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>
35 #define NFSDBG_FACILITY NFSDBG_PNFS
40 * protects pnfs_modules_tbl.
42 static DEFINE_SPINLOCK(pnfs_spinlock
);
45 * pnfs_modules_tbl holds all pnfs modules
47 static LIST_HEAD(pnfs_modules_tbl
);
49 /* Return the registered pnfs layout driver module matching given id */
50 static struct pnfs_layoutdriver_type
*
51 find_pnfs_driver_locked(u32 id
)
53 struct pnfs_layoutdriver_type
*local
;
55 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
60 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
64 static struct pnfs_layoutdriver_type
*
65 find_pnfs_driver(u32 id
)
67 struct pnfs_layoutdriver_type
*local
;
69 spin_lock(&pnfs_spinlock
);
70 local
= find_pnfs_driver_locked(id
);
71 spin_unlock(&pnfs_spinlock
);
76 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
78 if (nfss
->pnfs_curr_ld
)
79 module_put(nfss
->pnfs_curr_ld
->owner
);
80 nfss
->pnfs_curr_ld
= NULL
;
84 * Try to set the server's pnfs module to the pnfs layout type specified by id.
85 * Currently only one pNFS layout driver per filesystem is supported.
87 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
90 set_pnfs_layoutdriver(struct nfs_server
*server
, u32 id
)
92 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
96 if (!(server
->nfs_client
->cl_exchange_flags
&
97 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
98 printk(KERN_ERR
"%s: id %u cl_exchange_flags 0x%x\n", __func__
,
99 id
, server
->nfs_client
->cl_exchange_flags
);
102 ld_type
= find_pnfs_driver(id
);
104 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
, id
);
105 ld_type
= find_pnfs_driver(id
);
107 dprintk("%s: No pNFS module found for %u.\n",
112 if (!try_module_get(ld_type
->owner
)) {
113 dprintk("%s: Could not grab reference on module\n", __func__
);
116 server
->pnfs_curr_ld
= ld_type
;
118 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
122 dprintk("%s: Using NFSv4 I/O\n", __func__
);
123 server
->pnfs_curr_ld
= NULL
;
127 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
129 int status
= -EINVAL
;
130 struct pnfs_layoutdriver_type
*tmp
;
132 if (ld_type
->id
== 0) {
133 printk(KERN_ERR
"%s id 0 is reserved\n", __func__
);
136 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
137 printk(KERN_ERR
"%s Layout driver must provide "
138 "alloc_lseg and free_lseg.\n", __func__
);
142 spin_lock(&pnfs_spinlock
);
143 tmp
= find_pnfs_driver_locked(ld_type
->id
);
145 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
147 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
150 printk(KERN_ERR
"%s Module with id %d already loaded!\n",
151 __func__
, ld_type
->id
);
153 spin_unlock(&pnfs_spinlock
);
157 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
160 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
162 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
163 spin_lock(&pnfs_spinlock
);
164 list_del(&ld_type
->pnfs_tblid
);
165 spin_unlock(&pnfs_spinlock
);
167 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
170 * pNFS client layout cache
173 /* Need to hold i_lock if caller does not already hold reference */
175 get_layout_hdr(struct pnfs_layout_hdr
*lo
)
177 atomic_inc(&lo
->plh_refcount
);
181 destroy_layout_hdr(struct pnfs_layout_hdr
*lo
)
183 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
184 BUG_ON(!list_empty(&lo
->plh_layouts
));
185 NFS_I(lo
->plh_inode
)->layout
= NULL
;
190 put_layout_hdr_locked(struct pnfs_layout_hdr
*lo
)
192 if (atomic_dec_and_test(&lo
->plh_refcount
))
193 destroy_layout_hdr(lo
