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"
41 #define NFSDBG_FACILITY NFSDBG_PNFS
42 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
47 * protects pnfs_modules_tbl.
49 static DEFINE_SPINLOCK(pnfs_spinlock
);
52 * pnfs_modules_tbl holds all pnfs modules
54 static LIST_HEAD(pnfs_modules_tbl
);
56 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
);
57 static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
58 struct list_head
*free_me
,
59 const struct pnfs_layout_range
*range
,
61 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
62 struct list_head
*tmp_list
);
64 /* Return the registered pnfs layout driver module matching given id */
65 static struct pnfs_layoutdriver_type
*
66 find_pnfs_driver_locked(u32 id
)
68 struct pnfs_layoutdriver_type
*local
;
70 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
75 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
79 static struct pnfs_layoutdriver_type
*
80 find_pnfs_driver(u32 id
)
82 struct pnfs_layoutdriver_type
*local
;
84 spin_lock(&pnfs_spinlock
);
85 local
= find_pnfs_driver_locked(id
);
86 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
87 dprintk("%s: Could not grab reference on module\n", __func__
);
90 spin_unlock(&pnfs_spinlock
);
95 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
97 if (nfss
->pnfs_curr_ld
) {
98 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
99 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
100 /* Decrement the MDS count. Purge the deviceid cache if zero */
101 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
102 nfs4_deviceid_purge_client(nfss
->nfs_client
);
103 module_put(nfss
->pnfs_curr_ld
->owner
);
105 nfss
->pnfs_curr_ld
= NULL
;
109 * When the server sends a list of layout types, we choose one in the order
110 * given in the list below.
112 * FIXME: should this list be configurable in some fashion? module param?
113 * mount option? something else?
115 static const u32 ld_prefs
[] = {
120 LAYOUT_NFSV4_1_FILES
,
125 ld_cmp(const void *e1
, const void *e2
)
127 u32 ld1
= *((u32
*)e1
);
128 u32 ld2
= *((u32
*)e2
);
131 for (i
= 0; ld_prefs
[i
] != 0; i
++) {
132 if (ld1
== ld_prefs
[i
])
135 if (ld2
== ld_prefs
[i
])
142 * Try to set the server's pnfs module to the pnfs layout type specified by id.
143 * Currently only one pNFS layout driver per filesystem is supported.
145 * @ids array of layout types supported by MDS.
148 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
149 struct nfs_fsinfo
*fsinfo
)
151 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
155 if (fsinfo
->nlayouttypes
== 0)
157 if (!(server
->nfs_client
->cl_exchange_flags
&
158 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
159 printk(KERN_ERR
"NFS: %s: cl_exchange_flags 0x%x\n",
160 __func__
, server
->nfs_client
->cl_exchange_flags
);
164 sort(fsinfo
->layouttype
, fsinfo
->nlayouttypes
,
165 sizeof(*fsinfo
->layouttype
), ld_cmp
, NULL
);
167 for (i
= 0; i
< fsinfo
->nlayouttypes
; i
++) {
168 id
= fsinfo
->layouttype
[i
];
169 ld_type
= find_pnfs_driver(id
);
171 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
,
173 ld_type
= find_pnfs_driver(id
);
180 dprintk("%s: No pNFS module found!\n", __func__
);
184 server
->pnfs_curr_ld
= ld_type
;
185 if (ld_type
->set_layoutdriver
186 && ld_type
->set_layoutdriver(server
, mntfh
)) {
187 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
188 "driver %u.\n", __func__
, id
);
189 module_put(ld_type
->owner
);
192 /* Bump the MDS count */
193 atomic_inc(&server
->nfs_client
->cl_mds_count
);
195 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
199 dprintk("%s: Using NFSv4 I/O\n", __func__
);
200 server
->pnfs_curr_ld
= NULL
;
204 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
206 int status
= -EINVAL
;
207 struct pnfs_layoutdriver_type
*tmp
;
209 if (ld_type
->id
== 0) {
210 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
213 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
214 printk(KERN_ERR
"NFS: %s Layout driver must provide "
215 "alloc_lseg and free_lseg.\n", __func__
);
219 spin_lock(&pnfs_spinlock
);
220 tmp
= find_pnfs_driver_locked(ld_type
->id
);
222 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
224 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
227 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
228 __func__
, ld_type
->id
);
230 spin_unlock(&pnfs_spinlock
);
234 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
237 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
239 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
240 spin_lock(&pnfs_spinlock
);
241 list_del(&ld_type
->pnfs_tblid
);
242 spin_unlock(&pnfs_spinlock
);
244 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
247 * pNFS client layout cache
250 /* Need to hold i_lock if caller does not already hold reference */
252 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
254 atomic_inc(&lo
->plh_refcount
);
257 static struct pnfs_layout_hdr
*
258 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
260 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
261 return ld
->alloc_layout_hdr(ino
, gfp_flags
);
265 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
267 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
268 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
270 if (!list_empty(&lo
->plh_layouts
)) {
271 struct nfs_client
*clp
= server
->nfs_client
;
273 spin_lock(&clp
->cl_lock
);
274 list_del_init(&lo
->plh_layouts
);
275 spin_unlock(&clp
->cl_lock
);
277 put_rpccred(lo
->plh_lc_cred
);
278 return ld
->free_layout_hdr(lo
);
282 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
284 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
285 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
287 /* Reset MDS Threshold I/O counters */
293 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
299 inode
= lo
->plh_inode
;
300 pnfs_layoutreturn_before_put_layout_hdr(lo
);
302 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
303 if (!list_empty(&lo
->plh_segs
))
304 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
305 pnfs_detach_layout_hdr(lo
);
306 spin_unlock(&inode
->i_lock
);
307 pnfs_free_layout_hdr(lo
);
312 pnfs_set_plh_return_info(struct pnfs_layout_hdr
*lo
, enum pnfs_iomode iomode
,
315 if (lo
->plh_return_iomode
!= 0 && lo
->plh_return_iomode
!= iomode
)
317 lo
->plh_return_iomode
= iomode
;
318 set_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
320 WARN_ON_ONCE(lo
->plh_return_seq
!= 0 && lo
->plh_return_seq
!= seq
);
321 lo
->plh_return_seq
= seq
;
326 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr
*lo
)
328 struct pnfs_layout_segment
*lseg
;
329 lo
->plh_return_iomode
= 0;
330 lo
->plh_return_seq
= 0;
331 clear_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
332 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
333 if (!test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
335 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
339 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr
*lo
)
341 clear_bit_unlock(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
342 clear_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
);
343 smp_mb__after_atomic();
344 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
);
345 rpc_wake_up(&NFS_SERVER(lo
->plh_inode
)->roc_rpcwaitq
);
349 pnfs_clear_lseg_state(struct pnfs_layout_segment
*lseg
,
350 struct list_head
*free_me
)
352 clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
353 clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
354 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
355 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
356 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
357 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
361 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
363 * In order to continue using the pnfs_layout_hdr, a full recovery
365 * Note that caller must hold inode->i_lock.
368 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr
*lo
,
369 struct list_head
*lseg_list
)
371 struct pnfs_layout_range range
= {
372 .iomode
= IOMODE_ANY
,
374 .length
= NFS4_MAX_UINT64
,
376 struct pnfs_layout_segment
*lseg
, *next
;
378 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
379 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
380 pnfs_clear_lseg_state(lseg
, lseg_list
);
381 pnfs_clear_layoutreturn_info(lo
);
382 pnfs_free_returned_lsegs(lo
, lseg_list
, &range
, 0);
383 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
) &&
384 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
385 pnfs_clear_layoutreturn_waitbit(lo
);
386 return !list_empty(&lo
->plh_segs
);
390 pnfs_iomode_to_fail_bit(u32 iomode
)
392 return iomode
== IOMODE_RW
?
