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>
37 #define NFSDBG_FACILITY NFSDBG_PNFS
42 * protects pnfs_modules_tbl.
44 static DEFINE_SPINLOCK(pnfs_spinlock
);
47 * pnfs_modules_tbl holds all pnfs modules
49 static LIST_HEAD(pnfs_modules_tbl
);
51 /* Return the registered pnfs layout driver module matching given id */
52 static struct pnfs_layoutdriver_type
*
53 find_pnfs_driver_locked(u32 id
)
55 struct pnfs_layoutdriver_type
*local
;
57 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
62 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
66 static struct pnfs_layoutdriver_type
*
67 find_pnfs_driver(u32 id
)
69 struct pnfs_layoutdriver_type
*local
;
71 spin_lock(&pnfs_spinlock
);
72 local
= find_pnfs_driver_locked(id
);
73 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
74 dprintk("%s: Could not grab reference on module\n", __func__
);
77 spin_unlock(&pnfs_spinlock
);
82 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
84 if (nfss
->pnfs_curr_ld
) {
85 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
86 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
87 /* Decrement the MDS count. Purge the deviceid cache if zero */
88 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
89 nfs4_deviceid_purge_client(nfss
->nfs_client
);
90 module_put(nfss
->pnfs_curr_ld
->owner
);
92 nfss
->pnfs_curr_ld
= NULL
;
96 * Try to set the server's pnfs module to the pnfs layout type specified by id.
97 * Currently only one pNFS layout driver per filesystem is supported.
99 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
102 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
105 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
109 if (!(server
->nfs_client
->cl_exchange_flags
&
110 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
111 printk(KERN_ERR
"NFS: %s: id %u cl_exchange_flags 0x%x\n",
112 __func__
, id
, server
->nfs_client
->cl_exchange_flags
);
115 ld_type
= find_pnfs_driver(id
);
117 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
, id
);
118 ld_type
= find_pnfs_driver(id
);
120 dprintk("%s: No pNFS module found for %u.\n",
125 server
->pnfs_curr_ld
= ld_type
;
126 if (ld_type
->set_layoutdriver
127 && ld_type
->set_layoutdriver(server
, mntfh
)) {
128 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
129 "driver %u.\n", __func__
, id
);
130 module_put(ld_type
->owner
);
133 /* Bump the MDS count */
134 atomic_inc(&server
->nfs_client
->cl_mds_count
);
136 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
140 dprintk("%s: Using NFSv4 I/O\n", __func__
);
141 server
->pnfs_curr_ld
= NULL
;
145 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
147 int status
= -EINVAL
;
148 struct pnfs_layoutdriver_type
*tmp
;
150 if (ld_type
->id
== 0) {
151 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
154 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
155 printk(KERN_ERR
"NFS: %s Layout driver must provide "
156 "alloc_lseg and free_lseg.\n", __func__
);
160 spin_lock(&pnfs_spinlock
);
161 tmp
= find_pnfs_driver_locked(ld_type
->id
);
163 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
165 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
168 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
169 __func__
, ld_type
->id
);
171 spin_unlock(&pnfs_spinlock
);
175 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
178 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
180 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
181 spin_lock(&pnfs_spinlock
);
182 list_del(&ld_type
->pnfs_tblid
);
183 spin_unlock(&pnfs_spinlock
);
185 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
188 * pNFS client layout cache
191 /* Need to hold i_lock if caller does not already hold reference */
193 get_layout_hdr(struct pnfs_layout_hdr
*lo
)
195 atomic_inc(&lo
->plh_refcount
);
198 static struct pnfs_layout_hdr
*
199 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
201 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
202 return ld
->alloc_layout_hdr
? ld
->alloc_layout_hdr(ino
, gfp_flags
) :
203 kzalloc(sizeof(struct pnfs_layout_hdr
), gfp_flags
);
207 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
209 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(lo
->plh_inode
)->pnfs_curr_ld
;
210 put_rpccred(lo
->plh_lc_cred
);
211 return ld
->alloc_layout_hdr
? ld
->free_layout_hdr(lo
) : kfree(lo
);
215 destroy_layout_hdr(struct pnfs_layout_hdr
*lo
)
217 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
218 BUG_ON(!list_empty(&lo
->plh_layouts
));
219 NFS_I(lo
->plh_inode
)->layout
= NULL
;
220 pnfs_free_layout_hdr(lo
);
224 put_layout_hdr_locked(struct pnfs_layout_hdr
*lo
)
226 if (atomic_dec_and_test(&lo
->plh_refcount
))
227 destroy_layout_hdr(lo
);
231 put_layout_hdr(struct pnfs_layout_hdr
*lo
)
233 struct inode
*inode
= lo
->plh_inode
;
235 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
236 destroy_layout_hdr(lo
);
237 spin_unlock(&inode
->i_lock
);
242 init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
)
244 INIT_LIST_HEAD(&lseg
->pls_list
);
245 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
246 atomic_set(&lseg
->pls_refcount
, 1);
248 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
249 lseg
->pls_layout
= lo
;
252 static void free_lseg(struct pnfs_layout_segment
*lseg
)
254 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
256 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
257 /* Matched by get_layout_hdr in pnfs_insert_layout */
258 put_layout_hdr(NFS_I(ino
)->layout
);
262 put_lseg_common(struct pnfs_layout_segment
*lseg
)
264 struct inode
*inode
= lseg
->pls_layout
->plh_inode
;
266 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
267 list_del_init(&lseg
->pls_list
);
268 if (list_empty(&lseg
->pls_layout
->plh_segs
)) {
269 set_bit(NFS_LAYOUT_DESTROYED
, &lseg
->pls_layout
->plh_flags
);
270 /* Matched by initial refcount set in alloc_init_layout_hdr */
271 put_layout_hdr_locked(lseg
->pls_layout
);
273 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
277 put_lseg(struct pnfs_layout_segment
*lseg
)
284 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
285 atomic_read(&lseg
->pls_refcount
),
286 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
287 inode
= lseg
->pls_layout
->plh_inode
;
288 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
291 put_lseg_common(lseg
);
292 list_add(&lseg
->pls_list
, &free_me
);
293 spin_unlock(&inode
->i_lock
);
294 pnfs_free_lseg_list(&free_me
);
297 EXPORT_SYMBOL_GPL(put_lseg
);
300 end_offset(u64 start
, u64 len
)
305 return end
>= start
? end
: NFS4_MAX_UINT64
;
308 /* last octet in a range */
310 last_byte_offset(u64 start
, u64 len
)
316 return end
> start
? end
- 1 : NFS4_MAX_UINT64
;
320 * is l2 fully contained in l1?