);
197 put_layout_hdr(struct pnfs_layout_hdr
*lo
)
199 struct inode
*inode
= lo
->plh_inode
;
201 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
202 destroy_layout_hdr(lo
);
203 spin_unlock(&inode
->i_lock
);
208 init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
)
210 INIT_LIST_HEAD(&lseg
->pls_list
);
211 atomic_set(&lseg
->pls_refcount
, 1);
213 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
214 lseg
->pls_layout
= lo
;
217 static void free_lseg(struct pnfs_layout_segment
*lseg
)
219 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
221 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
222 /* Matched by get_layout_hdr in pnfs_insert_layout */
223 put_layout_hdr(NFS_I(ino
)->layout
);
227 put_lseg_common(struct pnfs_layout_segment
*lseg
)
229 struct inode
*inode
= lseg
->pls_layout
->plh_inode
;
231 BUG_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
232 list_del_init(&lseg
->pls_list
);
233 if (list_empty(&lseg
->pls_layout
->plh_segs
)) {
234 set_bit(NFS_LAYOUT_DESTROYED
, &lseg
->pls_layout
->plh_flags
);
235 /* Matched by initial refcount set in alloc_init_layout_hdr */
236 put_layout_hdr_locked(lseg
->pls_layout
);
238 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
242 put_lseg(struct pnfs_layout_segment
*lseg
)
249 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
250 atomic_read(&lseg
->pls_refcount
),
251 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
252 inode
= lseg
->pls_layout
->plh_inode
;
253 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
256 put_lseg_common(lseg
);
257 list_add(&lseg
->pls_list
, &free_me
);
258 spin_unlock(&inode
->i_lock
);
259 pnfs_free_lseg_list(&free_me
);
262 EXPORT_SYMBOL_GPL(put_lseg
);
265 should_free_lseg(u32 lseg_iomode
, u32 recall_iomode
)
267 return (recall_iomode
== IOMODE_ANY
||
268 lseg_iomode
== recall_iomode
);
271 /* Returns 1 if lseg is removed from list, 0 otherwise */
272 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
273 struct list_head
*tmp_list
)
277 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
278 /* Remove the reference keeping the lseg in the
279 * list. It will now be removed when all
280 * outstanding io is finished.
282 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
283 atomic_read(&lseg
->pls_refcount
));
284 if (atomic_dec_and_test(&lseg
->pls_refcount
)) {
285 put_lseg_common(lseg
);
286 list_add(&lseg
->pls_list
, tmp_list
);
293 /* Returns count of number of matching invalid lsegs remaining in list
297 mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
298 struct list_head
*tmp_list
,
301 struct pnfs_layout_segment
*lseg
, *next
;
302 int invalid
= 0, removed
= 0;
304 dprintk("%s:Begin lo %p\n", __func__
, lo
);
306 if (list_empty(&lo
->plh_segs
)) {
307 if (!test_and_set_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
))
308 put_layout_hdr_locked(lo
);
311 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
312 if (should_free_lseg(lseg
->pls_range
.iomode
, iomode
)) {
313 dprintk("%s: freeing lseg %p iomode %d "
314 "offset %llu length %llu\n", __func__
,
315 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_range
.offset
,
316 lseg
->pls_range
.length
);
318 removed
+= mark_lseg_invalid(lseg
, tmp_list
);
320 dprintk("%s:Return %i\n", __func__
, invalid
- removed
);
321 return invalid
- removed
;
324 /* note free_me must contain lsegs from a single layout_hdr */
326 pnfs_free_lseg_list(struct list_head
*free_me
)
328 struct pnfs_layout_segment
*lseg
, *tmp
;
329 struct pnfs_layout_hdr
*lo
;