393 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
397 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
399 lo
->plh_retry_timestamp
= jiffies
;
400 if (!test_and_set_bit(fail_bit
, &lo
->plh_flags
))
401 atomic_inc(&lo
->plh_refcount
);
405 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
407 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
408 atomic_dec(&lo
->plh_refcount
);
412 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
414 struct inode
*inode
= lo
->plh_inode
;
415 struct pnfs_layout_range range
= {
418 .length
= NFS4_MAX_UINT64
,
422 spin_lock(&inode
->i_lock
);
423 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
424 pnfs_mark_matching_lsegs_invalid(lo
, &head
, &range
, 0);
425 spin_unlock(&inode
->i_lock
);
426 pnfs_free_lseg_list(&head
);
427 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
428 iomode
== IOMODE_RW
? "RW" : "READ");
432 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
434 unsigned long start
, end
;
435 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
437 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
440 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
441 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
442 /* It is time to retry the failed layoutgets */
443 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
450 pnfs_init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
,
451 const struct pnfs_layout_range
*range
,
452 const nfs4_stateid
*stateid
)
454 INIT_LIST_HEAD(&lseg
->pls_list
);
455 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
456 atomic_set(&lseg
->pls_refcount
, 1);
457 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
458 lseg
->pls_layout
= lo
;
459 lseg
->pls_range
= *range
;
460 lseg
->pls_seq
= be32_to_cpu(stateid
->seqid
);
463 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
466 struct inode
*inode
= lseg
->pls_layout
->plh_inode
;
467 NFS_SERVER(inode
)->pnfs_curr_ld
->free_lseg(lseg
);
472 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
473 struct pnfs_layout_segment
*lseg
)
475 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
476 list_del_init(&lseg
->pls_list
);
477 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
478 atomic_dec(&lo
->plh_refcount
);
479 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
481 if (list_empty(&lo
->plh_segs
) &&
482 !test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
483 !test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
484 if (atomic_read(&lo
->plh_outstanding
) == 0)
485 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
486 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
491 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr
*lo
,
492 struct pnfs_layout_segment
*lseg
)
494 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
495 pnfs_layout_is_valid(lo
)) {
496 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
497 list_move_tail(&lseg
->pls_list
, &lo
->plh_return_segs
);
504 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
506 struct pnfs_layout_hdr
*lo
;
512 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
513 atomic_read(&lseg
->pls_refcount
),
514 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
516 lo
= lseg
->pls_layout
;
517 inode
= lo
->plh_inode
;
519 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
520 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
521 spin_unlock(&inode
->i_lock
);
524 pnfs_get_layout_hdr(lo
);
525 pnfs_layout_remove_lseg(lo
, lseg
);
526 if (pnfs_cache_lseg_for_layoutreturn(lo
, lseg
))
528 spin_unlock(&inode
->i_lock
);
529 pnfs_free_lseg(lseg
);
530 pnfs_put_layout_hdr(lo
);
533 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
536 * is l2 fully contained in l1?
538 * [----------------------------------)
543 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
544 const struct pnfs_layout_range
*l2
)
546 u64 start1
= l1
->offset
;
547 u64 end1
= pnfs_end_offset(start1
, l1
->length
);
548 u64 start2
= l2
->offset
;
549 u64 end2
= pnfs_end_offset(start2
, l2
->length
);
551 return (start1
<= start2
) && (end1
>= end2
);
554 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
555 struct list_head
*tmp_list
)
557 if (!atomic_dec_and_test(&lseg
->pls_refcount
))
559 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
560 list_add(&lseg
->pls_list
, tmp_list
);
564 /* Returns 1 if lseg is removed from list, 0 otherwise */
565 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
566 struct list_head
*tmp_list
)
570 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
571 /* Remove the reference keeping the lseg in the
572 * list. It will now be removed when all
573 * outstanding io is finished.
575 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
576 atomic_read(&lseg
->pls_refcount
));
577 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
584 * Compare 2 layout stateid sequence ids, to see which is newer,
585 * taking into account wraparound issues.
587 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
589 return (s32
)(s1
- s2
) > 0;
593 pnfs_should_free_range(const struct pnfs_layout_range
*lseg_range
,
594 const struct pnfs_layout_range
*recall_range
)
596 return (recall_range
->iomode
== IOMODE_ANY
||
597 lseg_range
->iomode
== recall_range
->iomode
) &&
598 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
602 pnfs_match_lseg_recall(const struct pnfs_layout_segment
*lseg
,
603 const struct pnfs_layout_range
*recall_range
,
606 if (seq
!= 0 && pnfs_seqid_is_newer(lseg
->pls_seq
, seq
))
608 if (recall_range
== NULL
)
610 return pnfs_should_free_range(&lseg
->pls_range
, recall_range
);
614 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
615 * @lo: layout header containing the lsegs
616 * @tmp_list: list head where doomed lsegs should go
617 * @recall_range: optional recall range argument to match (may be NULL)
618 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
620 * Walk the list of lsegs in the layout header, and tear down any that should
621 * be destroyed. If "recall_range" is specified then the segment must match
622 * that range. If "seq" is non-zero, then only match segments that were handed
623 * out at or before that sequence.
625 * Returns number of matching invalid lsegs remaining in list after scanning
626 * it and purging them.
629 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
630 struct list_head
*tmp_list
,
631 const struct pnfs_layout_range
*recall_range
,
634 struct pnfs_layout_segment
*lseg
, *next
;
637 dprintk("%s:Begin lo %p\n", __func__
, lo
);
639 if (list_empty(&lo
->plh_segs
))
641 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
642 if (pnfs_match_lseg_recall(lseg
, recall_range
, seq
)) {
643 dprintk("%s: freeing lseg %p iomode %d seq %u"
644 "offset %llu length %llu\n", __func__
,
645 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_seq
,
646 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
647 if (!mark_lseg_invalid(lseg
, tmp_list
))
650 dprintk("%s:Return %i\n", __func__
, remaining
);
655 pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
656 struct list_head
*free_me
,
657 const struct pnfs_layout_range
*range
,
660 struct pnfs_layout_segment
*lseg
, *next
;
662 list_for_each_entry_safe(lseg
, next
, &lo
->plh_return_segs
, pls_list
) {
663 if (pnfs_match_lseg_recall(lseg
, range
, seq
))
664 list_move_tail(&lseg
->pls_list
, free_me
);
668 /* note free_me must contain lsegs from a single layout_hdr */
670 pnfs_free_lseg_list(struct list_head
*free_me
)
672 struct pnfs_layout_segment
*lseg
, *tmp
;
674 if (list_empty(free_me
))
677 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
678 list_del(&lseg
->pls_list
);
679 pnfs_free_lseg(lseg
);
684 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
686 struct pnfs_layout_hdr
*lo
;
689 spin_lock(&nfsi
->vfs_inode
.i_lock
);
692 pnfs_get_layout_hdr(lo
);
693 pnfs_mark_layout_stateid_invalid(lo
, &tmp_list
);
694 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
695 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
696 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
697 pnfs_free_lseg_list(&tmp_list
);
698 nfs_commit_inode(&nfsi
->vfs_inode
, 0);
699 pnfs_put_layout_hdr(lo
);
701 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
703 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
706 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
707 struct list_head
*layout_list
)
709 struct pnfs_layout_hdr
*lo
;
712 spin_lock(&inode
->i_lock
);
713 lo
= NFS_I(inode
)->layout
;
714 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
715 pnfs_get_layout_hdr(lo
);
716 list_add(&lo
->plh_bulk_destroy
, layout_list
);
719 spin_unlock(&inode
->i_lock
);
723 /* Caller must hold rcu_read_lock and clp->cl_lock */
725 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
726 struct nfs_server
*server
,
727 struct list_head
*layout_list
)
728 __must_hold(&clp
->cl_lock
)
731 struct pnfs_layout_hdr
*lo
, *next
;
734 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
735 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
) ||
736 test_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
) ||
737 !list_empty(&lo
->plh_bulk_destroy
))
739 /* If the sb is being destroyed, just bail */
740 if (!nfs_sb_active(server
->super
))
742 inode
= igrab(lo
->plh_inode
);
744 list_del_init(&lo
->plh_layouts
);
745 if (pnfs_layout_add_bulk_destroy_list(inode
,
749 spin_unlock(&clp
->cl_lock
);
753 spin_unlock(&clp
->cl_lock
);
754 set_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
);
756 nfs_sb_deactive(server
->super
);
757 spin_lock(&clp
->cl_lock
);
765 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
768 struct pnfs_layout_hdr
*lo
;
770 LIST_HEAD(lseg_list
);
773 while (!list_empty(layout_list
)) {
774 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
776 dprintk("%s freeing layout for inode %lu\n", __func__
,
777 lo
->plh_inode
->i_ino
);
778 inode
= lo
->plh_inode
;
780 pnfs_layoutcommit_inode(inode
, false);
782 spin_lock(&inode
->i_lock
);
783 list_del_init(&lo
->plh_bulk_destroy
);
784 if (pnfs_mark_layout_stateid_invalid(lo
, &lseg_list
)) {
786 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
789 spin_unlock(&inode
->i_lock
);
790 pnfs_free_lseg_list(&lseg_list
);
791 /* Free all lsegs that are attached to commit buckets */
792 nfs_commit_inode(inode
, 0);
793 pnfs_put_layout_hdr(lo
);
794 nfs_iput_and_deactive(inode
);
800 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
801 struct nfs_fsid
*fsid
,
804 struct nfs_server
*server
;
805 LIST_HEAD(layout_list
);
807 spin_lock(&clp
->cl_lock
);
810 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
811 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
813 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
819 spin_unlock(&clp
->cl_lock
);
821 if (list_empty(&layout_list
))
823 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
827 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
830 struct nfs_server
*server
;
831 LIST_HEAD(layout_list
);
833 spin_lock(&clp
->cl_lock
);
836 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
837 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
843 spin_unlock(&clp
->cl_lock
);
845 if (list_empty(&layout_list
))
847 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
851 * Called by the state manger to remove all layouts established under an
855 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
857 nfs4_deviceid_mark_client_invalid(clp
);
858 nfs4_deviceid_purge_client(clp
);
860 pnfs_destroy_layouts_byclid(clp
, false);
863 /* update lo->plh_stateid with new if is more recent */
865 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
868 u32 oldseq
, newseq
, new_barrier
= 0;
870 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
871 newseq
= be32_to_cpu(new->seqid
);
873 if (!pnfs_layout_is_valid(lo
)) {
874 nfs4_stateid_copy(&lo
->plh_stateid
, new);
875 lo
->plh_barrier
= newseq
;
876 pnfs_clear_layoutreturn_info(lo
);
877 clear_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
880 if (pnfs_seqid_is_newer(newseq
, oldseq
)) {
881 nfs4_stateid_copy(&lo
->plh_stateid
, new);
883 * Because of wraparound, we want to keep the barrier
884 * "close" to the current seqids.
