoutcommit_inode(inode
, true);
1412 ret
= _pnfs_return_layout(inode
);
1413 spin_lock(&inode
->i_lock
);
1414 lo
->plh_block_lgets
--;
1415 spin_unlock(&inode
->i_lock
);
1416 pnfs_put_layout_hdr(lo
);
1420 bool pnfs_roc(struct inode
*ino
,
1421 struct nfs4_layoutreturn_args
*args
,
1422 struct nfs4_layoutreturn_res
*res
,
1423 const struct cred
*cred
)
1425 struct nfs_inode
*nfsi
= NFS_I(ino
);
1426 struct nfs_open_context
*ctx
;
1427 struct nfs4_state
*state
;
1428 struct pnfs_layout_hdr
*lo
;
1429 struct pnfs_layout_segment
*lseg
, *next
;
1430 const struct cred
*lc_cred
;
1431 nfs4_stateid stateid
;
1432 enum pnfs_iomode iomode
= 0;
1433 bool layoutreturn
= false, roc
= false;
1434 bool skip_read
= false;
1436 if (!nfs_have_layout(ino
))
1440 spin_lock(&ino
->i_lock
);
1442 if (!lo
|| !pnfs_layout_is_valid(lo
) ||
1443 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1447 pnfs_get_layout_hdr(lo
);
1448 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1449 spin_unlock(&ino
->i_lock
);
1451 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1452 TASK_UNINTERRUPTIBLE
);
1453 pnfs_put_layout_hdr(lo
);
1457 /* no roc if we hold a delegation */
1458 if (nfs4_check_delegation(ino
, FMODE_READ
)) {
1459 if (nfs4_check_delegation(ino
, FMODE_WRITE
))
1464 list_for_each_entry_rcu(ctx
, &nfsi
->open_files
, list
) {
1468 /* Don't return layout if there is open file state */
1469 if (state
->state
& FMODE_WRITE
)
1471 if (state
->state
& FMODE_READ
)
1476 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
) {
1477 if (skip_read
&& lseg
->pls_range
.iomode
== IOMODE_READ
)
1479 /* If we are sending layoutreturn, invalidate all valid lsegs */
1480 if (!test_and_clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
))
1483 * Note: mark lseg for return so pnfs_layout_remove_lseg
1484 * doesn't invalidate the layout for us.
1486 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
1487 if (!mark_lseg_invalid(lseg
, &lo
->plh_return_segs
))
1489 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
1492 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1495 /* ROC in two conditions:
1496 * 1. there are ROC lsegs
1497 * 2. we don't send layoutreturn
1499 /* lo ref dropped in pnfs_roc_release() */
1500 layoutreturn
= pnfs_prepare_layoutreturn(lo
, &stateid
, &lc_cred
, &iomode
);
1501 /* If the creds don't match, we can't compound the layoutreturn */
1502 if (!layoutreturn
|| cred_fscmp(cred
, lc_cred
) != 0)
1506 pnfs_init_layoutreturn_args(args
, lo
, &stateid
, iomode
);
1507 res
->lrs_present
= 0;
1508 layoutreturn
= false;
1512 spin_unlock(&ino
->i_lock
);
1514 pnfs_layoutcommit_inode(ino
, true);
1516 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1517 if (ld
->prepare_layoutreturn
)
1518 ld
->prepare_layoutreturn(args
);
1519 pnfs_put_layout_hdr(lo
);
1523 pnfs_send_layoutreturn(lo
, &stateid
, &lc_cred
, iomode
, true);
1524 pnfs_put_layout_hdr(lo
);
1528 int pnfs_roc_done(struct rpc_task
*task
, struct nfs4_layoutreturn_args
**argpp
,
1529 struct nfs4_layoutreturn_res
**respp
, int *ret
)
1531 struct nfs4_layoutreturn_args
*arg
= *argpp
;
1532 int retval
= -EAGAIN
;
1536 /* Handle Layoutreturn errors */
1541 case -NFS4ERR_NOMATCHING_LAYOUT
:
1542 /* Was there an RPC level error? If not, retry */
1543 if (task
->tk_rpc_status
== 0)
1545 /* If the call was not sent, let caller handle it */
1546 if (!RPC_WAS_SENT(task
))
1549 * Otherwise, assume the call succeeded and
1550 * that we need to release the layout
1553 (*respp
)->lrs_present
= 0;
1556 case -NFS4ERR_DELAY
:
1557 /* Let the caller handle the retry */
1558 *ret
= -NFS4ERR_NOMATCHING_LAYOUT
;
1560 case -NFS4ERR_OLD_STATEID
:
1561 if (!nfs4_layout_refresh_old_stateid(&arg
->stateid
,
1562 &arg
->range
, arg
->inode
))
1564 *ret
= -NFS4ERR_NOMATCHING_LAYOUT
;
1572 void pnfs_roc_release(struct nfs4_layoutreturn_args
*args
,
1573 struct nfs4_layoutreturn_res
*res
,
1576 struct pnfs_layout_hdr
*lo
= args
->layout
;
1577 struct inode
*inode
= args
->inode
;
1578 const nfs4_stateid
*res_stateid
= NULL
;
1579 struct nfs4_xdr_opaque_data
*ld_private
= args
->ld_private
;
1582 case -NFS4ERR_NOMATCHING_LAYOUT
:
1583 spin_lock(&inode
->i_lock
);
1584 if (pnfs_layout_is_valid(lo
) &&
1585 nfs4_stateid_match_other(&args
->stateid
, &lo
->plh_stateid
))
1586 pnfs_set_plh_return_info(lo
, args
->range
.iomode
, 0);
1587 pnfs_clear_layoutreturn_waitbit(lo
);
1588 spin_unlock(&inode
->i_lock
);
1591 if (res
->lrs_present
)
1592 res_stateid
= &res
->stateid
;
1595 pnfs_layoutreturn_free_lsegs(lo
, &args
->stateid
, &args
->range
,
1598 trace_nfs4_layoutreturn_on_close(args
->inode
, &args
->stateid
, ret
);
1599 if (ld_private
&& ld_private
->ops
&& ld_private
->ops
->free
)
1600 ld_private
->ops
->free(ld_private
);
1601 pnfs_put_layout_hdr(lo
);
1604 bool pnfs_wait_on_layoutreturn(struct inode
*ino
, struct rpc_task
*task
)
1606 struct nfs_inode
*nfsi
= NFS_I(ino
);
1607 struct pnfs_layout_hdr
*lo
;
1610 /* we might not have grabbed lo reference. so need to check under
1612 spin_lock(&ino
->i_lock
);
1614 if (lo
&& test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1615 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1618 spin_unlock(&ino
->i_lock
);
1623 * Compare two layout segments for sorting into layout cache.