322 * [----------------------------------)
327 lo_seg_contained(struct pnfs_layout_range
*l1
,
328 struct pnfs_layout_range
*l2
)
330 u64 start1
= l1
->offset
;
331 u64 end1
= end_offset(start1
, l1
->length
);
332 u64 start2
= l2
->offset
;
333 u64 end2
= end_offset(start2
, l2
->length
);
335 return (start1
<= start2
) && (end1
>= end2
);
339 * is l1 and l2 intersecting?
341 * [----------------------------------)
346 lo_seg_intersecting(struct pnfs_layout_range
*l1
,
347 struct pnfs_layout_range
*l2
)
349 u64 start1
= l1
->offset
;
350 u64 end1
= end_offset(start1
, l1
->length
);
351 u64 start2
= l2
->offset
;
352 u64 end2
= end_offset(start2
, l2
->length
);
354 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
355 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
359 should_free_lseg(struct pnfs_layout_range
*lseg_range
,
360 struct pnfs_layout_range
*recall_range
)
362 return (recall_range
->iomode
== IOMODE_ANY
||
363 lseg_range
->iomode
== recall_range
->iomode
) &&
364 lo_seg_intersecting(lseg_range
, recall_range
);
367 /* Returns 1 if lseg is removed from list, 0 otherwise */
368 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
369 struct list_head
*tmp_list
)
373 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
374 /* Remove the reference keeping the lseg in the
375 * list. It will now be removed when all
376 * outstanding io is finished.
378 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
379 atomic_read(&lseg
->pls_refcount
));
380 if (atomic_dec_and_test(&lseg
->pls_refcount
)) {
381 put_lseg_common(lseg
);
382 list_add(&lseg
->pls_list
, tmp_list
);
389 /* Returns count of number of matching invalid lsegs remaining in list
393 mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
394 struct list_head
*tmp_list
,
395 struct pnfs_layout_range
*recall_range
)
397 struct pnfs_layout_segment
*lseg
, *next
;
398 int invalid
= 0, removed
= 0;
400 dprintk("%s:Begin lo %p\n", __func__
, lo
);
402 if (list_empty(&lo
->plh_segs
)) {
403 /* Reset MDS Threshold I/O counters */
404 NFS_I(lo
->plh_inode
)->write_io
= 0;
405 NFS_I(lo
->plh_inode
)->read_io
= 0;
406 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
))
407 put_layout_hdr_locked(lo
);
410 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
412 should_free_lseg(&lseg
->pls_range
, recall_range
)) {
413 dprintk("%s: freeing lseg %p iomode %d "
414 "offset %llu length %llu\n", __func__
,
415 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_range
.offset
,
416 lseg
->pls_range
.length
);
418 removed
+= mark_lseg_invalid(lseg
, tmp_list
);
420 dprintk("%s:Return %i\n", __func__
, invalid
- removed
);
421 return invalid
- removed
;
424 /* note free_me must contain lsegs from a single layout_hdr */
426 pnfs_free_lseg_list(struct list_head
*free_me
)
428 struct pnfs_layout_segment
*lseg
, *tmp
;
429 struct pnfs_layout_hdr
*lo
;
431 if (list_empty(free_me
))
434 lo
= list_first_entry(free_me
, struct pnfs_layout_segment
,
435 pls_list
)->pls_layout
;
437 if (test_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
)) {
438 struct nfs_client
*clp
;
440 clp
= NFS_SERVER(lo
->plh_inode
)->nfs_client
;
441 spin_lock(&clp
->cl_lock
);
442 list_del_init(&lo
->plh_layouts
);
443 spin_unlock(&clp
->cl_lock
);
445 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
446 list_del(&lseg
->pls_list
);
452 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
454 struct pnfs_layout_hdr
*lo
;
457 spin_lock(&nfsi
->vfs_inode
.i_lock
);
460 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
461 mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
463 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
464 pnfs_free_lseg_list(&tmp_list
);
466 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
469 * Called by the state manger to remove all layouts established under an
473 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
475 struct nfs_server
*server
;
476 struct pnfs_layout_hdr
*lo
;
479 nfs4_deviceid_mark_client_invalid(clp
);
480 nfs4_deviceid_purge_client(clp
);
482 spin_lock(&clp
->cl_lock
);
484 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
485 if (!list_empty(&server
->layouts
))
486 list_splice_init(&server
->layouts
, &tmp_list
);
489 spin_unlock(&clp
->cl_lock
);
491 while (!list_empty(&tmp_list
)) {
492 lo
= list_entry(tmp_list
.next
, struct pnfs_layout_hdr
,
494 dprintk("%s freeing layout for inode %lu\n", __func__
,
495 lo
->plh_inode
->i_ino
);
496 list_del_init(&lo
->plh_layouts
);
497 pnfs_destroy_layout(NFS_I(lo
->plh_inode
));
501 /* update lo->plh_stateid with new if is more recent */
503 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
508 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
509 newseq
= be32_to_cpu(new->seqid
);
510 if ((int)(newseq
- oldseq
) > 0) {
511 nfs4_stateid_copy(&lo
->plh_stateid
, new);
512 if (update_barrier
) {
513 u32 new_barrier
= be32_to_cpu(new->seqid
);
515 if ((int)(new_barrier
- lo
->plh_barrier
))
516 lo
->plh_barrier
= new_barrier
;
518 /* Because of wraparound, we want to keep the barrier
519 * "close" to the current seqids. It needs to be
520 * within 2**31 to count as "behind", so if it
521 * gets too near that limit, give us a litle leeway
522 * and bring it to within 2**30.