331 if (list_empty(free_me
))
334 lo
= list_first_entry(free_me
, struct pnfs_layout_segment
,
335 pls_list
)->pls_layout
;
337 if (test_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
)) {
338 struct nfs_client
*clp
;
340 clp
= NFS_SERVER(lo
->plh_inode
)->nfs_client
;
341 spin_lock(&clp
->cl_lock
);
342 list_del_init(&lo
->plh_layouts
);
343 spin_unlock(&clp
->cl_lock
);
345 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
346 list_del(&lseg
->pls_list
);
352 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
354 struct pnfs_layout_hdr
*lo
;
357 spin_lock(&nfsi
->vfs_inode
.i_lock
);
360 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
361 mark_matching_lsegs_invalid(lo
, &tmp_list
, IOMODE_ANY
);
363 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
364 pnfs_free_lseg_list(&tmp_list
);
368 * Called by the state manger to remove all layouts established under an
372 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
374 struct pnfs_layout_hdr
*lo
;
377 spin_lock(&clp
->cl_lock
);
378 list_splice_init(&clp
->cl_layouts
, &tmp_list
);
379 spin_unlock(&clp
->cl_lock
);
381 while (!list_empty(&tmp_list
)) {
382 lo
= list_entry(tmp_list
.next
, struct pnfs_layout_hdr
,
384 dprintk("%s freeing layout for inode %lu\n", __func__
,
385 lo
->plh_inode
->i_ino
);
386 list_del_init(&lo
->plh_layouts
);
387 pnfs_destroy_layout(NFS_I(lo
->plh_inode
));
391 /* update lo->plh_stateid with new if is more recent */
393 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
398 oldseq
= be32_to_cpu(lo
->plh_stateid
.stateid
.seqid
);
399 newseq
= be32_to_cpu(new->stateid
.seqid
);
400 if ((int)(newseq
- oldseq
) > 0) {
401 memcpy(&lo
->plh_stateid
, &new->stateid
, sizeof(new->stateid
));
402 if (update_barrier
) {
403 u32 new_barrier
= be32_to_cpu(new->stateid
.seqid
);
405 if ((int)(new_barrier
- lo
->plh_barrier
))
406 lo
->plh_barrier
= new_barrier
;
408 /* Because of wraparound, we want to keep the barrier
409 * "close" to the current seqids. It needs to be
410 * within 2**31 to count as "behind", so if it
411 * gets too near that limit, give us a litle leeway
412 * and bring it to within 2**30.
413 * NOTE - and yes, this is all unsigned arithmetic.
415 if (unlikely((newseq
- lo
->plh_barrier
) > (3 << 29)))
416 lo
->plh_barrier
= newseq
- (1 << 30);
421 /* lget is set to 1 if called from inside send_layoutget call chain */
423 pnfs_layoutgets_blocked(struct pnfs_layout_hdr
*lo
, nfs4_stateid
*stateid
,
427 (int)(lo
->plh_barrier
- be32_to_cpu(stateid
->stateid
.seqid
)) >= 0)
429 return lo
->plh_block_lgets
||
430 test_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
) ||
431 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
) ||
432 (list_empty(&lo
->plh_segs
) &&
433 (atomic_read(&lo
->plh_outstanding
) > lget
));
437 pnfs_choose_layoutget_stateid(nfs4_stateid
*dst
, struct pnfs_layout_hdr
*lo
,
438 struct nfs4_state
*open_state
)
442 dprintk("--> %s\n", __func__
);
443 spin_lock(&lo
->plh_inode
->i_lock
);
444 if (pnfs_layoutgets_blocked(lo
, NULL
, 1)) {
446 } else if (list_empty(&lo
->plh_segs
)) {
450 seq
= read_seqbegin(&open_state
->seqlock
);
451 memcpy(dst
->data
, open_state
->stateid
.data
,
452 sizeof(open_state
->stateid
.data
));
453 } while (read_seqretry(&open_state
->seqlock
, seq
));
455 memcpy(dst
->data
, lo
->plh_stateid
.data
, sizeof(lo
->plh_stateid
.data
));
456 spin_unlock(&lo
->plh_inode
->i_lock
);
457 dprintk("<-- %s\n", __func__
);
462 * Get layout from server.
463 * for now, assume that whole file layouts are requested.