1624 * We want to preferentially return RW over RO layouts, so ensure those
1628 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1629 const struct pnfs_layout_range
*l2
)
1633 /* high offset > low offset */
1634 d
= l1
->offset
- l2
->offset
;
1638 /* short length > long length */
1639 d
= l2
->length
- l1
->length
;
1643 /* read > read/write */
1644 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1648 pnfs_lseg_range_is_after(const struct pnfs_layout_range
*l1
,
1649 const struct pnfs_layout_range
*l2
)
1651 return pnfs_lseg_range_cmp(l1
, l2
) > 0;
1655 pnfs_lseg_no_merge(struct pnfs_layout_segment
*lseg
,
1656 struct pnfs_layout_segment
*old
)
1662 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1663 struct pnfs_layout_segment
*lseg
,
1664 bool (*is_after
)(const struct pnfs_layout_range
*,
1665 const struct pnfs_layout_range
*),
1666 bool (*do_merge
)(struct pnfs_layout_segment
*,
1667 struct pnfs_layout_segment
*),
1668 struct list_head
*free_me
)
1670 struct pnfs_layout_segment
*lp
, *tmp
;
1672 dprintk("%s:Begin\n", __func__
);
1674 list_for_each_entry_safe(lp
, tmp
, &lo
->plh_segs
, pls_list
) {
1675 if (test_bit(NFS_LSEG_VALID
, &lp
->pls_flags
) == 0)
1677 if (do_merge(lseg
, lp
)) {
1678 mark_lseg_invalid(lp
, free_me
);
1681 if (is_after(&lseg
->pls_range
, &lp
->pls_range
))
1683 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1684 dprintk("%s: inserted lseg %p "
1685 "iomode %d offset %llu length %llu before "
1686 "lp %p iomode %d offset %llu length %llu\n",
1687 __func__
, lseg
, lseg
->pls_range
.iomode
,
1688 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1689 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1690 lp
->pls_range
.length
);
1693 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1694 dprintk("%s: inserted lseg %p "
1695 "iomode %d offset %llu length %llu at tail\n",
1696 __func__
, lseg
, lseg
->pls_range
.iomode
,
1697 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1699 pnfs_get_layout_hdr(lo
);
1701 dprintk("%s:Return\n", __func__
);
1703 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg
);
1706 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1707 struct pnfs_layout_segment
*lseg
,
1708 struct list_head
*free_me
)
1710 struct inode
*inode
= lo
->plh_inode
;
1711 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
1713 if (ld
->add_lseg
!= NULL
)
1714 ld
->add_lseg(lo
, lseg
, free_me
);
1716 pnfs_generic_layout_insert_lseg(lo
, lseg
,
1717 pnfs_lseg_range_is_after
,
1722 static struct pnfs_layout_hdr
*
1723 alloc_init_layout_hdr(struct inode
*ino
,
1724 struct nfs_open_context
*ctx
,
1727 struct pnfs_layout_hdr
*lo
;
1729 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1732 refcount_set(&lo
->plh_refcount
, 1);
1733 INIT_LIST_HEAD(&lo
->plh_layouts
);
1734 INIT_LIST_HEAD(&lo
->plh_segs
);
1735 INIT_LIST_HEAD(&lo
->plh_return_segs
);
1736 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1737 lo
->plh_inode
= ino
;
1738 lo
->plh_lc_cred
= get_cred(ctx
->cred
);
1739 lo
->plh_flags
|= 1 << NFS_LAYOUT_INVALID_STID
;
1743 static struct pnfs_layout_hdr
*
1744 pnfs_find_alloc_layout(struct inode
*ino
,
1745 struct nfs_open_context
*ctx
,
1747 __releases(&ino
->i_lock
)
1748 __acquires(&ino
->i_lock
)
1750 struct nfs_inode
*nfsi
= NFS_I(ino
);
1751 struct pnfs_layout_hdr
*new = NULL
;
1753 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1755 if (nfsi
->layout
!= NULL
)
1757 spin_unlock(&ino
->i_lock
);
1758 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1759 spin_lock(&ino
->i_lock
);
1761 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1764 } else if (new != NULL
)
1765 pnfs_free_layout_hdr(new);
1767 pnfs_get_layout_hdr(nfsi
->layout
);
1768 return nfsi
->layout
;
1772 * iomode matching rules:
1773 * iomode lseg strict match
1775 * ----- ----- ------ -----
1780 * READ READ N/A true
1781 * READ RW true false
1782 * READ RW false true
1785 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1786 const struct pnfs_layout_range
*range
,
1789 struct pnfs_layout_range range1
;
1791 if ((range
->iomode
== IOMODE_RW
&&
1792 ls_range
->iomode
!= IOMODE_RW
) ||
1793 (range
->iomode
!= ls_range
->iomode
&&
1795 !pnfs_lseg_range_intersecting(ls_range
, range
))
1798 /* range1 covers only the first byte in the range */
1801 return pnfs_lseg_range_contained(ls_range
, &range1
);
1805 * lookup range in layout
1807 static struct pnfs_layout_segment
*
1808 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1809 struct pnfs_layout_range
*range
,
1812 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1814 dprintk("%s:Begin\n", __func__
);
1816 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1817 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1818 pnfs_lseg_range_match(&lseg
->pls_range
, range
,
1820 ret
= pnfs_get_lseg(lseg
);
1825 dprintk("%s:Return lseg %p ref %d\n",
1826 __func__
, ret
, ret
? refcount_read(&ret
->pls_refcount
) : 0);
1831 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1832 * to the MDS or over pNFS
1834 * The nfs_inode read_io and write_io fields are cumulative counters reset
1835 * when there are no layout segments. Note that in pnfs_update_layout iomode
1836 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1839 * A return of true means use MDS I/O.
1842 * If a file's size is smaller than the file size threshold, data accesses
1843 * SHOULD be sent to the metadata server. If an I/O request has a length that
1844 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1845 * server. If both file size and I/O size are provided, the client SHOULD
1846 * reach or exceed both thresholds before sending its read or write
1847 * requests to the data server.