523 * NOTE - and yes, this is all unsigned arithmetic.
525 if (unlikely((newseq
- lo
->plh_barrier
) > (3 << 29)))
526 lo
->plh_barrier
= newseq
- (1 << 30);
531 /* lget is set to 1 if called from inside send_layoutget call chain */
533 pnfs_layoutgets_blocked(struct pnfs_layout_hdr
*lo
, nfs4_stateid
*stateid
,
537 (int)(lo
->plh_barrier
- be32_to_cpu(stateid
->seqid
)) >= 0)
539 return lo
->plh_block_lgets
||
540 test_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
) ||
541 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
) ||
542 (list_empty(&lo
->plh_segs
) &&
543 (atomic_read(&lo
->plh_outstanding
) > lget
));
547 pnfs_choose_layoutget_stateid(nfs4_stateid
*dst
, struct pnfs_layout_hdr
*lo
,
548 struct nfs4_state
*open_state
)
552 dprintk("--> %s\n", __func__
);
553 spin_lock(&lo
->plh_inode
->i_lock
);
554 if (pnfs_layoutgets_blocked(lo
, NULL
, 1)) {
556 } else if (list_empty(&lo
->plh_segs
)) {
560 seq
= read_seqbegin(&open_state
->seqlock
);
561 nfs4_stateid_copy(dst
, &open_state
->stateid
);
562 } while (read_seqretry(&open_state
->seqlock
, seq
));
564 nfs4_stateid_copy(dst
, &lo
->plh_stateid
);
565 spin_unlock(&lo
->plh_inode
->i_lock
);
566 dprintk("<-- %s\n", __func__
);
571 * Get layout from server.
572 * for now, assume that whole file layouts are requested.
574 * arg->length: all ones
576 static struct pnfs_layout_segment
*
577 send_layoutget(struct pnfs_layout_hdr
*lo
,
578 struct nfs_open_context
*ctx
,
579 struct pnfs_layout_range
*range
,
582 struct inode
*ino
= lo
->plh_inode
;
583 struct nfs_server
*server
= NFS_SERVER(ino
);
584 struct nfs4_layoutget
*lgp
;
585 struct pnfs_layout_segment
*lseg
= NULL
;
586 struct page
**pages
= NULL
;
588 u32 max_resp_sz
, max_pages
;
590 dprintk("--> %s\n", __func__
);
593 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
597 /* allocate pages for xdr post processing */
598 max_resp_sz
= server
->nfs_client
->cl_session
->fc_attrs
.max_resp_sz
;
599 max_pages
= nfs_page_array_len(0, max_resp_sz
);
601 pages
= kcalloc(max_pages
, sizeof(struct page
*), gfp_flags
);
605 for (i
= 0; i
< max_pages
; i
++) {
606 pages
[i
] = alloc_page(gfp_flags
);
611 lgp
->args
.minlength
= PAGE_CACHE_SIZE
;
612 if (lgp
->args
.minlength
> range
->length
)
613 lgp
->args
.minlength
= range
->length
;
614 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
615 lgp
->args
.range
= *range
;
616 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
617 lgp
->args
.inode
= ino
;
618 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
619 lgp
->args
.layout
.pages
= pages
;
620 lgp
->args
.layout
.pglen
= max_pages
* PAGE_SIZE
;
622 lgp
->gfp_flags
= gfp_flags
;
624 /* Synchronously retrieve layout information from server and
627 nfs4_proc_layoutget(lgp
);
629 /* remember that LAYOUTGET failed and suspend trying */
630 set_bit(lo_fail_bit(range
->iomode
), &lo
->plh_flags
);
634 for (i
= 0; i
< max_pages
; i
++)
635 __free_page(pages
[i
]);
641 /* free any allocated xdr pages, lgp as it's not used */
643 for (i
= 0; i
< max_pages
; i
++) {
646 __free_page(pages
[i
]);
654 /* Initiates a LAYOUTRETURN(FILE) */
656 _pnfs_return_layout(struct inode
*ino
)
658 struct pnfs_layout_hdr
*lo
= NULL
;
659 struct nfs_inode
*nfsi
= NFS_I(ino
);
661 struct nfs4_layoutreturn
*lrp
;
662 nfs4_stateid stateid
;
665 dprintk("--> %s\n", __func__
);
667 spin_lock(&ino
->i_lock
);
670 spin_unlock(&ino
->i_lock
);
671 dprintk("%s: no layout to return\n", __func__
);
674 stateid
= nfsi
->layout
->plh_stateid
;
675 /* Reference matched in nfs4_layoutreturn_release */
677 mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
678 lo
->plh_block_lgets
++;
679 spin_unlock(&ino
->i_lock
);
680 pnfs_free_lseg_list(&tmp_list
);
682 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
));
684 lrp
= kzalloc(sizeof(*lrp
), GFP_KERNEL
);
685 if (unlikely(lrp
== NULL
)) {
687 set_bit(NFS_LAYOUT_RW_FAILED
, &lo
->plh_flags
);
688 set_bit(NFS_LAYOUT_RO_FAILED
, &lo
->plh_flags
);
693 lrp
->args
.stateid
= stateid
;
694 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
695 lrp
->args
.inode
= ino
;
696 lrp
->args
.layout
= lo
;
697 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
699 status
= nfs4_proc_layoutreturn(lrp
);
701 dprintk("<-- %s status: %d\n", __func__
, status
);
704 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
706 bool pnfs_roc(struct inode
*ino
)
708 struct pnfs_layout_hdr
*lo
;
709 struct pnfs_layout_segment
*lseg
, *tmp
;
713 spin_lock(&ino
->i_lock
);
714 lo
= NFS_I(ino
)->layout
;
715 if (!lo
|| !test_and_clear_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