886 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
889 new_barrier
= be32_to_cpu(new->seqid
);
890 else if (new_barrier
== 0)
892 if (pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
893 lo
->plh_barrier
= new_barrier
;
897 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
898 const nfs4_stateid
*stateid
)
900 u32 seqid
= be32_to_cpu(stateid
->seqid
);
902 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
905 /* lget is set to 1 if called from inside send_layoutget call chain */
907 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
)
909 return lo
->plh_block_lgets
||
910 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
914 * Get layout from server.
915 * for now, assume that whole file layouts are requested.
917 * arg->length: all ones
919 static struct pnfs_layout_segment
*
920 send_layoutget(struct pnfs_layout_hdr
*lo
,
921 struct nfs_open_context
*ctx
,
922 nfs4_stateid
*stateid
,
923 const struct pnfs_layout_range
*range
,
924 long *timeout
, gfp_t gfp_flags
)
926 struct inode
*ino
= lo
->plh_inode
;
927 struct nfs_server
*server
= NFS_SERVER(ino
);
928 struct nfs4_layoutget
*lgp
;
931 dprintk("--> %s\n", __func__
);
934 * Synchronously retrieve layout information from server and
935 * store in lseg. If we race with a concurrent seqid morphing
936 * op, then re-send the LAYOUTGET.
938 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
940 return ERR_PTR(-ENOMEM
);
942 i_size
= i_size_read(ino
);
944 lgp
->args
.minlength
= PAGE_SIZE
;
945 if (lgp
->args
.minlength
> range
->length
)
946 lgp
->args
.minlength
= range
->length
;
947 if (range
->iomode
== IOMODE_READ
) {
948 if (range
->offset
>= i_size
)
949 lgp
->args
.minlength
= 0;
950 else if (i_size
- range
->offset
< lgp
->args
.minlength
)
951 lgp
->args
.minlength
= i_size
- range
->offset
;
953 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
954 pnfs_copy_range(&lgp
->args
.range
, range
);
955 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
956 lgp
->args
.inode
= ino
;
957 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
958 nfs4_stateid_copy(&lgp
->args
.stateid
, stateid
);
959 lgp
->gfp_flags
= gfp_flags
;
960 lgp
->cred
= lo
->plh_lc_cred
;
962 return nfs4_proc_layoutget(lgp
, timeout
, gfp_flags
);
965 static void pnfs_clear_layoutcommit(struct inode
*inode
,
966 struct list_head
*head
)
968 struct nfs_inode
*nfsi
= NFS_I(inode
);
969 struct pnfs_layout_segment
*lseg
, *tmp
;
971 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
973 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
974 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
976 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
980 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr
*lo
,
981 const nfs4_stateid
*arg_stateid
,
982 const struct pnfs_layout_range
*range
,
983 const nfs4_stateid
*stateid
)
985 struct inode
*inode
= lo
->plh_inode
;
988 spin_lock(&inode
->i_lock
);
989 if (!pnfs_layout_is_valid(lo
) || !arg_stateid
||
990 !nfs4_stateid_match_other(&lo
->plh_stateid
, arg_stateid
))
993 u32 seq
= be32_to_cpu(arg_stateid
->seqid
);
995 pnfs_mark_matching_lsegs_invalid(lo
, &freeme
, range
, seq
);
996 pnfs_free_returned_lsegs(lo
, &freeme
, range
, seq
);
997 pnfs_set_layout_stateid(lo
, stateid
, true);
999 pnfs_mark_layout_stateid_invalid(lo
, &freeme
);
1001 pnfs_clear_layoutreturn_waitbit(lo
);
1002 spin_unlock(&inode
->i_lock
);
1003 pnfs_free_lseg_list(&freeme
);
1008 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr
*lo
,
1009 nfs4_stateid
*stateid
,
1010 enum pnfs_iomode
*iomode
)
1012 /* Serialise LAYOUTGET/LAYOUTRETURN */
1013 if (atomic_read(&lo
->plh_outstanding
) != 0)
1015 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
1017 set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
1018 pnfs_get_layout_hdr(lo
);
1019 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
)) {
1020 if (stateid
!= NULL
) {
1021 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
1022 if (lo
->plh_return_seq
!= 0)
1023 stateid
->seqid
= cpu_to_be32(lo
->plh_return_seq
);
1026 *iomode
= lo
->plh_return_iomode
;
1027 pnfs_clear_layoutreturn_info(lo
);
1030 if (stateid
!= NULL
)
1031 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
1033 *iomode
= IOMODE_ANY
;
1038 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args
*args
,
1039 struct pnfs_layout_hdr
*lo
,
1040 const nfs4_stateid
*stateid
,
1041 enum pnfs_iomode iomode
)
1043 struct inode
*inode
= lo
->plh_inode
;
1045 args
->layout_type
= NFS_SERVER(inode
)->pnfs_curr_ld
->id
;
1046 args
->inode
= inode
;
1047 args
->range
.iomode
= iomode
;
1048 args
->range
.offset
= 0;
1049 args
->range
.length
= NFS4_MAX_UINT64
;
1051 nfs4_stateid_copy(&args
->stateid
, stateid
);
1055 pnfs_send_layoutreturn(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*stateid
,
1056 enum pnfs_iomode iomode
, bool sync
)
1058 struct inode
*ino
= lo
->plh_inode
;
1059 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1060 struct nfs4_layoutreturn
*lrp
;
1063 lrp
= kzalloc(sizeof(*lrp
), GFP_NOFS
);
1064 if (unlikely(lrp
== NULL
)) {
1066 spin_lock(&ino
->i_lock
);
1067 pnfs_clear_layoutreturn_waitbit(lo
);
1068 spin_unlock(&ino
->i_lock
);
1069 pnfs_put_layout_hdr(lo
);
1073 pnfs_init_layoutreturn_args(&lrp
->args
, lo
, stateid
, iomode
);
1074 lrp
->args
.ld_private
= &lrp
->ld_private
;
1075 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
1076 lrp
->cred
= lo
->plh_lc_cred
;
1077 if (ld
->prepare_layoutreturn
)
1078 ld
->prepare_layoutreturn(&lrp
->args
);
1080 status
= nfs4_proc_layoutreturn(lrp
, sync
);
1082 dprintk("<-- %s status: %d\n", __func__
, status
);
1086 /* Return true if layoutreturn is needed */
1088 pnfs_layout_need_return(struct pnfs_layout_hdr
*lo
)
1090 struct pnfs_layout_segment
*s
;
1092 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1095 /* Defer layoutreturn until all lsegs are done */
1096 list_for_each_entry(s
, &lo
->plh_segs
, pls_list
) {
1097 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &s
->pls_flags
))
1104 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
1106 struct inode
*inode
= lo
->plh_inode
;
1108 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1110 spin_lock(&inode
->i_lock
);
1111 if (pnfs_layout_need_return(lo
)) {
1112 nfs4_stateid stateid
;
1113 enum pnfs_iomode iomode
;
1116 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
1117 spin_unlock(&inode
->i_lock
);
1119 /* Send an async layoutreturn so we dont deadlock */
1120 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
1123 spin_unlock(&inode
->i_lock
);
1127 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1128 * when the layout segment list is empty.