465 * arg->length: all ones
467 static struct pnfs_layout_segment
*
468 send_layoutget(struct pnfs_layout_hdr
*lo
,
469 struct nfs_open_context
*ctx
,
473 struct inode
*ino
= lo
->plh_inode
;
474 struct nfs_server
*server
= NFS_SERVER(ino
);
475 struct nfs4_layoutget
*lgp
;
476 struct pnfs_layout_segment
*lseg
= NULL
;
477 struct page
**pages
= NULL
;
479 u32 max_resp_sz
, max_pages
;
481 dprintk("--> %s\n", __func__
);
484 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
488 /* allocate pages for xdr post processing */
489 max_resp_sz
= server
->nfs_client
->cl_session
->fc_attrs
.max_resp_sz
;
490 max_pages
= max_resp_sz
>> PAGE_SHIFT
;
492 pages
= kzalloc(max_pages
* sizeof(struct page
*), gfp_flags
);
496 for (i
= 0; i
< max_pages
; i
++) {
497 pages
[i
] = alloc_page(gfp_flags
);
502 lgp
->args
.minlength
= NFS4_MAX_UINT64
;
503 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
504 lgp
->args
.range
.iomode
= iomode
;
505 lgp
->args
.range
.offset
= 0;
506 lgp
->args
.range
.length
= NFS4_MAX_UINT64
;
507 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
508 lgp
->args
.inode
= ino
;
509 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
510 lgp
->args
.layout
.pages
= pages
;
511 lgp
->args
.layout
.pglen
= max_pages
* PAGE_SIZE
;
513 lgp
->gfp_flags
= gfp_flags
;
515 /* Synchronously retrieve layout information from server and
518 nfs4_proc_layoutget(lgp
);
520 /* remember that LAYOUTGET failed and suspend trying */
521 set_bit(lo_fail_bit(iomode
), &lo
->plh_flags
);
525 for (i
= 0; i
< max_pages
; i
++)
526 __free_page(pages
[i
]);
532 /* free any allocated xdr pages, lgp as it's not used */
534 for (i
= 0; i
< max_pages
; i
++) {
537 __free_page(pages
[i
]);
545 bool pnfs_roc(struct inode
*ino
)
547 struct pnfs_layout_hdr
*lo
;
548 struct pnfs_layout_segment
*lseg
, *tmp
;
552 spin_lock(&ino
->i_lock
);
553 lo
= NFS_I(ino
)->layout
;
554 if (!lo
|| !test_and_clear_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
) ||
555 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
557 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
558 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
559 mark_lseg_invalid(lseg
, &tmp_list
);
564 lo
->plh_block_lgets
++;
565 get_layout_hdr(lo
); /* matched in pnfs_roc_release */
566 spin_unlock(&ino
->i_lock
);
567 pnfs_free_lseg_list(&tmp_list
);
571 spin_unlock(&ino
->i_lock
);
575 void pnfs_roc_release(struct inode
*ino
)
577 struct pnfs_layout_hdr
*lo
;
579 spin_lock(&ino
->i_lock
);
580 lo
= NFS_I(ino
)->layout
;
581 lo
->plh_block_lgets
--;
582 put_layout_hdr_locked(lo
);
583 spin_unlock(&ino
->i_lock
);
586 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
588 struct pnfs_layout_hdr
*lo
;
590 spin_lock(&ino
->i_lock
);
591 lo
= NFS_I(ino
)->layout
;
592 if ((int)(barrier
- lo
->plh_barrier
) > 0)
593 lo
->plh_barrier
= barrier
;
594 spin_unlock(&ino
->i_lock
);
597 bool pnfs_roc_drain(struct inode
*ino
, u32
*barrier
)
599 struct nfs_inode
*nfsi
= NFS_I(ino
);
600 struct pnfs_layout_segment
*lseg
;
603 spin_lock(&ino
->i_lock
);
604 list_for_each_entry(lseg
, &nfsi
->layout
->plh_segs
, pls_list
)
605 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
610 struct pnfs_layout_hdr
*lo
= nfsi
->layout
;
611 u32 current_seqid
= be32_to_cpu(lo
->plh_stateid
.stateid
.seqid
);
613 /* Since close does not return a layout stateid for use as
614 * a barrier, we choose the worst-case barrier.
616 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
618 spin_unlock(&ino
->i_lock
);
623 * Compare two layout segments for sorting into layout cache.