1849 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1850 struct inode
*ino
, int iomode
)
1852 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1853 struct nfs_inode
*nfsi
= NFS_I(ino
);
1854 loff_t fsize
= i_size_read(ino
);
1855 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1860 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1861 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1865 if (t
->bm
& THRESHOLD_RD
) {
1866 dprintk("%s fsize %llu\n", __func__
, fsize
);
1868 if (fsize
< t
->rd_sz
)
1871 if (t
->bm
& THRESHOLD_RD_IO
) {
1872 dprintk("%s nfsi->read_io %llu\n", __func__
,
1875 if (nfsi
->read_io
< t
->rd_io_sz
)
1880 if (t
->bm
& THRESHOLD_WR
) {
1881 dprintk("%s fsize %llu\n", __func__
, fsize
);
1883 if (fsize
< t
->wr_sz
)
1886 if (t
->bm
& THRESHOLD_WR_IO
) {
1887 dprintk("%s nfsi->write_io %llu\n", __func__
,
1890 if (nfsi
->write_io
< t
->wr_io_sz
)
1895 if (size_set
&& io_set
) {
1898 } else if (size
|| io
)
1901 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1905 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr
*lo
)
1908 * send layoutcommit as it can hold up layoutreturn due to lseg
1911 pnfs_layoutcommit_inode(lo
->plh_inode
, false);
1912 return wait_on_bit_action(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1913 nfs_wait_bit_killable
,
1914 TASK_KILLABLE
|TASK_FREEZABLE_UNSAFE
);
1917 static void nfs_layoutget_begin(struct pnfs_layout_hdr
*lo
)
1919 atomic_inc(&lo
->plh_outstanding
);
1922 static void nfs_layoutget_end(struct pnfs_layout_hdr
*lo
)
1924 if (atomic_dec_and_test(&lo
->plh_outstanding
) &&
1925 test_and_clear_bit(NFS_LAYOUT_DRAIN
, &lo
->plh_flags
))
1926 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_DRAIN
);
1929 static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr
*lo
)
1931 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET
, &lo
->plh_flags
);
1934 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr
*lo
)
1936 unsigned long *bitlock
= &lo
->plh_flags
;
1938 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET
, bitlock
);
1939 smp_mb__after_atomic();
1940 wake_up_bit(bitlock
, NFS_LAYOUT_FIRST_LAYOUTGET
);
1943 static void _add_to_server_list(struct pnfs_layout_hdr
*lo
,
1944 struct nfs_server
*server
)
1946 if (!test_and_set_bit(NFS_LAYOUT_HASHED
, &lo
->plh_flags
)) {
1947 struct nfs_client
*clp
= server
->nfs_client
;
1949 /* The lo must be on the clp list if there is any
1950 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1952 spin_lock(&clp
->cl_lock
);
1953 list_add_tail_rcu(&lo
->plh_layouts
, &server
->layouts
);
1954 spin_unlock(&clp
->cl_lock
);
1959 * Layout segment is retreived from the server if not cached.
1960 * The appropriate layout segment is referenced and returned to the caller.
1962 struct pnfs_layout_segment
*
1963 pnfs_update_layout(struct inode
*ino
,
1964 struct nfs_open_context
*ctx
,
1967 enum pnfs_iomode iomode
,
1971 struct pnfs_layout_range arg
= {
1977 struct nfs_server
*server
= NFS_SERVER(ino
);
1978 struct nfs_client
*clp
= server
->nfs_client
;
1979 struct pnfs_layout_hdr
*lo
= NULL
;
1980 struct pnfs_layout_segment
*lseg
= NULL
;
1981 struct nfs4_layoutget
*lgp
;
1982 nfs4_stateid stateid
;
1984 unsigned long giveup
= jiffies
+ (clp
->cl_lease_time
<< 1);
1987 if (!pnfs_enabled_sb(NFS_SERVER(ino
))) {
1988 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1989 PNFS_UPDATE_LAYOUT_NO_PNFS
);
1993 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
)) {
1994 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1995 PNFS_UPDATE_LAYOUT_MDSTHRESH
);
2000 lseg
= ERR_PTR(nfs4_client_recover_expired_lease(clp
));
2004 spin_lock(&ino
->i_lock
);
2005 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
2007 spin_unlock(&ino
->i_lock
);
2008 lseg
= ERR_PTR(-ENOMEM
);
2009 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2010 PNFS_UPDATE_LAYOUT_NOMEM
);
2014 /* Do we even need to bother with this? */
2015 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
2016 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2017 PNFS_UPDATE_LAYOUT_BULK_RECALL
);
2018 dprintk("%s matches recall, use MDS\n", __func__
);
2022 /* if LAYOUTGET already failed once we don't try again */
2023 if (pnfs_layout_io_test_failed(lo
, iomode
)) {
2024 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2025 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL
);
2030 * If the layout segment list is empty, but there are outstanding
2031 * layoutget calls, then they might be subject to a layoutrecall.
2033 if (test_bit(NFS_LAYOUT_DRAIN
, &lo
->plh_flags
) &&
2034 atomic_read(&lo
->plh_outstanding
) != 0) {
2035 spin_unlock(&ino
->i_lock
);
2036 lseg
= ERR_PTR(wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_DRAIN
,
2039 goto out_put_layout_hdr
;
2040 pnfs_put_layout_hdr(lo
);
2045 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2048 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
2049 spin_unlock(&ino
->i_lock
);
2050 dprintk("%s wait for layoutreturn\n", __func__
);
2051 lseg
= ERR_PTR(pnfs_prepare_to_retry_layoutget(lo
));
2052 if (!IS_ERR(lseg
)) {
2053 pnfs_put_layout_hdr(lo
);
2054 dprintk("%s retrying\n", __func__
);
2055 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
,
2057 PNFS_UPDATE_LAYOUT_RETRY
);
2060 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2061 PNFS_UPDATE_LAYOUT_RETURN
);
2062 goto out_put_layout_hdr
;
2065 lseg
= pnfs_find_lseg(lo
, &arg
, strict_iomode
);
2067 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2068 PNFS_UPDATE_LAYOUT_FOUND_CACHED
);
2073 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2074 * stateid, or it has been invalidated, then we must use the open
2077 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
)) {
2081 * The first layoutget for the file. Need to serialize per
2082 * RFC 5661 Errata 3208.