1130 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1131 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1132 * deviceid is marked invalid.
1135 _pnfs_return_layout(struct inode
*ino
)
1137 struct pnfs_layout_hdr
*lo
= NULL
;
1138 struct nfs_inode
*nfsi
= NFS_I(ino
);
1139 LIST_HEAD(tmp_list
);
1140 nfs4_stateid stateid
;
1142 bool send
, valid_layout
;
1144 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
1146 spin_lock(&ino
->i_lock
);
1149 spin_unlock(&ino
->i_lock
);
1150 dprintk("NFS: %s no layout to return\n", __func__
);
1153 /* Reference matched in nfs4_layoutreturn_release */
1154 pnfs_get_layout_hdr(lo
);
1155 /* Is there an outstanding layoutreturn ? */
1156 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1157 spin_unlock(&ino
->i_lock
);
1158 if (wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1159 TASK_UNINTERRUPTIBLE
))
1160 goto out_put_layout_hdr
;
1161 spin_lock(&ino
->i_lock
);
1163 valid_layout
= pnfs_layout_is_valid(lo
);
1164 pnfs_clear_layoutcommit(ino
, &tmp_list
);
1165 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
, 0);
1167 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
1168 struct pnfs_layout_range range
= {
1169 .iomode
= IOMODE_ANY
,
1171 .length
= NFS4_MAX_UINT64
,
1173 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
, &range
);
1176 /* Don't send a LAYOUTRETURN if list was initially empty */
1177 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) ||
1179 spin_unlock(&ino
->i_lock
);
1180 dprintk("NFS: %s no layout segments to return\n", __func__
);
1181 goto out_put_layout_hdr
;
1184 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, NULL
);
1185 spin_unlock(&ino
->i_lock
);
1187 status
= pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1189 pnfs_free_lseg_list(&tmp_list
);
1190 pnfs_put_layout_hdr(lo
);
1192 dprintk("<-- %s status: %d\n", __func__
, status
);
1197 pnfs_commit_and_return_layout(struct inode
*inode
)
1199 struct pnfs_layout_hdr
*lo
;
1202 spin_lock(&inode
->i_lock
);
1203 lo
= NFS_I(inode
)->layout
;
1205 spin_unlock(&inode
->i_lock
);
1208 pnfs_get_layout_hdr(lo
);
1209 /* Block new layoutgets and read/write to ds */
1210 lo
->plh_block_lgets
++;
1211 spin_unlock(&inode
->i_lock
);
1212 filemap_fdatawait(inode
->i_mapping
);
1213 ret
= pnfs_layoutcommit_inode(inode
, true);
1215 ret
= _pnfs_return_layout(inode
);
1216 spin_lock(&inode
->i_lock
);
1217 lo
->plh_block_lgets
--;
1218 spin_unlock(&inode
->i_lock
);
1219 pnfs_put_layout_hdr(lo
);
1223 bool pnfs_roc(struct inode
*ino
,
1224 struct nfs4_layoutreturn_args
*args
,
1225 struct nfs4_layoutreturn_res
*res
,
1226 const struct rpc_cred
*cred
)
1228 struct nfs_inode
*nfsi
= NFS_I(ino
);
1229 struct nfs_open_context
*ctx
;
1230 struct nfs4_state
*state
;
1231 struct pnfs_layout_hdr
*lo
;
1232 struct pnfs_layout_segment
*lseg
, *next
;
1233 nfs4_stateid stateid
;
1234 enum pnfs_iomode iomode
= 0;
1235 bool layoutreturn
= false, roc
= false;
1236 bool skip_read
= false;
1238 if (!nfs_have_layout(ino
))
1241 spin_lock(&ino
->i_lock
);
1243 if (!lo
|| !pnfs_layout_is_valid(lo
) ||
1244 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1248 pnfs_get_layout_hdr(lo
);
1249 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1250 spin_unlock(&ino
->i_lock
);
1251 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1252 TASK_UNINTERRUPTIBLE
);
1253 pnfs_put_layout_hdr(lo
);
1257 /* no roc if we hold a delegation */
1258 if (nfs4_check_delegation(ino
, FMODE_READ
)) {
1259 if (nfs4_check_delegation(ino
, FMODE_WRITE
))
1264 list_for_each_entry(ctx
, &nfsi
->open_files
, list
) {
1268 /* Don't return layout if there is open file state */
1269 if (state
->state
& FMODE_WRITE
)
1271 if (state
->state
& FMODE_READ
)
1276 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
) {
1277 if (skip_read
&& lseg
->pls_range
.iomode
== IOMODE_READ
)
1279 /* If we are sending layoutreturn, invalidate all valid lsegs */
1280 if (!test_and_clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
))
1283 * Note: mark lseg for return so pnfs_layout_remove_lseg
1284 * doesn't invalidate the layout for us.
1286 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
1287 if (!mark_lseg_invalid(lseg
, &lo
->plh_return_segs
))
1289 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
1292 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1295 /* ROC in two conditions:
1296 * 1. there are ROC lsegs
1297 * 2. we don't send layoutreturn
1299 /* lo ref dropped in pnfs_roc_release() */
1300 layoutreturn
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
1301 /* If the creds don't match, we can't compound the layoutreturn */
1302 if (!layoutreturn
|| cred
!= lo
->plh_lc_cred
)
1306 pnfs_init_layoutreturn_args(args
, lo
, &stateid
, iomode
);
1307 res
->lrs_present
= 0;
1308 layoutreturn
= false;
1311 spin_unlock(&ino
->i_lock
);
1312 pnfs_layoutcommit_inode(ino
, true);
1314 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1315 if (ld
->prepare_layoutreturn
)
1316 ld
->prepare_layoutreturn(args
);
1317 pnfs_put_layout_hdr(lo
);
1321 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, true);
1322 pnfs_put_layout_hdr(lo
);
1326 void pnfs_roc_release(struct nfs4_layoutreturn_args
*args
,
1327 struct nfs4_layoutreturn_res
*res
,
1330 struct pnfs_layout_hdr
*lo
= args
->layout
;
1331 const nfs4_stateid
*arg_stateid
= NULL
;
1332 const nfs4_stateid
*res_stateid
= NULL
;
1333 struct nfs4_xdr_opaque_data
*ld_private
= args
->ld_private
;
1336 case -NFS4ERR_NOMATCHING_LAYOUT
:
1339 if (res
->lrs_present
)
1340 res_stateid
= &res
->stateid
;
1343 arg_stateid
= &args
->stateid
;
1345 pnfs_layoutreturn_free_lsegs(lo
, arg_stateid
, &args
->range
,
1347 if (ld_private
&& ld_private
->ops
&& ld_private
->ops
->free
)
1348 ld_private
->ops
->free(ld_private
);
1349 pnfs_put_layout_hdr(lo
);
1350 trace_nfs4_layoutreturn_on_close(args
->inode
, 0);
1353 bool pnfs_wait_on_layoutreturn(struct inode
*ino
, struct rpc_task
*task
)
1355 struct nfs_inode
*nfsi
= NFS_I(ino
);
1356 struct pnfs_layout_hdr
*lo
;
1359 /* we might not have grabbed lo reference. so need to check under
1361 spin_lock(&ino
->i_lock
);
1363 if (lo
&& test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1364 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1367 spin_unlock(&ino
->i_lock
);
1372 * Compare two layout segments for sorting into layout cache.