) ||
716 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
718 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
719 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
720 mark_lseg_invalid(lseg
, &tmp_list
);
725 lo
->plh_block_lgets
++;
726 get_layout_hdr(lo
); /* matched in pnfs_roc_release */
727 spin_unlock(&ino
->i_lock
);
728 pnfs_free_lseg_list(&tmp_list
);
732 spin_unlock(&ino
->i_lock
);
736 void pnfs_roc_release(struct inode
*ino
)
738 struct pnfs_layout_hdr
*lo
;
740 spin_lock(&ino
->i_lock
);
741 lo
= NFS_I(ino
)->layout
;
742 lo
->plh_block_lgets
--;
743 put_layout_hdr_locked(lo
);
744 spin_unlock(&ino
->i_lock
);
747 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
749 struct pnfs_layout_hdr
*lo
;
751 spin_lock(&ino
->i_lock
);
752 lo
= NFS_I(ino
)->layout
;
753 if ((int)(barrier
- lo
->plh_barrier
) > 0)
754 lo
->plh_barrier
= barrier
;
755 spin_unlock(&ino
->i_lock
);
758 bool pnfs_roc_drain(struct inode
*ino
, u32
*barrier
)
760 struct nfs_inode
*nfsi
= NFS_I(ino
);
761 struct pnfs_layout_segment
*lseg
;
764 spin_lock(&ino
->i_lock
);
765 list_for_each_entry(lseg
, &nfsi
->layout
->plh_segs
, pls_list
)
766 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
771 struct pnfs_layout_hdr
*lo
= nfsi
->layout
;
772 u32 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
774 /* Since close does not return a layout stateid for use as
775 * a barrier, we choose the worst-case barrier.
777 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
779 spin_unlock(&ino
->i_lock
);
784 * Compare two layout segments for sorting into layout cache.
785 * We want to preferentially return RW over RO layouts, so ensure those
789 cmp_layout(struct pnfs_layout_range
*l1
,
790 struct pnfs_layout_range
*l2
)
794 /* high offset > low offset */
795 d
= l1
->offset
- l2
->offset
;
799 /* short length > long length */
800 d
= l2
->length
- l1
->length
;
804 /* read > read/write */
805 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
809 pnfs_insert_layout(struct pnfs_layout_hdr
*lo
,
810 struct pnfs_layout_segment
*lseg
)
812 struct pnfs_layout_segment
*lp
;
814 dprintk("%s:Begin\n", __func__
);
816 assert_spin_locked(&lo
->plh_inode
->i_lock
);
817 list_for_each_entry(lp
, &lo
->plh_segs
, pls_list
) {
818 if (cmp_layout(&lseg
->pls_range
, &lp
->pls_range
) > 0)
820 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
821 dprintk("%s: inserted lseg %p "
822 "iomode %d offset %llu length %llu before "
823 "lp %p iomode %d offset %llu length %llu\n",
824 __func__
, lseg
, lseg
->pls_range
.iomode
,
825 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
826 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
827 lp
->pls_range
.length
);
830 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
831 dprintk("%s: inserted lseg %p "
832 "iomode %d offset %llu length %llu at tail\n",
833 __func__
, lseg
, lseg
->pls_range
.iomode
,
834 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
838 dprintk("%s:Return\n", __func__
);
841 static struct pnfs_layout_hdr
*
842 alloc_init_layout_hdr(struct inode
*ino
,
843 struct nfs_open_context
*ctx
,
846 struct pnfs_layout_hdr
*lo
;
848 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
851 atomic_set(&lo
->plh_refcount
, 1);
852 INIT_LIST_HEAD(&lo
->plh_layouts
);
853 INIT_LIST_HEAD(&lo
->plh_segs
);
854 INIT_LIST_HEAD(&lo
->plh_bulk_recall
);
856 lo
->plh_lc_cred
= get_rpccred(ctx
->state
->owner
->so_cred
);
860 static struct pnfs_layout_hdr
*
861 pnfs_find_alloc_layout(struct inode
*ino
,
862 struct nfs_open_context
*ctx
,
865 struct nfs_inode
*nfsi
= NFS_I(ino
);
866 struct pnfs_layout_hdr
*new = NULL
;
868 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
870 assert_spin_locked(&ino
->i_lock
);
872 if (test_bit(NFS_LAYOUT_DESTROYED
, &nfsi
->layout
->plh_flags
))
877 spin_unlock(&ino
->i_lock
);
878 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
879 spin_lock(&ino
->i_lock
);
881 if (likely(nfsi
->layout
== NULL
)) /* Won the race? */
884 pnfs_free_layout_hdr(new);
889 * iomode matching rules:
900 is_matching_lseg(struct pnfs_layout_range
*ls_range
,
901 struct pnfs_layout_range
*range
)
903 struct pnfs_layout_range range1
;
905 if ((range
->iomode
== IOMODE_RW
&&
906 ls_range
->iomode
!= IOMODE_RW
) ||
907 !lo_seg_intersecting(ls_range
, range
))
910 /* range1 covers only the first byte in the range */
913 return lo_seg_contained(ls_range
, &range1
);
917 * lookup range in layout
919 static struct pnfs_layout_segment
*
920 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