2084 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
,
2086 spin_unlock(&ino
->i_lock
);
2087 lseg
= ERR_PTR(wait_on_bit(&lo
->plh_flags
,
2088 NFS_LAYOUT_FIRST_LAYOUTGET
,
2091 goto out_put_layout_hdr
;
2092 pnfs_put_layout_hdr(lo
);
2093 dprintk("%s retrying\n", __func__
);
2097 spin_unlock(&ino
->i_lock
);
2099 status
= nfs4_select_rw_stateid(ctx
->state
,
2100 iomode
== IOMODE_RW
? FMODE_WRITE
: FMODE_READ
,
2101 NULL
, &stateid
, NULL
);
2103 lseg
= ERR_PTR(status
);
2104 trace_pnfs_update_layout(ino
, pos
, count
,
2106 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
2107 nfs4_schedule_stateid_recovery(server
, ctx
->state
);
2108 pnfs_clear_first_layoutget(lo
);
2109 pnfs_put_layout_hdr(lo
);
2112 spin_lock(&ino
->i_lock
);
2114 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
2117 if (pnfs_layoutgets_blocked(lo
)) {
2118 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2119 PNFS_UPDATE_LAYOUT_BLOCKED
);
2122 nfs_layoutget_begin(lo
);
2123 spin_unlock(&ino
->i_lock
);
2125 _add_to_server_list(lo
, server
);
2127 pg_offset
= arg
.offset
& ~PAGE_MASK
;
2129 arg
.offset
-= pg_offset
;
2130 arg
.length
+= pg_offset
;
2132 if (arg
.length
!= NFS4_MAX_UINT64
)
2133 arg
.length
= PAGE_ALIGN(arg
.length
);
2135 lgp
= pnfs_alloc_init_layoutget_args(ino
, ctx
, &stateid
, &arg
, gfp_flags
);
2137 lseg
= ERR_PTR(-ENOMEM
);
2138 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, NULL
,
2139 PNFS_UPDATE_LAYOUT_NOMEM
);
2140 nfs_layoutget_end(lo
);
2141 goto out_put_layout_hdr
;
2145 pnfs_get_layout_hdr(lo
);
2147 lseg
= nfs4_proc_layoutget(lgp
, &timeout
);
2148 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2149 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET
);
2150 nfs_layoutget_end(lo
);
2152 switch(PTR_ERR(lseg
)) {
2154 if (time_after(jiffies
, giveup
))
2157 case -ERECALLCONFLICT
:
2161 /* The server returned NFS4ERR_LAYOUTUNAVAILABLE */
2162 pnfs_layout_set_fail_bit(
2163 lo
, pnfs_iomode_to_fail_bit(iomode
));
2165 goto out_put_layout_hdr
;
2167 if (!nfs_error_is_fatal(PTR_ERR(lseg
))) {
2168 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2171 goto out_put_layout_hdr
;
2175 pnfs_clear_first_layoutget(lo
);
2176 trace_pnfs_update_layout(ino
, pos
, count
,
2177 iomode
, lo
, lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
2178 pnfs_put_layout_hdr(lo
);
2182 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2187 pnfs_clear_first_layoutget(lo
);
2188 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2189 PNFS_UPDATE_LAYOUT_EXIT
);
2190 pnfs_put_layout_hdr(lo
);
2192 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2193 "(%s, offset: %llu, length: %llu)\n",
2194 __func__
, ino
->i_sb
->s_id
,
2195 (unsigned long long)NFS_FILEID(ino
),
2196 IS_ERR_OR_NULL(lseg
) ? "not found" : "found",
2197 iomode
==IOMODE_RW
? "read/write" : "read-only",
2198 (unsigned long long)pos
,
2199 (unsigned long long)count
);
2202 spin_unlock(&ino
->i_lock
);
2203 goto out_put_layout_hdr
;
2205 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
2208 pnfs_sanity_check_layout_range(struct pnfs_layout_range
*range
)
2210 switch (range
->iomode
) {
2217 if (range
->offset
== NFS4_MAX_UINT64
)
2219 if (range
->length
== 0)
2221 if (range
->length
!= NFS4_MAX_UINT64
&&
2222 range
->length
> NFS4_MAX_UINT64
- range
->offset
)
2227 static struct pnfs_layout_hdr
*
2228 _pnfs_grab_empty_layout(struct inode
*ino
, struct nfs_open_context
*ctx
)
2230 struct pnfs_layout_hdr
*lo
;
2232 spin_lock(&ino
->i_lock
);
2233 lo
= pnfs_find_alloc_layout(ino
, ctx
, nfs_io_gfp_mask());
2236 if (!test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
))
2238 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
2240 if (pnfs_layoutgets_blocked(lo
))
2242 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
, &lo
->plh_flags
))
2244 nfs_layoutget_begin(lo
);
2245 spin_unlock(&ino
->i_lock
);
2246 _add_to_server_list(lo
, NFS_SERVER(ino
));
2250 spin_unlock(&ino
->i_lock
);
2251 pnfs_put_layout_hdr(lo
);
2255 static void _lgopen_prepare_attached(struct nfs4_opendata
*data
,
2256 struct nfs_open_context
*ctx
)
2258 struct inode
*ino
= data
->dentry
->d_inode
;
2259 struct pnfs_layout_range rng
= {
2260 .iomode
= (data
->o_arg
.fmode
& FMODE_WRITE
) ?
2261 IOMODE_RW
: IOMODE_READ
,
2263 .length
= NFS4_MAX_UINT64
,
2265 struct nfs4_layoutget
*lgp
;
2266 struct pnfs_layout_hdr
*lo
;
2268 /* Heuristic: don't send layoutget if we have cached data */
2269 if (rng
.iomode
== IOMODE_READ
&&
2270 (i_size_read(ino
) == 0 || ino
->i_mapping
->nrpages
!= 0))
2273 lo
= _pnfs_grab_empty_layout(ino
, ctx
);
2276 lgp
= pnfs_alloc_init_layoutget_args(ino
, ctx
, ¤t_stateid
, &rng
,
2279 pnfs_clear_first_layoutget(lo
);
2280 nfs_layoutget_end(lo
);
2281 pnfs_put_layout_hdr(lo
);
2286 data
->o_arg
.lg_args
= &lgp
->args
;
2287 data
->o_res
.lg_res
= &lgp
->res
;
2290 static void _lgopen_prepare_floating(struct nfs4_opendata
*data
,
2291 struct nfs_open_context
*ctx
)
2293 struct inode
*ino
= data
->dentry
->d_inode
;
2294 struct pnfs_layout_range rng
= {
2295 .iomode
= (data
->o_arg
.fmode
& FMODE_WRITE
) ?
2296 IOMODE_RW
: IOMODE_READ
,
2298 .length
= NFS4_MAX_UINT64
,
2300 struct nfs4_layoutget
*lgp
;
2302 lgp
= pnfs_alloc_init_layoutget_args(ino
, ctx
, ¤t_stateid
, &rng
,
2307 data
->o_arg
.lg_args
= &lgp
->args
;
2308 data
->o_res
.lg_res
= &lgp
->res
;
2311 void pnfs_lgopen_prepare(struct nfs4_opendata
*data
,
2312 struct nfs_open_context
*ctx
)
2314 struct nfs_server
*server
= NFS_SERVER(data
->dir
->d_inode
);
2316 if (!(pnfs_enabled_sb(server
) &&
2317 server
->pnfs_curr_ld
->flags
& PNFS_LAYOUTGET_ON_OPEN
))
2319 /* Could check on max_ops, but currently hardcoded high enough */
2320 if (!nfs_server_capable(data
->dir
->d_inode
, NFS_CAP_LGOPEN
))
2325 _lgopen_prepare_attached(data
, ctx
);
2327 _lgopen_prepare_floating(data
, ctx
);
2330 void pnfs_parse_lgopen(struct inode
*ino
, struct nfs4_layoutget
*lgp
,
2331 struct nfs_open_context
*ctx
)
2333 struct pnfs_layout_hdr
*lo
;
2334 struct pnfs_layout_segment
*lseg
;
2335 struct nfs_server
*srv
= NFS_SERVER(ino
);
2340 dprintk("%s: entered with status %i\n", __func__
, lgp
->res
.status
);
2341 if (lgp
->res
.status
) {
2342 switch (lgp
->res
.status
) {
2346 * Halt lgopen attempts if the server doesn't recognise
2347 * the "current stateid" value, the layout type, or the
2348 * layoutget operation as being valid.