1373 * We want to preferentially return RW over RO layouts, so ensure those
1377 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1378 const struct pnfs_layout_range
*l2
)
1382 /* high offset > low offset */
1383 d
= l1
->offset
- l2
->offset
;
1387 /* short length > long length */
1388 d
= l2
->length
- l1
->length
;
1392 /* read > read/write */
1393 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1397 pnfs_lseg_range_is_after(const struct pnfs_layout_range
*l1
,
1398 const struct pnfs_layout_range
*l2
)
1400 return pnfs_lseg_range_cmp(l1
, l2
) > 0;
1404 pnfs_lseg_no_merge(struct pnfs_layout_segment
*lseg
,
1405 struct pnfs_layout_segment
*old
)
1411 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1412 struct pnfs_layout_segment
*lseg
,
1413 bool (*is_after
)(const struct pnfs_layout_range
*,
1414 const struct pnfs_layout_range
*),
1415 bool (*do_merge
)(struct pnfs_layout_segment
*,
1416 struct pnfs_layout_segment
*),
1417 struct list_head
*free_me
)
1419 struct pnfs_layout_segment
*lp
, *tmp
;
1421 dprintk("%s:Begin\n", __func__
);
1423 list_for_each_entry_safe(lp
, tmp
, &lo
->plh_segs
, pls_list
) {
1424 if (test_bit(NFS_LSEG_VALID
, &lp
->pls_flags
) == 0)
1426 if (do_merge(lseg
, lp
)) {
1427 mark_lseg_invalid(lp
, free_me
);
1430 if (is_after(&lseg
->pls_range
, &lp
->pls_range
))
1432 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1433 dprintk("%s: inserted lseg %p "
1434 "iomode %d offset %llu length %llu before "
1435 "lp %p iomode %d offset %llu length %llu\n",
1436 __func__
, lseg
, lseg
->pls_range
.iomode
,
1437 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1438 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1439 lp
->pls_range
.length
);
1442 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1443 dprintk("%s: inserted lseg %p "
1444 "iomode %d offset %llu length %llu at tail\n",
1445 __func__
, lseg
, lseg
->pls_range
.iomode
,
1446 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1448 pnfs_get_layout_hdr(lo
);
1450 dprintk("%s:Return\n", __func__
);
1452 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg
);
1455 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1456 struct pnfs_layout_segment
*lseg
,
1457 struct list_head
*free_me
)
1459 struct inode
*inode
= lo
->plh_inode
;
1460 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
1462 if (ld
->add_lseg
!= NULL
)
1463 ld
->add_lseg(lo
, lseg
, free_me
);
1465 pnfs_generic_layout_insert_lseg(lo
, lseg
,
1466 pnfs_lseg_range_is_after
,
1471 static struct pnfs_layout_hdr
*
1472 alloc_init_layout_hdr(struct inode
*ino
,
1473 struct nfs_open_context
*ctx
,
1476 struct pnfs_layout_hdr
*lo
;
1478 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1481 atomic_set(&lo
->plh_refcount
, 1);
1482 INIT_LIST_HEAD(&lo
->plh_layouts
);
1483 INIT_LIST_HEAD(&lo
->plh_segs
);
1484 INIT_LIST_HEAD(&lo
->plh_return_segs
);
1485 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1486 lo
->plh_inode
= ino
;
1487 lo
->plh_lc_cred
= get_rpccred(ctx
->cred
);
1488 lo
->plh_flags
|= 1 << NFS_LAYOUT_INVALID_STID
;
1492 static struct pnfs_layout_hdr
*
1493 pnfs_find_alloc_layout(struct inode
*ino
,
1494 struct nfs_open_context
*ctx
,
1496 __releases(&ino
->i_lock
)
1497 __acquires(&ino
->i_lock
)
1499 struct nfs_inode
*nfsi
= NFS_I(ino
);
1500 struct pnfs_layout_hdr
*new = NULL
;
1502 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1504 if (nfsi
->layout
!= NULL
)
1506 spin_unlock(&ino
->i_lock
);
1507 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1508 spin_lock(&ino
->i_lock
);
1510 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1513 } else if (new != NULL
)
1514 pnfs_free_layout_hdr(new);
1516 pnfs_get_layout_hdr(nfsi
->layout
);
1517 return nfsi
->layout
;
1521 * iomode matching rules:
1522 * iomode lseg strict match
1524 * ----- ----- ------ -----
1529 * READ READ N/A true
1530 * READ RW true false
1531 * READ RW false true
1534 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1535 const struct pnfs_layout_range
*range
,
1538 struct pnfs_layout_range range1
;
1540 if ((range
->iomode
== IOMODE_RW
&&
1541 ls_range
->iomode
!= IOMODE_RW
) ||
1542 (range
->iomode
!= ls_range
->iomode
&&
1543 strict_iomode
== true) ||
1544 !pnfs_lseg_range_intersecting(ls_range
, range
))
1547 /* range1 covers only the first byte in the range */
1550 return pnfs_lseg_range_contained(ls_range
, &range1
);
1554 * lookup range in layout
1556 static struct pnfs_layout_segment
*
1557 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1558 struct pnfs_layout_range
*range
,
1561 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1563 dprintk("%s:Begin\n", __func__
);
1565 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1566 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1567 !test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
1568 pnfs_lseg_range_match(&lseg
->pls_range
, range
,
1570 ret
= pnfs_get_lseg(lseg
);
1575 dprintk("%s:Return lseg %p ref %d\n",
1576 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
1581 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1582 * to the MDS or over pNFS
1584 * The nfs_inode read_io and write_io fields are cumulative counters reset
1585 * when there are no layout segments. Note that in pnfs_update_layout iomode
1586 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1589 * A return of true means use MDS I/O.
1592 * If a file's size is smaller than the file size threshold, data accesses
1593 * SHOULD be sent to the metadata server. If an I/O request has a length that
1594 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1595 * server. If both file size and I/O size are provided, the client SHOULD
1596 * reach or exceed both thresholds before sending its read or write
1597 * requests to the data server.
1599 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1600 struct inode
*ino
, int iomode
)
1602 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1603 struct nfs_inode
*nfsi
= NFS_I(ino
);
1604 loff_t fsize
= i_size_read(ino
);
1605 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1610 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1611 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1615 if (t
->bm
& THRESHOLD_RD
) {
1616 dprintk("%s fsize %llu\n", __func__
, fsize
);
1618 if (fsize
< t
->rd_sz
)
1621 if (t
->bm
& THRESHOLD_RD_IO
) {
1622 dprintk("%s nfsi->read_io %llu\n", __func__
,
1625 if (nfsi
->read_io
< t
->rd_io_sz
)
1630 if (t
->bm
& THRESHOLD_WR
) {
1631 dprintk("%s fsize %llu\n", __func__
, fsize
);
1633 if (fsize
< t
->wr_sz
)
1636 if (t
->bm
& THRESHOLD_WR_IO
) {
1637 dprintk("%s nfsi->write_io %llu\n", __func__
,
1640 if (nfsi
->write_io
< t
->wr_io_sz
)
1645 if (size_set
&& io_set
) {
1648 } else if (size
|| io
)
1651 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1655 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr
*lo
)
1658 * send layoutcommit as it can hold up layoutreturn due to lseg
1661 pnfs_layoutcommit_inode(lo
->plh_inode
, false);
1662 return !wait_on_bit_action(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1663 nfs_wait_bit_killable
,
1664 TASK_UNINTERRUPTIBLE
);
1667 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr
*lo
)
1669 unsigned long *bitlock
= &lo
->plh_flags
;
1671 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET
, bitlock
);
1672 smp_mb__after_atomic();
1673 wake_up_bit(bitlock
, NFS_LAYOUT_FIRST_LAYOUTGET
);
1677 * Layout segment is retreived from the server if not cached.
1678 * The appropriate layout segment is referenced and returned to the caller.