921 struct pnfs_layout_range
*range
)
923 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
925 dprintk("%s:Begin\n", __func__
);
927 assert_spin_locked(&lo
->plh_inode
->i_lock
);
928 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
929 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
930 is_matching_lseg(&lseg
->pls_range
, range
)) {
931 ret
= get_lseg(lseg
);
934 if (lseg
->pls_range
.offset
> range
->offset
)
938 dprintk("%s:Return lseg %p ref %d\n",
939 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
944 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
945 * to the MDS or over pNFS
947 * The nfs_inode read_io and write_io fields are cumulative counters reset
948 * when there are no layout segments. Note that in pnfs_update_layout iomode
949 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
952 * A return of true means use MDS I/O.
955 * If a file's size is smaller than the file size threshold, data accesses
956 * SHOULD be sent to the metadata server. If an I/O request has a length that
957 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
958 * server. If both file size and I/O size are provided, the client SHOULD
959 * reach or exceed both thresholds before sending its read or write
960 * requests to the data server.
962 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
963 struct inode
*ino
, int iomode
)
965 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
966 struct nfs_inode
*nfsi
= NFS_I(ino
);
967 loff_t fsize
= i_size_read(ino
);
968 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
973 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
974 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
978 if (t
->bm
& THRESHOLD_RD
) {
979 dprintk("%s fsize %llu\n", __func__
, fsize
);
981 if (fsize
< t
->rd_sz
)
984 if (t
->bm
& THRESHOLD_RD_IO
) {
985 dprintk("%s nfsi->read_io %llu\n", __func__
,
988 if (nfsi
->read_io
< t
->rd_io_sz
)
993 if (t
->bm
& THRESHOLD_WR
) {
994 dprintk("%s fsize %llu\n", __func__
, fsize
);
996 if (fsize
< t
->wr_sz
)
999 if (t
->bm
& THRESHOLD_WR_IO
) {
1000 dprintk("%s nfsi->write_io %llu\n", __func__
,
1003 if (nfsi
->write_io
< t
->wr_io_sz
)
1008 if (size_set
&& io_set
) {
1011 } else if (size
|| io
)
1014 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1019 * Layout segment is retreived from the server if not cached.
1020 * The appropriate layout segment is referenced and returned to the caller.
1022 struct pnfs_layout_segment
*
1023 pnfs_update_layout(struct inode
*ino
,
1024 struct nfs_open_context
*ctx
,
1027 enum pnfs_iomode iomode
,
1030 struct pnfs_layout_range arg
= {
1036 struct nfs_inode
*nfsi
= NFS_I(ino
);
1037 struct nfs_server
*server
= NFS_SERVER(ino
);
1038 struct nfs_client
*clp
= server
->nfs_client
;
1039 struct pnfs_layout_hdr
*lo
;
1040 struct pnfs_layout_segment
*lseg
= NULL
;
1043 if (!pnfs_enabled_sb(NFS_SERVER(ino
)))
1046 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
))
1049 spin_lock(&ino
->i_lock
);
1050 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1052 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__
);
1056 /* Do we even need to bother with this? */
1057 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1058 dprintk("%s matches recall, use MDS\n", __func__
);
1062 /* if LAYOUTGET already failed once we don't try again */
1063 if (test_bit(lo_fail_bit(iomode
), &nfsi
->layout
->plh_flags
))
1066 /* Check to see if the layout for the given range already exists */
1067 lseg
= pnfs_find_lseg(lo
, &arg
);
1071 if (pnfs_layoutgets_blocked(lo
, NULL
, 0))
1073 atomic_inc(&lo
->plh_outstanding
);
1076 if (list_empty(&lo
->plh_segs
))
1078 spin_unlock(&ino
->i_lock
);
1080 /* The lo must be on the clp list if there is any
1081 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1083 spin_lock(&clp
->cl_lock
);
1084 BUG_ON(!list_empty(&lo
->plh_layouts
));
1085 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1086 spin_unlock(&clp
->cl_lock
);
1089 pg_offset
= arg
.offset
& ~PAGE_CACHE_MASK
;
1091 arg
.offset
-= pg_offset
;
1092 arg
.length
+= pg_offset
;
1094 if (arg
.length
!= NFS4_MAX_UINT64
)
1095 arg
.length
= PAGE_CACHE_ALIGN(arg
.length
);
1097 lseg
= send_layoutget(lo
, ctx
, &arg
, gfp_flags
);
1098 if (!lseg
&& first
) {
1099 spin_lock(&clp
->cl_lock
);
1100 list_del_init(&lo
->plh_layouts
);
1101 spin_unlock(&clp
->cl_lock
);
1103 atomic_dec(&lo
->plh_outstanding
);
1106 dprintk("%s end, state 0x%lx lseg %p\n", __func__
,
1107 nfsi
->layout
? nfsi
->layout
->plh_flags
: -1, lseg
);
1110 spin_unlock(&ino
->i_lock
);