2349 * Also if it complains about too many ops in the compound
2350 * or of the request/reply being too big.
2352 case -NFS4ERR_BAD_STATEID
:
2353 case -NFS4ERR_NOTSUPP
:
2354 case -NFS4ERR_REP_TOO_BIG
:
2355 case -NFS4ERR_REP_TOO_BIG_TO_CACHE
:
2356 case -NFS4ERR_REQ_TOO_BIG
:
2357 case -NFS4ERR_TOO_MANY_OPS
:
2358 case -NFS4ERR_UNKNOWN_LAYOUTTYPE
:
2359 srv
->caps
&= ~NFS_CAP_LGOPEN
;
2364 lo
= _pnfs_grab_empty_layout(ino
, ctx
);
2371 lseg
= pnfs_layout_process(lgp
);
2372 if (!IS_ERR(lseg
)) {
2373 iomode
= lgp
->args
.range
.iomode
;
2374 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2375 pnfs_put_lseg(lseg
);
2379 void nfs4_lgopen_release(struct nfs4_layoutget
*lgp
)
2383 pnfs_clear_first_layoutget(lgp
->lo
);
2384 nfs_layoutget_end(lgp
->lo
);
2386 pnfs_layoutget_free(lgp
);
2390 struct pnfs_layout_segment
*
2391 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
2393 struct pnfs_layout_hdr
*lo
= lgp
->lo
;
2394 struct nfs4_layoutget_res
*res
= &lgp
->res
;
2395 struct pnfs_layout_segment
*lseg
;
2396 struct inode
*ino
= lo
->plh_inode
;
2399 if (!pnfs_sanity_check_layout_range(&res
->range
))
2400 return ERR_PTR(-EINVAL
);
2402 /* Inject layout blob into I/O device driver */
2403 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
2404 if (IS_ERR_OR_NULL(lseg
)) {
2406 lseg
= ERR_PTR(-ENOMEM
);
2408 dprintk("%s: Could not allocate layout: error %ld\n",
2409 __func__
, PTR_ERR(lseg
));
2413 pnfs_init_lseg(lo
, lseg
, &res
->range
, &res
->stateid
);
2415 spin_lock(&ino
->i_lock
);
2416 if (pnfs_layoutgets_blocked(lo
)) {
2417 dprintk("%s forget reply due to state\n", __func__
);
2421 if (test_bit(NFS_LAYOUT_DRAIN
, &lo
->plh_flags
) &&
2422 !pnfs_is_first_layoutget(lo
))
2425 if (nfs4_stateid_match_other(&lo
->plh_stateid
, &res
->stateid
)) {
2426 /* existing state ID, make sure the sequence number matches. */
2427 if (pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
2428 if (!pnfs_layout_is_valid(lo
))
2429 lo
->plh_barrier
= 0;
2430 dprintk("%s forget reply due to sequence\n", __func__
);
2433 pnfs_set_layout_stateid(lo
, &res
->stateid
, lgp
->cred
, false);
2434 } else if (pnfs_layout_is_valid(lo
)) {
2436 * We got an entirely new state ID. Mark all segments for the
2437 * inode invalid, and retry the layoutget
2439 struct pnfs_layout_range range
= {
2440 .iomode
= IOMODE_ANY
,
2441 .length
= NFS4_MAX_UINT64
,
2443 pnfs_mark_matching_lsegs_return(lo
, &free_me
, &range
, 0);
2446 /* We have a completely new layout */
2447 pnfs_set_layout_stateid(lo
, &res
->stateid
, lgp
->cred
, true);
2450 pnfs_get_lseg(lseg
);
2451 pnfs_layout_insert_lseg(lo
, lseg
, &free_me
);
2454 if (res
->return_on_close
)
2455 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
2457 spin_unlock(&ino
->i_lock
);
2458 pnfs_free_lseg_list(&free_me
);
2462 spin_unlock(&ino
->i_lock
);
2463 lseg
->pls_layout
= lo
;
2464 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
2465 return ERR_PTR(-EAGAIN
);
2469 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2470 * @lo: pointer to layout header
2471 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2472 * @return_range: describe layout segment ranges to be returned
2473 * @seq: stateid seqid to match
2475 * This function is mainly intended for use by layoutrecall. It attempts
2476 * to free the layout segment immediately, or else to mark it for return
2477 * as soon as its reference count drops to zero.
2480 * - 0: a layoutreturn needs to be scheduled.
2481 * - EBUSY: there are layout segment that are still in use.
2482 * - ENOENT: there are no layout segments that need to be returned.
2485 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr
*lo
,
2486 struct list_head
*tmp_list
,
2487 const struct pnfs_layout_range
*return_range
,
2490 struct pnfs_layout_segment
*lseg
, *next
;
2491 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
2494 dprintk("%s:Begin lo %p\n", __func__
, lo
);
2496 assert_spin_locked(&lo
->plh_inode
->i_lock
);
2498 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
2499 tmp_list
= &lo
->plh_return_segs
;
2501 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
2502 if (pnfs_match_lseg_recall(lseg
, return_range
, seq
)) {
2503 dprintk("%s: marking lseg %p iomode %d "
2504 "offset %llu length %llu\n", __func__
,
2505 lseg
, lseg
->pls_range
.iomode
,
2506 lseg
->pls_range
.offset
,
2507 lseg
->pls_range
.length
);
2508 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
2509 tmp_list
= &lo
->plh_return_segs
;
2510 if (mark_lseg_invalid(lseg
, tmp_list
))
2513 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
2514 pnfs_lseg_cancel_io(server
, lseg
);
2518 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
2522 if (!list_empty(&lo
->plh_return_segs
)) {
2523 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
2531 pnfs_mark_layout_for_return(struct inode
*inode
,
2532 const struct pnfs_layout_range
*range
)
2534 struct pnfs_layout_hdr
*lo
;
2535 bool return_now
= false;
2537 spin_lock(&inode
->i_lock
);
2538 lo
= NFS_I(inode
)->layout
;
2539 if (!pnfs_layout_is_valid(lo
)) {
2540 spin_unlock(&inode
->i_lock
);
2543 pnfs_set_plh_return_info(lo
, range
->iomode
, 0);
2545 * mark all matching lsegs so that we are sure to have no live
2546 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2549 if (pnfs_mark_matching_lsegs_return(lo
, &lo
->plh_return_segs
, range
, 0) != -EBUSY
) {
2550 const struct cred
*cred
;
2551 nfs4_stateid stateid
;
2552 enum pnfs_iomode iomode
;
2554 return_now
= pnfs_prepare_layoutreturn(lo
, &stateid
, &cred
, &iomode
);
2555 spin_unlock(&inode
->i_lock
);
2557 pnfs_send_layoutreturn(lo
, &stateid
, &cred
, iomode
, false);
2559 spin_unlock(&inode
->i_lock
);
2560 nfs_commit_inode(inode
, 0);
2564 void pnfs_error_mark_layout_for_return(struct inode
*inode
,
2565 struct pnfs_layout_segment