1680 struct pnfs_layout_segment
*
1681 pnfs_update_layout(struct inode
*ino
,
1682 struct nfs_open_context
*ctx
,
1685 enum pnfs_iomode iomode
,
1689 struct pnfs_layout_range arg
= {
1695 struct nfs_server
*server
= NFS_SERVER(ino
);
1696 struct nfs_client
*clp
= server
->nfs_client
;
1697 struct pnfs_layout_hdr
*lo
= NULL
;
1698 struct pnfs_layout_segment
*lseg
= NULL
;
1699 nfs4_stateid stateid
;
1701 unsigned long giveup
= jiffies
+ (clp
->cl_lease_time
<< 1);
1704 if (!pnfs_enabled_sb(NFS_SERVER(ino
))) {
1705 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1706 PNFS_UPDATE_LAYOUT_NO_PNFS
);
1710 if (iomode
== IOMODE_READ
&& i_size_read(ino
) == 0) {
1711 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1712 PNFS_UPDATE_LAYOUT_RD_ZEROLEN
);
1716 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
)) {
1717 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1718 PNFS_UPDATE_LAYOUT_MDSTHRESH
);
1723 nfs4_client_recover_expired_lease(clp
);
1725 spin_lock(&ino
->i_lock
);
1726 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1728 spin_unlock(&ino
->i_lock
);
1729 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1730 PNFS_UPDATE_LAYOUT_NOMEM
);
1734 /* Do we even need to bother with this? */
1735 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1736 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1737 PNFS_UPDATE_LAYOUT_BULK_RECALL
);
1738 dprintk("%s matches recall, use MDS\n", __func__
);
1742 /* if LAYOUTGET already failed once we don't try again */
1743 if (pnfs_layout_io_test_failed(lo
, iomode
)) {
1744 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1745 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL
);
1749 lseg
= pnfs_find_lseg(lo
, &arg
, strict_iomode
);
1751 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1752 PNFS_UPDATE_LAYOUT_FOUND_CACHED
);
1756 if (!nfs4_valid_open_stateid(ctx
->state
)) {
1757 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1758 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1763 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1764 * stateid, or it has been invalidated, then we must use the open
1767 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
)) {
1770 * The first layoutget for the file. Need to serialize per
1771 * RFC 5661 Errata 3208.
1773 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
,
1775 spin_unlock(&ino
->i_lock
);
1776 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_FIRST_LAYOUTGET
,
1777 TASK_UNINTERRUPTIBLE
);
1778 pnfs_put_layout_hdr(lo
);
1779 dprintk("%s retrying\n", __func__
);
1784 if (nfs4_select_rw_stateid(ctx
->state
,
1785 iomode
== IOMODE_RW
? FMODE_WRITE
: FMODE_READ
,
1786 NULL
, &stateid
, NULL
) != 0) {
1787 trace_pnfs_update_layout(ino
, pos
, count
,
1789 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1793 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1797 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1798 * for LAYOUTRETURN even if first is true.
1800 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1801 spin_unlock(&ino
->i_lock
);
1802 dprintk("%s wait for layoutreturn\n", __func__
);
1803 if (pnfs_prepare_to_retry_layoutget(lo
)) {
1805 pnfs_clear_first_layoutget(lo
);
1806 pnfs_put_layout_hdr(lo
);
1807 dprintk("%s retrying\n", __func__
);
1808 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
,
1809 lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1812 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1813 PNFS_UPDATE_LAYOUT_RETURN
);
1814 goto out_put_layout_hdr
;
1817 if (pnfs_layoutgets_blocked(lo
)) {
1818 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1819 PNFS_UPDATE_LAYOUT_BLOCKED
);
1822 atomic_inc(&lo
->plh_outstanding
);
1823 spin_unlock(&ino
->i_lock
);
1825 if (list_empty(&lo
->plh_layouts
)) {
1826 /* The lo must be on the clp list if there is any
1827 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1829 spin_lock(&clp
->cl_lock
);
1830 if (list_empty(&lo
->plh_layouts
))
1831 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1832 spin_unlock(&clp
->cl_lock
);
1835 pg_offset
= arg
.offset
& ~PAGE_MASK
;
1837 arg
.offset
-= pg_offset
;
1838 arg
.length
+= pg_offset
;
1840 if (arg
.length
!= NFS4_MAX_UINT64
)
1841 arg
.length
= PAGE_ALIGN(arg
.length
);
1843 lseg
= send_layoutget(lo
, ctx
, &stateid
, &arg
, &timeout
, gfp_flags
);
1844 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1845 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET
);
1846 atomic_dec(&lo
->plh_outstanding
);
1848 switch(PTR_ERR(lseg
)) {
1850 if (time_after(jiffies
, giveup
))
1853 case -ERECALLCONFLICT
:
1854 /* Huh? We hold no layouts, how is there a recall? */
1859 /* Destroy the existing layout and start over */
1860 if (time_after(jiffies
, giveup
))
1861 pnfs_destroy_layout(NFS_I(ino
));
1866 if (!nfs_error_is_fatal(PTR_ERR(lseg
))) {
1867 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1870 goto out_put_layout_hdr
;
1874 pnfs_clear_first_layoutget(lo
);
1875 trace_pnfs_update_layout(ino
, pos
, count
,
1876 iomode
, lo
, lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1877 pnfs_put_layout_hdr(lo
);
1881 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1886 pnfs_clear_first_layoutget(lo
);
1887 pnfs_put_layout_hdr(lo
);
1889 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1890 "(%s, offset: %llu, length: %llu)\n",
1891 __func__
, ino
->i_sb
->s_id
,
1892 (unsigned long long)NFS_FILEID(ino
),
1893 IS_ERR_OR_NULL(lseg
) ? "not found" : "found",
1894 iomode
==IOMODE_RW
? "read/write" : "read-only",
1895 (unsigned long long)pos
,
1896 (unsigned long long)count
);
1899 spin_unlock(&ino
->i_lock
);
1900 goto out_put_layout_hdr
;
1902 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1905 pnfs_sanity_check_layout_range(struct pnfs_layout_range
*range
)
1907 switch (range
->iomode
) {
1914 if (range
->offset
== NFS4_MAX_UINT64
)
1916 if (range
->length
== 0)
1918 if (range
->length
!= NFS4_MAX_UINT64
&&
1919 range
->length
> NFS4_MAX_UINT64
- range
->offset
)
1924 struct pnfs_layout_segment
*
1925 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1927 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1928 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1929 struct pnfs_layout_segment
*lseg
;
1930 struct inode
*ino
= lo
->plh_inode
;
1933 if (!pnfs_sanity_check_layout_range(&res
->range
))
1934 return ERR_PTR(-EINVAL
);
1936 /* Inject layout blob into I/O device driver */
1937 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1938 if (IS_ERR_OR_NULL(lseg
)) {
1940 lseg
= ERR_PTR(-ENOMEM
);
1942 dprintk("%s: Could not allocate layout: error %ld\n",
1943 __func__
, PTR_ERR(lseg
));
1947 pnfs_init_lseg(lo
, lseg
, &res
->range
, &res
->stateid
);
1949 spin_lock(&ino
->i_lock
);
1950 if (pnfs_layoutgets_blocked(lo
)) {
1951 dprintk("%s forget reply due to state\n", __func__
);
1955 if (!pnfs_layout_is_valid(lo
)) {
1956 /* We have a completely new layout */
1957 pnfs_set_layout_stateid(lo
, &res
->stateid
, true);
1958 } else if (nfs4_stateid_match_other(&lo
->plh_stateid
, &res
->stateid
)) {
1959 /* existing state ID, make sure the sequence number matches. */
1960 if (pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
1961 dprintk("%s forget reply due to sequence\n", __func__
);
1964 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1967 * We got an entirely new state ID. Mark all segments for the
1968 * inode invalid, and retry the layoutget
1970 pnfs_mark_layout_stateid_invalid(lo
, &free_me
);
1974 pnfs_get_lseg(lseg
);
1975 pnfs_layout_insert_lseg(lo
, lseg
, &free_me
);
1978 if (res
->return_on_close
)
1979 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1981 spin_unlock(&ino
->i_lock
);
1982 pnfs_free_lseg_list(&free_me
);
1986 spin_unlock(&ino
->i_lock
);
1987 lseg
->pls_layout
= lo
;
1988 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1989 if (!pnfs_layout_is_valid(lo
))
1990 nfs_commit_inode(ino
, 0);
1991 return ERR_PTR(-EAGAIN
);
1995 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
1996 * @lo: pointer to layout header
1997 * @tmp_list: list header to be used with pnfs_free_lseg_list()
1998 * @return_range: describe layout segment ranges to be returned
2000 * This function is mainly intended for use by layoutrecall. It attempts
2001 * to free the layout segment immediately, or else to mark it for return
2002 * as soon as its reference count drops to zero.