1113 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1116 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1118 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1119 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1120 struct pnfs_layout_segment
*lseg
;
1121 struct inode
*ino
= lo
->plh_inode
;
1124 /* Inject layout blob into I/O device driver */
1125 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1126 if (!lseg
|| IS_ERR(lseg
)) {
1130 status
= PTR_ERR(lseg
);
1131 dprintk("%s: Could not allocate layout: error %d\n",
1136 spin_lock(&ino
->i_lock
);
1137 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1138 dprintk("%s forget reply due to recall\n", __func__
);
1139 goto out_forget_reply
;
1142 if (pnfs_layoutgets_blocked(lo
, &res
->stateid
, 1)) {
1143 dprintk("%s forget reply due to state\n", __func__
);
1144 goto out_forget_reply
;
1146 init_lseg(lo
, lseg
);
1147 lseg
->pls_range
= res
->range
;
1148 *lgp
->lsegpp
= get_lseg(lseg
);
1149 pnfs_insert_layout(lo
, lseg
);
1151 if (res
->return_on_close
) {
1152 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1153 set_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
);
1156 /* Done processing layoutget. Set the layout stateid */
1157 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1158 spin_unlock(&ino
->i_lock
);
1163 spin_unlock(&ino
->i_lock
);
1164 lseg
->pls_layout
= lo
;
1165 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1170 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1172 BUG_ON(pgio
->pg_lseg
!= NULL
);
1174 if (req
->wb_offset
!= req
->wb_pgbase
) {
1175 nfs_pageio_reset_read_mds(pgio
);
1178 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1184 /* If no lseg, fall back to read through mds */
1185 if (pgio
->pg_lseg
== NULL
)
1186 nfs_pageio_reset_read_mds(pgio
);
1189 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1192 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1194 BUG_ON(pgio
->pg_lseg
!= NULL
);
1196 if (req
->wb_offset
!= req
->wb_pgbase
) {
1197 nfs_pageio_reset_write_mds(pgio
);
1200 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1206 /* If no lseg, fall back to write through mds */
1207 if (pgio
->pg_lseg
== NULL
)
1208 nfs_pageio_reset_write_mds(pgio
);
1210 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1213 pnfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1214 const struct nfs_pgio_completion_ops
*compl_ops
)
1216 struct nfs_server
*server
= NFS_SERVER(inode
);
1217 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1221 nfs_pageio_init(pgio
, inode
, ld
->pg_read_ops
, compl_ops
,
1227 pnfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1229 const struct nfs_pgio_completion_ops
*compl_ops
)
1231 struct nfs_server
*server
= NFS_SERVER(inode
);
1232 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1236 nfs_pageio_init(pgio
, inode
, ld
->pg_write_ops
, compl_ops
,
1237 server
->wsize
, ioflags
);
1242 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
1243 struct nfs_page
*req
)
1245 if (pgio
->pg_lseg
== NULL
)
1246 return nfs_generic_pg_test(pgio
, prev
, req
);
1249 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1250 * Note that this test makes several assumptions:
1251 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1252 * is known to lie within the range.
1253 * - that the nfs_page being tested is known to be contiguous with the
1254 * previous nfs_page.
1255 * - Layout ranges are page aligned, so we only have to test the
1256 * start offset of the request.
1258 * Please also note that 'end_offset' is actually the offset of the
1259 * first byte that lies outside the pnfs_layout_range. FIXME?
1262 return req_offset(req
) < end_offset(pgio
->pg_lseg
->pls_range
.offset
,
1263 pgio
->pg_lseg
->pls_range
.length
);
1265 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
1267 int pnfs_write_done_resend_to_mds(struct inode
*inode
,
1268 struct list_head
*head
,
1269 const struct nfs_pgio_completion_ops
*compl_ops
)
1271 struct nfs_pageio_descriptor pgio
;
1274 /* Resend all requests through the MDS */
1275 nfs_pageio_init_write_mds(&pgio
, inode
, FLUSH_STABLE
, compl_ops
);
1276 while (!list_empty(head
)) {
1277 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1279 nfs_list_remove_request(req
);
1280 if (!nfs_pageio_add_request(&pgio
, req
))
1281 nfs_list_add_request(req
, &failed
);
1283 nfs_pageio_complete(&pgio
);
1285 if (!list_empty(&failed
)) {
1286 /* For some reason our attempt to resend pages. Mark the
1287 * overall send request as having failed, and let
1288 * nfs_writeback_release_full deal with the error.