624 * We want to preferentially return RW over RO layouts, so ensure those
628 cmp_layout(u32 iomode1
, u32 iomode2
)
630 /* read > read/write */
631 return (int)(iomode2
== IOMODE_READ
) - (int)(iomode1
== IOMODE_READ
);
635 pnfs_insert_layout(struct pnfs_layout_hdr
*lo
,
636 struct pnfs_layout_segment
*lseg
)
638 struct pnfs_layout_segment
*lp
;
641 dprintk("%s:Begin\n", __func__
);
643 assert_spin_locked(&lo
->plh_inode
->i_lock
);
644 list_for_each_entry(lp
, &lo
->plh_segs
, pls_list
) {
645 if (cmp_layout(lp
->pls_range
.iomode
, lseg
->pls_range
.iomode
) > 0)
647 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
648 dprintk("%s: inserted lseg %p "
649 "iomode %d offset %llu length %llu before "
650 "lp %p iomode %d offset %llu length %llu\n",
651 __func__
, lseg
, lseg
->pls_range
.iomode
,
652 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
653 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
654 lp
->pls_range
.length
);
659 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
660 dprintk("%s: inserted lseg %p "
661 "iomode %d offset %llu length %llu at tail\n",
662 __func__
, lseg
, lseg
->pls_range
.iomode
,
663 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
667 dprintk("%s:Return\n", __func__
);
670 static struct pnfs_layout_hdr
*
671 alloc_init_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
673 struct pnfs_layout_hdr
*lo
;
675 lo
= kzalloc(sizeof(struct pnfs_layout_hdr
), gfp_flags
);
678 atomic_set(&lo
->plh_refcount
, 1);
679 INIT_LIST_HEAD(&lo
->plh_layouts
);
680 INIT_LIST_HEAD(&lo
->plh_segs
);
681 INIT_LIST_HEAD(&lo
->plh_bulk_recall
);
686 static struct pnfs_layout_hdr
*
687 pnfs_find_alloc_layout(struct inode
*ino
, gfp_t gfp_flags
)
689 struct nfs_inode
*nfsi
= NFS_I(ino
);
690 struct pnfs_layout_hdr
*new = NULL
;
692 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
694 assert_spin_locked(&ino
->i_lock
);
696 if (test_bit(NFS_LAYOUT_DESTROYED
, &nfsi
->layout
->plh_flags
))
701 spin_unlock(&ino
->i_lock
);
702 new = alloc_init_layout_hdr(ino
, gfp_flags
);
703 spin_lock(&ino
->i_lock
);
705 if (likely(nfsi
->layout
== NULL
)) /* Won the race? */
713 * iomode matching rules:
724 is_matching_lseg(struct pnfs_layout_segment
*lseg
, u32 iomode
)
726 return (iomode
!= IOMODE_RW
|| lseg
->pls_range
.iomode
== IOMODE_RW
);
730 * lookup range in layout
732 static struct pnfs_layout_segment
*
733 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
, u32 iomode
)
735 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
737 dprintk("%s:Begin\n", __func__
);
739 assert_spin_locked(&lo
->plh_inode
->i_lock
);
740 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
741 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
742 is_matching_lseg(lseg
, iomode
)) {
743 ret
= get_lseg(lseg
);
746 if (cmp_layout(iomode
, lseg
->pls_range
.iomode
) > 0)
750 dprintk("%s:Return lseg %p ref %d\n",
751 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
756 * Layout segment is retreived from the server if not cached.
757 * The appropriate layout segment is referenced and returned to the caller.
759 struct pnfs_layout_segment
*
760 pnfs_update_layout(struct inode
*ino
,
761 struct nfs_open_context
*ctx
,
762 enum pnfs_iomode iomode
,
765 struct nfs_inode
*nfsi
= NFS_I(ino
);
766 struct nfs_client
*clp
= NFS_SERVER(ino
)->nfs_client
;
767 struct pnfs_layout_hdr
*lo
;
768 struct pnfs_layout_segment
*lseg
= NULL
;
771 if (!pnfs_enabled_sb(NFS_SERVER(ino
)))
773 spin_lock(&ino
->i_lock
);
774 lo
= pnfs_find_alloc_layout(ino
, gfp_flags
);
776 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__
);
780 /* Do we even need to bother with this? */
781 if (test_bit(NFS4CLNT_LAYOUTRECALL
, &clp
->cl_state
) ||
782 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
783 dprintk("%s matches recall, use MDS\n", __func__
);
787 /* if LAYOUTGET already failed once we don't try again */
788 if (test_bit(lo_fail_bit(iomode
), &nfsi
->layout
->plh_flags
))
791 /* Check to see if the layout for the given range already exists */
792 lseg
= pnfs_find_lseg(lo
, iomode
);
796 if (pnfs_layoutgets_blocked(lo
, NULL
, 0))
798 atomic_inc(&lo
->plh_outstanding
);
801 if (list_empty(&lo
->plh_segs
))
803 spin_unlock(&ino
->i_lock
);
805 /* The lo must be on the clp list if there is any
806 * chance of a CB_LAYOUTRECALL(FILE) coming in.