2005 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr
*lo
,
2006 struct list_head
*tmp_list
,
2007 const struct pnfs_layout_range
*return_range
,
2010 struct pnfs_layout_segment
*lseg
, *next
;
2013 dprintk("%s:Begin lo %p\n", __func__
, lo
);
2015 if (list_empty(&lo
->plh_segs
))
2018 assert_spin_locked(&lo
->plh_inode
->i_lock
);
2020 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
2021 if (pnfs_match_lseg_recall(lseg
, return_range
, seq
)) {
2022 dprintk("%s: marking lseg %p iomode %d "
2023 "offset %llu length %llu\n", __func__
,
2024 lseg
, lseg
->pls_range
.iomode
,
2025 lseg
->pls_range
.offset
,
2026 lseg
->pls_range
.length
);
2027 if (mark_lseg_invalid(lseg
, tmp_list
))
2030 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
2034 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
2039 void pnfs_error_mark_layout_for_return(struct inode
*inode
,
2040 struct pnfs_layout_segment
*lseg
)
2042 struct pnfs_layout_hdr
*lo
= NFS_I(inode
)->layout
;
2043 struct pnfs_layout_range range
= {
2044 .iomode
= lseg
->pls_range
.iomode
,
2046 .length
= NFS4_MAX_UINT64
,
2048 bool return_now
= false;
2050 spin_lock(&inode
->i_lock
);
2051 if (!pnfs_layout_is_valid(lo
)) {
2052 spin_unlock(&inode
->i_lock
);
2055 pnfs_set_plh_return_info(lo
, range
.iomode
, 0);
2057 * mark all matching lsegs so that we are sure to have no live
2058 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2061 if (!pnfs_mark_matching_lsegs_return(lo
, &lo
->plh_return_segs
, &range
, 0)) {
2062 nfs4_stateid stateid
;
2063 enum pnfs_iomode iomode
;
2065 return_now
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
2066 spin_unlock(&inode
->i_lock
);
2068 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
2070 spin_unlock(&inode
->i_lock
);
2071 nfs_commit_inode(inode
, 0);
2074 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return
);
2077 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor
*pgio
)
2079 if (pgio
->pg_lseg
== NULL
||
2080 test_bit(NFS_LSEG_VALID
, &pgio
->pg_lseg
->pls_flags
))
2082 pnfs_put_lseg(pgio
->pg_lseg
);
2083 pgio
->pg_lseg
= NULL
;
2085 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout
);
2088 * Check for any intersection between the request and the pgio->pg_lseg,
2089 * and if none, put this pgio->pg_lseg away.
2092 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2094 if (pgio
->pg_lseg
&& !pnfs_lseg_request_intersecting(pgio
->pg_lseg
, req
)) {
2095 pnfs_put_lseg(pgio
->pg_lseg
);
2096 pgio
->pg_lseg
= NULL
;
2101 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2103 u64 rd_size
= req
->wb_bytes
;
2105 pnfs_generic_pg_check_layout(pgio
);
2106 pnfs_generic_pg_check_range(pgio
, req
);
2107 if (pgio
->pg_lseg
== NULL
) {
2108 if (pgio
->pg_dreq
== NULL
)
2109 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
2111 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
2113 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
2120 if (IS_ERR(pgio
->pg_lseg
)) {
2121 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2122 pgio
->pg_lseg
= NULL
;
2126 /* If no lseg, fall back to read through mds */
2127 if (pgio
->pg_lseg
== NULL
)
2128 nfs_pageio_reset_read_mds(pgio
);
2131 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
2134 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
2135 struct nfs_page
*req
, u64 wb_size
)
2137 pnfs_generic_pg_check_layout(pgio
);
2138 pnfs_generic_pg_check_range(pgio
, req
);
2139 if (pgio
->pg_lseg
== NULL
) {
2140 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
2147 if (IS_ERR(pgio
->pg_lseg
)) {
2148 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2149 pgio
->pg_lseg
= NULL
;
2153 /* If no lseg, fall back to write through mds */
2154 if (pgio
->pg_lseg
== NULL
)
2155 nfs_pageio_reset_write_mds(pgio
);
2157 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
2160 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor
*desc
)
2162 if (desc
->pg_lseg
) {
2163 pnfs_put_lseg(desc
->pg_lseg
);
2164 desc
->pg_lseg
= NULL
;
2167 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup
);
2170 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2171 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2174 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
,
2175 struct nfs_page
*prev
, struct nfs_page
*req
)
2178 u64 seg_end
, req_start
, seg_left
;
2180 size
= nfs_generic_pg_test(pgio
, prev
, req
);
2185 * 'size' contains the number of bytes left in the current page (up
2186 * to the original size asked for in @req->wb_bytes).
2188 * Calculate how many bytes are left in the layout segment
2189 * and if there are less bytes than 'size', return that instead.
2191 * Please also note that 'end_offset' is actually the offset of the
2192 * first byte that lies outside the pnfs_layout_range. FIXME?
2195 if (pgio
->pg_lseg
) {
2196 seg_end
= pnfs_end_offset(pgio
->pg_lseg
->pls_range
.offset
,
2197 pgio
->pg_lseg
->pls_range
.length
);
2198 req_start
= req_offset(req
);
2200 /* start of request is past the last byte of this segment */
2201 if (req_start
>= seg_end
)
2204 /* adjust 'size' iff there are fewer bytes left in the
2205 * segment than what nfs_generic_pg_test returned */
2206 seg_left
= seg_end
- req_start
;
2207 if (seg_left
< size
)
2208 size
= (unsigned int)seg_left
;
2213 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
2215 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2217 struct nfs_pageio_descriptor pgio
;
2219 /* Resend all requests through the MDS */
2220 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
2221 hdr
->completion_ops
);
2222 set_bit(NFS_CONTEXT_RESEND_WRITES
, &hdr
->args
.context
->flags
);
2223 return nfs_pageio_resend(&pgio
, hdr
);
2225 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
2227 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
2230 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
2231 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2232 PNFS_LAYOUTRET_ON_ERROR
) {
2233 pnfs_return_layout(hdr
->inode
);
2235 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2236 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
2240 * Called by non rpc-based layout drivers
2242 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
2244 if (likely(!hdr
->pnfs_error
)) {
2245 pnfs_set_layoutcommit(hdr
->inode
, hdr
->lseg
,
2246 hdr
->mds_offset
+ hdr
->res
.count
);
2247 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2249 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
2250 if (unlikely(hdr
->pnfs_error
))
2251 pnfs_ld_handle_write_error(hdr
);
2252 hdr
->mds_ops
->rpc_release(hdr
);
2254 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
2257 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
2258 struct nfs_pgio_header
*hdr
)
2260 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2262 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2263 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2264 nfs_pageio_reset_write_mds(desc
);
2265 mirror
->pg_recoalesce
= 1;
2267 hdr
->completion_ops
->completion(hdr
);
2270 static enum pnfs_try_status
2271 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
2272 const struct rpc_call_ops
*call_ops
,
2273 struct pnfs_layout_segment
*lseg
,
2276 struct inode
*inode
= hdr
->inode
;
2277 enum pnfs_try_status trypnfs
;
2278 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2280 hdr
->mds_ops
= call_ops
;
2282 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
2283 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
2284 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
2285 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2286 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
2287 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2292 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
2293 struct nfs_pgio_header
*hdr
, int how
)
2295 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2296 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2297 enum pnfs_try_status trypnfs
;
2299 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
2301 case PNFS_NOT_ATTEMPTED
:
2302 pnfs_write_through_mds(desc
, hdr
);
2303 case PNFS_ATTEMPTED
:
2305 case PNFS_TRY_AGAIN
:
2306 /* cleanup hdr and prepare to redo pnfs */
2307 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2308 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2309 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
2310 mirror
->pg_recoalesce
= 1;
2312 hdr
->mds_ops
->rpc_release(hdr
);
2316 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
2318 pnfs_put_lseg(hdr
->lseg
);
2319 nfs_pgio_header_free(hdr
);
2323 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
2325 struct nfs_pgio_header
*hdr
;
2328 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2330 desc
->pg_error
= -ENOMEM
;
2331 return desc
->pg_error
;
2333 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
2335 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2336 ret
= nfs_generic_pgio(desc
, hdr
);
2338 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
2342 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
2344 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2346 struct nfs_pageio_descriptor pgio
;
2348 /* Resend all requests through the MDS */
2349 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
2350 return nfs_pageio_resend(&pgio
, hdr
);
2352 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
2354 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
2356 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
2357 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2358 PNFS_LAYOUTRET_ON_ERROR
) {
2359 pnfs_return_layout(hdr
->inode
);
2361 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2362 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
2366 * Called by non rpc-based layout drivers
2368 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
2370 if (likely(!hdr
->pnfs_error
))
2371 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2372 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
2373 if (unlikely(hdr
->pnfs_error
))
2374 pnfs_ld_handle_read_error(hdr
);
2375 hdr
->mds_ops
->rpc_release(hdr
);
2377 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
2380 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
2381 struct nfs_pgio_header
*hdr
)
2383 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2385 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2386 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2387 nfs_pageio_reset_read_mds(desc
);
2388 mirror
->pg_recoalesce
= 1;
2390 hdr
->completion_ops
->completion(hdr
);
2394 * Call the appropriate parallel I/O subsystem read function.