*lseg
)
2567 struct pnfs_layout_range range
= {
2568 .iomode
= lseg
->pls_range
.iomode
,
2570 .length
= NFS4_MAX_UINT64
,
2573 pnfs_mark_layout_for_return(inode
, &range
);
2575 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return
);
2578 pnfs_layout_can_be_returned(struct pnfs_layout_hdr
*lo
)
2580 return pnfs_layout_is_valid(lo
) &&
2581 !test_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
) &&
2582 !test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
2585 static struct pnfs_layout_segment
*
2586 pnfs_find_first_lseg(struct pnfs_layout_hdr
*lo
,
2587 const struct pnfs_layout_range
*range
,
2588 enum pnfs_iomode iomode
)
2590 struct pnfs_layout_segment
*lseg
;
2592 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
2593 if (!test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
2595 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
2597 if (lseg
->pls_range
.iomode
!= iomode
&& iomode
!= IOMODE_ANY
)
2599 if (pnfs_lseg_range_intersecting(&lseg
->pls_range
, range
))
2605 /* Find open file states whose mode matches that of the range */
2607 pnfs_should_return_unused_layout(struct pnfs_layout_hdr
*lo
,
2608 const struct pnfs_layout_range
*range
)
2610 struct list_head
*head
;
2611 struct nfs_open_context
*ctx
;
2614 if (!pnfs_layout_can_be_returned(lo
) ||
2615 !pnfs_find_first_lseg(lo
, range
, range
->iomode
))
2618 head
= &NFS_I(lo
->plh_inode
)->open_files
;
2619 list_for_each_entry_rcu(ctx
, head
, list
) {
2621 mode
|= ctx
->state
->state
& (FMODE_READ
|FMODE_WRITE
);
2624 switch (range
->iomode
) {
2628 mode
&= ~FMODE_WRITE
;
2631 if (pnfs_find_first_lseg(lo
, range
, IOMODE_READ
))
2632 mode
&= ~FMODE_READ
;
2637 static int pnfs_layout_return_unused_byserver(struct nfs_server
*server
,
2640 const struct pnfs_layout_range
*range
= data
;
2641 const struct cred
*cred
;
2642 struct pnfs_layout_hdr
*lo
;
2643 struct inode
*inode
;
2644 nfs4_stateid stateid
;
2645 enum pnfs_iomode iomode
;
2649 list_for_each_entry_rcu(lo
, &server
->layouts
, plh_layouts
) {
2650 inode
= lo
->plh_inode
;
2651 if (!inode
|| !pnfs_layout_can_be_returned(lo
) ||
2652 test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
2654 spin_lock(&inode
->i_lock
);
2655 if (!lo
->plh_inode
||
2656 !pnfs_should_return_unused_layout(lo
, range
)) {
2657 spin_unlock(&inode
->i_lock
);
2660 pnfs_get_layout_hdr(lo
);
2661 pnfs_set_plh_return_info(lo
, range
->iomode
, 0);
2662 if (pnfs_mark_matching_lsegs_return(lo
, &lo
->plh_return_segs
,
2664 !pnfs_prepare_layoutreturn(lo
, &stateid
, &cred
, &iomode
)) {
2665 spin_unlock(&inode
->i_lock
);
2667 pnfs_put_layout_hdr(lo
);
2671 spin_unlock(&inode
->i_lock
);
2673 pnfs_send_layoutreturn(lo
, &stateid
, &cred
, iomode
, false);
2674 pnfs_put_layout_hdr(lo
);
2683 pnfs_layout_return_unused_byclid(struct nfs_client
*clp
,
2684 enum pnfs_iomode iomode
)
2686 struct pnfs_layout_range range
= {
2689 .length
= NFS4_MAX_UINT64
,
2692 nfs_client_for_each_server(clp
, pnfs_layout_return_unused_byserver
,
2697 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor
*pgio
)
2699 if (pgio
->pg_lseg
== NULL
||
2700 test_bit(NFS_LSEG_VALID
, &pgio
->pg_lseg
->pls_flags
))
2702 pnfs_put_lseg(pgio
->pg_lseg
);
2703 pgio
->pg_lseg
= NULL
;
2705 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout
);
2708 * Check for any intersection between the request and the pgio->pg_lseg,
2709 * and if none, put this pgio->pg_lseg away.
2712 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2714 if (pgio
->pg_lseg
&& !pnfs_lseg_request_intersecting(pgio
->pg_lseg
, req
)) {
2715 pnfs_put_lseg(pgio
->pg_lseg
);
2716 pgio
->pg_lseg
= NULL
;
2719 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range
);
2722 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2726 pnfs_generic_pg_check_layout(pgio
);
2727 pnfs_generic_pg_check_range(pgio
, req
);
2728 if (pgio
->pg_lseg
== NULL
) {
2729 if (pgio
->pg_dreq
== NULL
)
2730 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
2732 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
2735 pnfs_update_layout(pgio
->pg_inode
, nfs_req_openctx(req
),
2736 req_offset(req
), rd_size
,
2739 if (IS_ERR(pgio
->pg_lseg
)) {
2740 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2741 pgio
->pg_lseg
= NULL
;
2745 /* If no lseg, fall back to read through mds */
2746 if (pgio
->pg_lseg
== NULL
)
2747 nfs_pageio_reset_read_mds(pgio
);
2750 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
2753 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
2754 struct nfs_page
*req
, u64 wb_size
)
2756 pnfs_generic_pg_check_layout(pgio
);
2757 pnfs_generic_pg_check_range(pgio
, req
);
2758 if (pgio
->pg_lseg
== NULL
) {
2760 pnfs_update_layout(pgio
->pg_inode
, nfs_req_openctx(req
),
2761 req_offset(req
), wb_size
, IOMODE_RW
,
2762 false, nfs_io_gfp_mask());
2763 if (IS_ERR(pgio
->pg_lseg
)) {
2764 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2765 pgio
->pg_lseg
= NULL
;
2769 /* If no lseg, fall back to write through mds */
2770 if (pgio
->pg_lseg
== NULL
)
2771 nfs_pageio_reset_write_mds(pgio
);
2773 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
2776 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor
*desc
)
2778 if (desc
->pg_lseg
) {
2779 pnfs_put_lseg(desc
->pg_lseg
);
2780 desc
->pg_lseg
= NULL
;
2783 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup
);
2786 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2787 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2790 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
,
2791 struct nfs_page
*prev
, struct nfs_page
*req
)
2794 u64 seg_end
, req_start
, seg_left
;
2796 size
= nfs_generic_pg_test(pgio
, prev
, req
);
2801 * 'size' contains the number of bytes left in the current page (up
2802 * to the original size asked for in @req->wb_bytes).