1290 list_move(&failed
, head
);
1295 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
1297 static void pnfs_ld_handle_write_error(struct nfs_write_data
*data
)
1299 struct nfs_pgio_header
*hdr
= data
->header
;
1301 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
1302 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1303 PNFS_LAYOUTRET_ON_ERROR
) {
1304 clear_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(hdr
->inode
)->flags
);
1305 pnfs_return_layout(hdr
->inode
);
1307 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1308 data
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
->inode
,
1310 hdr
->completion_ops
);
1314 * Called by non rpc-based layout drivers
1316 void pnfs_ld_write_done(struct nfs_write_data
*data
)
1318 struct nfs_pgio_header
*hdr
= data
->header
;
1320 if (!hdr
->pnfs_error
) {
1321 pnfs_set_layoutcommit(data
);
1322 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1324 pnfs_ld_handle_write_error(data
);
1325 hdr
->mds_ops
->rpc_release(data
);
1327 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
1330 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
1331 struct nfs_write_data
*data
)
1333 struct nfs_pgio_header
*hdr
= data
->header
;
1335 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1336 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1337 nfs_pageio_reset_write_mds(desc
);
1338 desc
->pg_recoalesce
= 1;
1340 nfs_writedata_release(data
);
1343 static enum pnfs_try_status
1344 pnfs_try_to_write_data(struct nfs_write_data
*wdata
,
1345 const struct rpc_call_ops
*call_ops
,
1346 struct pnfs_layout_segment
*lseg
,
1349 struct nfs_pgio_header
*hdr
= wdata
->header
;
1350 struct inode
*inode
= hdr
->inode
;
1351 enum pnfs_try_status trypnfs
;
1352 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1354 hdr
->mds_ops
= call_ops
;
1356 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
1357 inode
->i_ino
, wdata
->args
.count
, wdata
->args
.offset
, how
);
1358 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(wdata
, how
);
1359 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1360 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
1361 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1366 pnfs_do_multiple_writes(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
, int how
)
1368 struct nfs_write_data
*data
;
1369 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1370 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1372 desc
->pg_lseg
= NULL
;
1373 while (!list_empty(head
)) {
1374 enum pnfs_try_status trypnfs
;
1376 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1377 list_del_init(&data
->list
);
1379 trypnfs
= pnfs_try_to_write_data(data
, call_ops
, lseg
, how
);
1380 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1381 pnfs_write_through_mds(desc
, data
);
1386 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
1388 put_lseg(hdr
->lseg
);
1389 nfs_writehdr_free(hdr
);
1393 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1395 struct nfs_write_header
*whdr
;
1396 struct nfs_pgio_header
*hdr
;
1399 whdr
= nfs_writehdr_alloc();
1401 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1402 put_lseg(desc
->pg_lseg
);
1403 desc
->pg_lseg
= NULL
;
1406 hdr
= &whdr
->header
;
1407 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
1408 hdr
->lseg
= get_lseg(desc
->pg_lseg
);
1409 atomic_inc(&hdr
->refcnt
);
1410 ret
= nfs_generic_flush(desc
, hdr
);
1412 put_lseg(desc
->pg_lseg
);
1413 desc
->pg_lseg
= NULL
;
1415 pnfs_do_multiple_writes(desc
, &hdr
->rpc_list
, desc
->pg_ioflags
);
1416 if (atomic_dec_and_test(&hdr
->refcnt
))
1417 hdr
->completion_ops
->completion(hdr
);
1420 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
1422 int pnfs_read_done_resend_to_mds(struct inode
*inode
,
1423 struct list_head
*head
,
1424 const struct nfs_pgio_completion_ops
*compl_ops
)
1426 struct nfs_pageio_descriptor pgio
;
1429 /* Resend all requests through the MDS */
1430 nfs_pageio_init_read_mds(&pgio
, inode
, compl_ops
);
1431 while (!list_empty(head
)) {
1432 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1434 nfs_list_remove_request(req
);
1435 if (!nfs_pageio_add_request(&pgio
, req
))
1436 nfs_list_add_request(req
, &failed
);
1438 nfs_pageio_complete(&pgio
);
1440 if (!list_empty(&failed
)) {
1441 list_move(&failed
, head
);
1446 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
1448 static void pnfs_ld_handle_read_error(struct nfs_read_data
*data
)
1450 struct nfs_pgio_header
*hdr
= data
->header
;
1452 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
1453 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1454 PNFS_LAYOUTRET_ON_ERROR
) {
1455 clear_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(hdr
->inode
)->flags
);
1456 pnfs_return_layout(hdr
->inode
);
1458 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1459 data
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
->inode
,
1461 hdr
->completion_ops
);
1465 * Called by non rpc-based layout drivers
1467 void pnfs_ld_read_done(struct nfs_read_data
*data
)
1469 struct nfs_pgio_header
*hdr
= data
->header
;
1471 if (likely(!hdr
->pnfs_error
)) {
1472 __nfs4_read_done_cb(data
);
1473 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1475 pnfs_ld_handle_read_error(data
);
1476 hdr
->mds_ops
->rpc_release(data
);
1478 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
1481 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
1482 struct nfs_read_data
*data
)
1484 struct nfs_pgio_header
*hdr
= data
->header
;
1486 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1487 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1488 nfs_pageio_reset_read_mds(desc
);
1489 desc
->pg_recoalesce
= 1;
1491 nfs_readdata_release(data
);
1495 * Call the appropriate parallel I/O subsystem read function.