2396 static enum pnfs_try_status
2397 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
2398 const struct rpc_call_ops
*call_ops
,
2399 struct pnfs_layout_segment
*lseg
)
2401 struct inode
*inode
= hdr
->inode
;
2402 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2403 enum pnfs_try_status trypnfs
;
2405 hdr
->mds_ops
= call_ops
;
2407 dprintk("%s: Reading ino:%lu %u@%llu\n",
2408 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
2410 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
2411 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2412 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
2413 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2417 /* Resend all requests through pnfs. */
2418 void pnfs_read_resend_pnfs(struct nfs_pgio_header
*hdr
)
2420 struct nfs_pageio_descriptor pgio
;
2422 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2423 /* Prevent deadlocks with layoutreturn! */
2424 pnfs_put_lseg(hdr
->lseg
);
2427 nfs_pageio_init_read(&pgio
, hdr
->inode
, false,
2428 hdr
->completion_ops
);
2429 hdr
->task
.tk_status
= nfs_pageio_resend(&pgio
, hdr
);
2432 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs
);
2435 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
2437 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2438 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2439 enum pnfs_try_status trypnfs
;
2441 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
2443 case PNFS_NOT_ATTEMPTED
:
2444 pnfs_read_through_mds(desc
, hdr
);
2445 case PNFS_ATTEMPTED
:
2447 case PNFS_TRY_AGAIN
:
2448 /* cleanup hdr and prepare to redo pnfs */
2449 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2450 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2451 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
2452 mirror
->pg_recoalesce
= 1;
2454 hdr
->mds_ops
->rpc_release(hdr
);
2458 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
2460 pnfs_put_lseg(hdr
->lseg
);
2461 nfs_pgio_header_free(hdr
);
2465 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
2467 struct nfs_pgio_header
*hdr
;
2470 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2472 desc
->pg_error
= -ENOMEM
;
2473 return desc
->pg_error
;
2475 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
2476 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2477 ret
= nfs_generic_pgio(desc
, hdr
);
2479 pnfs_do_read(desc
, hdr
);
2482 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
2484 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
2486 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
2488 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
2489 smp_mb__after_atomic();
2490 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
2494 * There can be multiple RW segments.
2496 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
2498 struct pnfs_layout_segment
*lseg
;
2500 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
2501 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
2502 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
2503 list_add(&lseg
->pls_lc_list
, listp
);
2507 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
2509 struct pnfs_layout_segment
*lseg
, *tmp
;
2511 /* Matched by references in pnfs_set_layoutcommit */
2512 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
2513 list_del_init(&lseg
->pls_lc_list
);
2514 pnfs_put_lseg(lseg
);
2517 pnfs_clear_layoutcommitting(inode
);
2520 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
2522 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
2524 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
2527 pnfs_set_layoutcommit(struct inode
*inode
, struct pnfs_layout_segment
*lseg
,
2530 struct nfs_inode
*nfsi
= NFS_I(inode
);
2531 bool mark_as_dirty
= false;
2533 spin_lock(&inode
->i_lock
);
2534 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
2535 nfsi
->layout
->plh_lwb
= end_pos
;
2536 mark_as_dirty
= true;
2537 dprintk("%s: Set layoutcommit for inode %lu ",
2538 __func__
, inode
->i_ino
);
2539 } else if (end_pos
> nfsi
->layout
->plh_lwb
)
2540 nfsi
->layout
->plh_lwb
= end_pos
;
2541 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
)) {
2542 /* references matched in nfs4_layoutcommit_release */
2543 pnfs_get_lseg(lseg
);
2545 spin_unlock(&inode
->i_lock
);
2546 dprintk("%s: lseg %p end_pos %llu\n",
2547 __func__
, lseg
, nfsi
->layout
->plh_lwb
);
2549 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2550 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2552 mark_inode_dirty_sync(inode
);
2554 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
2556 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
2558 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
2560 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
2561 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
2562 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
2566 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2567 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2568 * data to disk to allow the server to recover the data if it crashes.
2569 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2570 * is off, and a COMMIT is sent to a data server, or
2571 * if WRITEs to a data server return NFS_DATA_SYNC.
2574 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
2576 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2577 struct nfs4_layoutcommit_data
*data
;
2578 struct nfs_inode
*nfsi
= NFS_I(inode
);
2582 if (!pnfs_layoutcommit_outstanding(inode
))
2585 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
2588 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
2591 status
= wait_on_bit_lock_action(&nfsi
->flags
,
2592 NFS_INO_LAYOUTCOMMITTING
,
2593 nfs_wait_bit_killable
,
2600 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2601 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
2603 goto clear_layoutcommitting
;
2606 spin_lock(&inode
->i_lock
);
2607 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
2610 INIT_LIST_HEAD(&data
->lseg_list
);
2611 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
2613 end_pos
= nfsi
->layout
->plh_lwb
;
2615 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
2616 spin_unlock(&inode
->i_lock
);
2618 data
->args
.inode
= inode
;
2619 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
2620 nfs_fattr_init(&data
->fattr
);
2621 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
2622 data
->res
.fattr
= &data
->fattr
;
2624 data
->args
.lastbytewritten
= end_pos
- 1;
2626 data
->args
.lastbytewritten
= U64_MAX
;
2627 data
->res
.server
= NFS_SERVER(inode
);
2629 if (ld
->prepare_layoutcommit
) {
2630 status
= ld
->prepare_layoutcommit(&data
->args
);
2632 put_rpccred(data
->cred
);
2633 spin_lock(&inode
->i_lock
);
2634 set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
);
2635 if (end_pos
> nfsi
->layout
->plh_lwb
)
2636 nfsi
->layout
->plh_lwb
= end_pos
;
2642 status
= nfs4_proc_layoutcommit(data
, sync
);
2645 mark_inode_dirty_sync(inode
);
2646 dprintk("<-- %s status %d\n", __func__
, status
);
2649 spin_unlock(&inode
->i_lock
);
2651 clear_layoutcommitting
:
2652 pnfs_clear_layoutcommitting(inode
);
2655 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode
);
2658 pnfs_generic_sync(struct inode
*inode
, bool datasync
)
2660 return pnfs_layoutcommit_inode(inode
, true);
2662 EXPORT_SYMBOL_GPL(pnfs_generic_sync
);
2664 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
2666 struct nfs4_threshold
*thp
;
2668 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
2670 dprintk("%s mdsthreshold allocation failed\n", __func__
);
2676 #if IS_ENABLED(CONFIG_NFS_V4_2)
2678 pnfs_report_layoutstat(struct inode
*inode
, gfp_t gfp_flags
)
2680 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2681 struct nfs_server
*server
= NFS_SERVER(inode
);
2682 struct nfs_inode
*nfsi
= NFS_I(inode
);
2683 struct nfs42_layoutstat_data
*data
;
2684 struct pnfs_layout_hdr
*hdr
;
2687 if (!pnfs_enabled_sb(server
) || !ld
->prepare_layoutstats
)
2690 if (!nfs_server_capable(inode
, NFS_CAP_LAYOUTSTATS
))
2693 if (test_and_set_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
))
2696 spin_lock(&inode
->i_lock
);
2697 if (!NFS_I(inode
)->layout
) {
2698 spin_unlock(&inode
->i_lock
);
2699 goto out_clear_layoutstats
;
2701 hdr
= NFS_I(inode
)->layout
;
2702 pnfs_get_layout_hdr(hdr
);
2703 spin_unlock(&inode
->i_lock
);
2705 data
= kzalloc(sizeof(*data
), gfp_flags
);
2711 data
->args
.fh
= NFS_FH(inode
);
2712 data
->args
.inode
= inode
;
2713 status
= ld
->prepare_layoutstats(&data
->args
);
2717 status
= nfs42_proc_layoutstats_generic(NFS_SERVER(inode
), data
);
2720 dprintk("%s returns %d\n", __func__
, status
);
2726 pnfs_put_layout_hdr(hdr
);
2727 out_clear_layoutstats
:
2728 smp_mb__before_atomic();
2729 clear_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
);
2730 smp_mb__after_atomic();
2733 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat
);
2736 unsigned int layoutstats_timer
;
2737 module_param(layoutstats_timer
, uint
, 0644);
2738 EXPORT_SYMBOL_GPL(layoutstats_timer
);