2804 * Calculate how many bytes are left in the layout segment
2805 * and if there are less bytes than 'size', return that instead.
2807 * Please also note that 'end_offset' is actually the offset of the
2808 * first byte that lies outside the pnfs_layout_range. FIXME?
2811 if (pgio
->pg_lseg
) {
2812 seg_end
= pnfs_end_offset(pgio
->pg_lseg
->pls_range
.offset
,
2813 pgio
->pg_lseg
->pls_range
.length
);
2814 req_start
= req_offset(req
);
2816 /* start of request is past the last byte of this segment */
2817 if (req_start
>= seg_end
)
2820 /* adjust 'size' iff there are fewer bytes left in the
2821 * segment than what nfs_generic_pg_test returned */
2822 seg_left
= seg_end
- req_start
;
2823 if (seg_left
< size
)
2824 size
= (unsigned int)seg_left
;
2829 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
2831 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2833 struct nfs_pageio_descriptor pgio
;
2835 /* Resend all requests through the MDS */
2836 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
2837 hdr
->completion_ops
);
2838 return nfs_pageio_resend(&pgio
, hdr
);
2840 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
2842 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
2845 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
2846 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2847 PNFS_LAYOUTRET_ON_ERROR
) {
2848 pnfs_return_layout(hdr
->inode
);
2850 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2851 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
2855 * Called by non rpc-based layout drivers
2857 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
2859 if (likely(!hdr
->pnfs_error
)) {
2860 pnfs_set_layoutcommit(hdr
->inode
, hdr
->lseg
,
2861 hdr
->mds_offset
+ hdr
->res
.count
);
2862 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2864 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
2865 if (unlikely(hdr
->pnfs_error
))
2866 pnfs_ld_handle_write_error(hdr
);
2867 hdr
->mds_ops
->rpc_release(hdr
);
2869 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
2872 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
2873 struct nfs_pgio_header
*hdr
)
2875 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2877 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2878 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2879 nfs_pageio_reset_write_mds(desc
);
2880 mirror
->pg_recoalesce
= 1;
2882 hdr
->completion_ops
->completion(hdr
);
2885 static enum pnfs_try_status
2886 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
2887 const struct rpc_call_ops
*call_ops
,
2888 struct pnfs_layout_segment
*lseg
,
2891 struct inode
*inode
= hdr
->inode
;
2892 enum pnfs_try_status trypnfs
;
2893 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2895 hdr
->mds_ops
= call_ops
;
2897 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
2898 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
2899 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
2900 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2901 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
2902 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2907 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
2908 struct nfs_pgio_header
*hdr
, int how
)
2910 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2911 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2912 enum pnfs_try_status trypnfs
;
2914 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
2916 case PNFS_NOT_ATTEMPTED
:
2917 pnfs_write_through_mds(desc
, hdr
);
2919 case PNFS_ATTEMPTED
:
2921 case PNFS_TRY_AGAIN
:
2922 /* cleanup hdr and prepare to redo pnfs */
2923 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2924 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2925 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
2926 mirror
->pg_recoalesce
= 1;
2928 hdr
->mds_ops
->rpc_release(hdr
);
2932 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
2934 pnfs_put_lseg(hdr
->lseg
);
2935 nfs_pgio_header_free(hdr
);
2939 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
2941 struct nfs_pgio_header
*hdr
;
2944 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2946 desc
->pg_error
= -ENOMEM
;
2947 return desc
->pg_error
;
2949 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
2951 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2952 ret
= nfs_generic_pgio(desc
, hdr
);
2954 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
2958 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
2960 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2962 struct nfs_pageio_descriptor pgio
;
2964 /* Resend all requests through the MDS */
2965 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
2966 return nfs_pageio_resend(&pgio
, hdr
);
2968 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
2970 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
2972 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
2973 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2974 PNFS_LAYOUTRET_ON_ERROR
) {
2975 pnfs_return_layout(hdr
->inode
);
2977 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2978 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
2982 * Called by non rpc-based layout drivers
2984 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
2986 if (likely(!hdr
->pnfs_error
))
2987 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2988 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
2989 if (unlikely(hdr
->pnfs_error
))
2990 pnfs_ld_handle_read_error(hdr
);
2991 hdr
->mds_ops
->rpc_release(hdr
);
2993 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
2996 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
2997 struct nfs_pgio_header
*hdr
)
2999 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
3001 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
3002 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
3003 nfs_pageio_reset_read_mds(desc
);
3004 mirror
->pg_recoalesce
= 1;
3006 hdr
->completion_ops
->completion(hdr
);
3010 * Call the appropriate parallel I/O subsystem read function.
3012 static enum pnfs_try_status
3013 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
3014 const struct rpc_call_ops
*call_ops
,
3015 struct pnfs_layout_segment
*lseg
)
3017 struct inode
*inode
= hdr
->inode
;
3018 struct nfs_server
*nfss
= NFS_SERVER(inode
);
3019 enum pnfs_try_status trypnfs
;
3021 hdr
->mds_ops
= call_ops
;
3023 dprintk("%s: Reading ino:%lu %u@%llu\n",
3024 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
3026 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
3027 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
3028 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
3029 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
3033 /* Resend all requests through pnfs. */
3034 void pnfs_read_resend_pnfs(struct nfs_pgio_header
*hdr
,
3035 unsigned int mirror_idx
)
3037 struct nfs_pageio_descriptor pgio
;
3039 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
3040 /* Prevent deadlocks with layoutreturn! */
3041 pnfs_put_lseg(hdr
->lseg
);
3044 nfs_pageio_init_read(&pgio
, hdr
->inode
, false,
3045 hdr
->completion_ops
);
3046 pgio
.pg_mirror_idx
= mirror_idx
;
3047 hdr
->task
.tk_status
= nfs_pageio_resend(&pgio
, hdr
);
3050 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs
);
3053 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
3055 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
3056 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
3057 enum pnfs_try_status trypnfs
;
3059 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
3061 case PNFS_NOT_ATTEMPTED
:
3062 pnfs_read_through_mds(desc
, hdr
);
3064 case PNFS_ATTEMPTED
:
3066 case PNFS_TRY_AGAIN
:
3067 /* cleanup hdr and prepare to redo pnfs */
3068 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
3069 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
3070 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
3071 mirror
->pg_recoalesce
= 1;
3073 hdr
->mds_ops
->rpc_release(hdr
);
3077 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
3079 pnfs_put_lseg(hdr
->lseg
);
3080 nfs_pgio_header_free(hdr
);
3084 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
3086 struct nfs_pgio_header
*hdr
;
3089 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
3091 desc
->pg_error
= -ENOMEM
;
3092 return desc
->pg_error
;
3094 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
3095 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
3096 ret
= nfs_generic_pgio(desc
, hdr
);
3098 pnfs_do_read(desc
, hdr
);
3101 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
3103 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
3105 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
3107 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
3108 smp_mb__after_atomic();
3109 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
3113 * There can be multiple RW segments.