1497 static enum pnfs_try_status
1498 pnfs_try_to_read_data(struct nfs_read_data
*rdata
,
1499 const struct rpc_call_ops
*call_ops
,
1500 struct pnfs_layout_segment
*lseg
)
1502 struct nfs_pgio_header
*hdr
= rdata
->header
;
1503 struct inode
*inode
= hdr
->inode
;
1504 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1505 enum pnfs_try_status trypnfs
;
1507 hdr
->mds_ops
= call_ops
;
1509 dprintk("%s: Reading ino:%lu %u@%llu\n",
1510 __func__
, inode
->i_ino
, rdata
->args
.count
, rdata
->args
.offset
);
1512 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(rdata
);
1513 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1514 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
1515 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1520 pnfs_do_multiple_reads(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
)
1522 struct nfs_read_data
*data
;
1523 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1524 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1526 desc
->pg_lseg
= NULL
;
1527 while (!list_empty(head
)) {
1528 enum pnfs_try_status trypnfs
;
1530 data
= list_first_entry(head
, struct nfs_read_data
, list
);
1531 list_del_init(&data
->list
);
1533 trypnfs
= pnfs_try_to_read_data(data
, call_ops
, lseg
);
1534 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1535 pnfs_read_through_mds(desc
, data
);
1540 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
1542 put_lseg(hdr
->lseg
);
1543 nfs_readhdr_free(hdr
);
1547 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
1549 struct nfs_read_header
*rhdr
;
1550 struct nfs_pgio_header
*hdr
;
1553 rhdr
= nfs_readhdr_alloc();
1555 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1557 put_lseg(desc
->pg_lseg
);
1558 desc
->pg_lseg
= NULL
;
1561 hdr
= &rhdr
->header
;
1562 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
1563 hdr
->lseg
= get_lseg(desc
->pg_lseg
);
1564 atomic_inc(&hdr
->refcnt
);
1565 ret
= nfs_generic_pagein(desc
, hdr
);
1567 put_lseg(desc
->pg_lseg
);
1568 desc
->pg_lseg
= NULL
;
1570 pnfs_do_multiple_reads(desc
, &hdr
->rpc_list
);
1571 if (atomic_dec_and_test(&hdr
->refcnt
))
1572 hdr
->completion_ops
->completion(hdr
);
1575 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
1578 * There can be multiple RW segments.
1580 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
1582 struct pnfs_layout_segment
*lseg
;
1584 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
1585 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
1586 test_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1587 list_add(&lseg
->pls_lc_list
, listp
);
1591 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
1593 if (lseg
->pls_range
.iomode
== IOMODE_RW
) {
1594 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__
);
1595 set_bit(lo_fail_bit(IOMODE_RW
), &lseg
->pls_layout
->plh_flags
);
1597 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__
);
1598 set_bit(lo_fail_bit(IOMODE_READ
), &lseg
->pls_layout
->plh_flags
);
1601 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
1604 pnfs_set_layoutcommit(struct nfs_write_data
*wdata
)
1606 struct nfs_pgio_header
*hdr
= wdata
->header
;
1607 struct inode
*inode
= hdr
->inode
;
1608 struct nfs_inode
*nfsi
= NFS_I(inode
);
1609 loff_t end_pos
= wdata
->mds_offset
+ wdata
->res
.count
;
1610 bool mark_as_dirty
= false;
1612 spin_lock(&inode
->i_lock
);
1613 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1614 mark_as_dirty
= true;
1615 dprintk("%s: Set layoutcommit for inode %lu ",
1616 __func__
, inode
->i_ino
);
1618 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &hdr
->lseg
->pls_flags
)) {
1619 /* references matched in nfs4_layoutcommit_release */
1620 get_lseg(hdr
->lseg
);
1622 if (end_pos
> nfsi
->layout
->plh_lwb
)
1623 nfsi
->layout
->plh_lwb
= end_pos
;
1624 spin_unlock(&inode
->i_lock
);
1625 dprintk("%s: lseg %p end_pos %llu\n",
1626 __func__
, hdr
->lseg
, nfsi
->layout
->plh_lwb
);
1628 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1629 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1631 mark_inode_dirty_sync(inode
);
1633 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
1635 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
1637 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
1639 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
1640 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
1644 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1645 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1646 * data to disk to allow the server to recover the data if it crashes.
1647 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1648 * is off, and a COMMIT is sent to a data server, or
1649 * if WRITEs to a data server return NFS_DATA_SYNC.
1652 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
1654 struct nfs4_layoutcommit_data
*data
;
1655 struct nfs_inode
*nfsi
= NFS_I(inode
);
1659 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
1661 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1664 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1665 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
1671 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1674 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
1679 status
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
,
1680 nfs_wait_bit_killable
, TASK_KILLABLE
);
1685 INIT_LIST_HEAD(&data
->lseg_list
);
1686 spin_lock(&inode
->i_lock
);
1687 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1688 clear_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
);
1689 spin_unlock(&inode
->i_lock
);
1690 wake_up_bit(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
);
1694 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
1696 end_pos
= nfsi
->layout
->plh_lwb
;
1697 nfsi
->layout
->plh_lwb
= 0;
1699 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
1700 spin_unlock(&inode
->i_lock
);
1702 data
->args
.inode
= inode
;
1703 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
1704 nfs_fattr_init(&data
->fattr
);
1705 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
1706 data
->res
.fattr
= &data
->fattr
;
1707 data
->args
.lastbytewritten
= end_pos
- 1;
1708 data
->res
.server
= NFS_SERVER(inode
);
1710 status
= nfs4_proc_layoutcommit(data
, sync
);
1713 mark_inode_dirty_sync(inode
);
1714 dprintk("<-- %s status %d\n", __func__
, status
);
1721 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
1723 struct nfs4_threshold
*thp
;
1725 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
1727 dprintk("%s mdsthreshold allocation failed\n", __func__
);