3115 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
3117 struct pnfs_layout_segment
*lseg
;
3119 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
3120 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
3121 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
3122 list_add(&lseg
->pls_lc_list
, listp
);
3126 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
3128 struct pnfs_layout_segment
*lseg
, *tmp
;
3130 /* Matched by references in pnfs_set_layoutcommit */
3131 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
3132 list_del_init(&lseg
->pls_lc_list
);
3133 pnfs_put_lseg(lseg
);
3136 pnfs_clear_layoutcommitting(inode
);
3139 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
3141 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
3143 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
3146 pnfs_set_layoutcommit(struct inode
*inode
, struct pnfs_layout_segment
*lseg
,
3149 struct nfs_inode
*nfsi
= NFS_I(inode
);
3150 bool mark_as_dirty
= false;
3152 spin_lock(&inode
->i_lock
);
3153 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
3154 nfsi
->layout
->plh_lwb
= end_pos
;
3155 mark_as_dirty
= true;
3156 dprintk("%s: Set layoutcommit for inode %lu ",
3157 __func__
, inode
->i_ino
);
3158 } else if (end_pos
> nfsi
->layout
->plh_lwb
)
3159 nfsi
->layout
->plh_lwb
= end_pos
;
3160 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
)) {
3161 /* references matched in nfs4_layoutcommit_release */
3162 pnfs_get_lseg(lseg
);
3164 spin_unlock(&inode
->i_lock
);
3165 dprintk("%s: lseg %p end_pos %llu\n",
3166 __func__
, lseg
, nfsi
->layout
->plh_lwb
);
3168 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3169 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3171 mark_inode_dirty_sync(inode
);
3173 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
3175 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
3177 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
3179 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
3180 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
3181 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
3185 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3186 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3187 * data to disk to allow the server to recover the data if it crashes.
3188 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3189 * is off, and a COMMIT is sent to a data server, or
3190 * if WRITEs to a data server return NFS_DATA_SYNC.
3193 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
3195 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
3196 struct nfs4_layoutcommit_data
*data
;
3197 struct nfs_inode
*nfsi
= NFS_I(inode
);
3201 if (!pnfs_layoutcommit_outstanding(inode
))
3204 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
3207 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
3210 status
= wait_on_bit_lock_action(&nfsi
->flags
,
3211 NFS_INO_LAYOUTCOMMITTING
,
3212 nfs_wait_bit_killable
,
3213 TASK_KILLABLE
|TASK_FREEZABLE_UNSAFE
);
3219 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3220 data
= kzalloc(sizeof(*data
), nfs_io_gfp_mask());
3222 goto clear_layoutcommitting
;
3225 spin_lock(&inode
->i_lock
);
3226 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
3229 INIT_LIST_HEAD(&data
->lseg_list
);
3230 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
3232 end_pos
= nfsi
->layout
->plh_lwb
;
3234 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
3235 data
->cred
= get_cred(nfsi
->layout
->plh_lc_cred
);
3236 spin_unlock(&inode
->i_lock
);
3238 data
->args
.inode
= inode
;
3239 nfs_fattr_init(&data
->fattr
);
3240 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
3241 data
->res
.fattr
= &data
->fattr
;
3243 data
->args
.lastbytewritten
= end_pos
- 1;
3245 data
->args
.lastbytewritten
= U64_MAX
;
3246 data
->res
.server
= NFS_SERVER(inode
);
3248 if (ld
->prepare_layoutcommit
) {
3249 status
= ld
->prepare_layoutcommit(&data
->args
);
3251 put_cred(data
->cred
);
3252 spin_lock(&inode
->i_lock
);
3253 set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
);
3254 if (end_pos
> nfsi
->layout
->plh_lwb
)
3255 nfsi
->layout
->plh_lwb
= end_pos
;
3261 status
= nfs4_proc_layoutcommit(data
, sync
);
3264 mark_inode_dirty_sync(inode
);
3265 dprintk("<-- %s status %d\n", __func__
, status
);
3268 spin_unlock(&inode
->i_lock
);
3270 clear_layoutcommitting
:
3271 pnfs_clear_layoutcommitting(inode
);
3274 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode
);
3277 pnfs_generic_sync(struct inode
*inode
, bool datasync
)
3279 return pnfs_layoutcommit_inode(inode
, true);
3281 EXPORT_SYMBOL_GPL(pnfs_generic_sync
);
3283 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
3285 struct nfs4_threshold
*thp
;
3287 thp
= kzalloc(sizeof(*thp
), nfs_io_gfp_mask());
3289 dprintk("%s mdsthreshold allocation failed\n", __func__
);
3295 #if IS_ENABLED(CONFIG_NFS_V4_2)
3297 pnfs_report_layoutstat(struct inode
*inode
, gfp_t gfp_flags
)
3299 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
3300 struct nfs_server
*server
= NFS_SERVER(inode
);
3301 struct nfs_inode
*nfsi
= NFS_I(inode
);
3302 struct nfs42_layoutstat_data
*data
;
3303 struct pnfs_layout_hdr
*hdr
;
3306 if (!pnfs_enabled_sb(server
) || !ld
->prepare_layoutstats
)
3309 if (!nfs_server_capable(inode
, NFS_CAP_LAYOUTSTATS
))
3312 if (test_and_set_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
))
3315 spin_lock(&inode
->i_lock
);
3316 if (!NFS_I(inode
)->layout
) {
3317 spin_unlock(&inode
->i_lock
);
3318 goto out_clear_layoutstats
;
3320 hdr
= NFS_I(inode
)->layout
;
3321 pnfs_get_layout_hdr(hdr
);
3322 spin_unlock(&inode
->i_lock
);
3324 data
= kzalloc(sizeof(*data
), gfp_flags
);
3330 data
->args
.fh
= NFS_FH(inode
);
3331 data
->args
.inode
= inode
;
3332 status
= ld
->prepare_layoutstats(&data
->args
);
3336 status
= nfs42_proc_layoutstats_generic(NFS_SERVER(inode
), data
);
3339 dprintk("%s returns %d\n", __func__
, status
);
3345 pnfs_put_layout_hdr(hdr
);
3346 out_clear_layoutstats
:
3347 smp_mb__before_atomic();
3348 clear_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
);
3349 smp_mb__after_atomic();
3352 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat
);
3355 unsigned int layoutstats_timer
;
3356 module_param(layoutstats_timer
, uint
, 0644);
3357 EXPORT_SYMBOL_GPL(layoutstats_timer
);