808 spin_lock(&clp
->cl_lock
);
809 BUG_ON(!list_empty(&lo
->plh_layouts
));
810 list_add_tail(&lo
->plh_layouts
, &clp
->cl_layouts
);
811 spin_unlock(&clp
->cl_lock
);
814 lseg
= send_layoutget(lo
, ctx
, iomode
, gfp_flags
);
815 if (!lseg
&& first
) {
816 spin_lock(&clp
->cl_lock
);
817 list_del_init(&lo
->plh_layouts
);
818 spin_unlock(&clp
->cl_lock
);
820 atomic_dec(&lo
->plh_outstanding
);
823 dprintk("%s end, state 0x%lx lseg %p\n", __func__
,
824 nfsi
->layout
? nfsi
->layout
->plh_flags
: -1, lseg
);
827 spin_unlock(&ino
->i_lock
);
832 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
834 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
835 struct nfs4_layoutget_res
*res
= &lgp
->res
;
836 struct pnfs_layout_segment
*lseg
;
837 struct inode
*ino
= lo
->plh_inode
;
838 struct nfs_client
*clp
= NFS_SERVER(ino
)->nfs_client
;
841 /* Verify we got what we asked for.
842 * Note that because the xdr parsing only accepts a single
843 * element array, this can fail even if the server is behaving
846 if (lgp
->args
.range
.iomode
> res
->range
.iomode
||
847 res
->range
.offset
!= 0 ||
848 res
->range
.length
!= NFS4_MAX_UINT64
) {
852 /* Inject layout blob into I/O device driver */
853 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
854 if (!lseg
|| IS_ERR(lseg
)) {
858 status
= PTR_ERR(lseg
);
859 dprintk("%s: Could not allocate layout: error %d\n",
864 spin_lock(&ino
->i_lock
);
865 if (test_bit(NFS4CLNT_LAYOUTRECALL
, &clp
->cl_state
) ||
866 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
867 dprintk("%s forget reply due to recall\n", __func__
);
868 goto out_forget_reply
;
871 if (pnfs_layoutgets_blocked(lo
, &res
->stateid
, 1)) {
872 dprintk("%s forget reply due to state\n", __func__
);
873 goto out_forget_reply
;
876 lseg
->pls_range
= res
->range
;
877 *lgp
->lsegpp
= get_lseg(lseg
);
878 pnfs_insert_layout(lo
, lseg
);
880 if (res
->return_on_close
) {
881 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
882 set_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
);
885 /* Done processing layoutget. Set the layout stateid */
886 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
887 spin_unlock(&ino
->i_lock
);
892 spin_unlock(&ino
->i_lock
);
893 lseg
->pls_layout
= lo
;
894 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
898 static int pnfs_read_pg_test(struct nfs_pageio_descriptor
*pgio
,
899 struct nfs_page
*prev
,
900 struct nfs_page
*req
)
902 if (pgio
->pg_count
== prev
->wb_bytes
) {
903 /* This is first coelesce call for a series of nfs_pages */
904 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
909 return NFS_SERVER(pgio
->pg_inode
)->pnfs_curr_ld
->pg_test(pgio
, prev
, req
);
913 pnfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
)
915 struct pnfs_layoutdriver_type
*ld
;
917 ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
918 pgio
->pg_test
= (ld
&& ld
->pg_test
) ? pnfs_read_pg_test
: NULL
;
921 static int pnfs_write_pg_test(struct nfs_pageio_descriptor
*pgio
,
922 struct nfs_page
*prev
,
923 struct nfs_page
*req
)
925 if (pgio
->pg_count
== prev
->wb_bytes
) {
926 /* This is first coelesce call for a series of nfs_pages */
927 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
932 return NFS_SERVER(pgio
->pg_inode
)->pnfs_curr_ld
->pg_test(pgio
, prev
, req
);
936 pnfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
)
938 struct pnfs_layoutdriver_type
*ld
;
940 ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
941 pgio
->pg_test
= (ld
&& ld
->pg_test
) ? pnfs_write_pg_test
: NULL
;
945 pnfs_try_to_write_data(struct nfs_write_data
*wdata
,
946 const struct rpc_call_ops
*call_ops
, int how
)
948 struct inode
*inode
= wdata
->inode
;
949 enum pnfs_try_status trypnfs
;
950 struct nfs_server
*nfss
= NFS_SERVER(inode
);
952 wdata
->mds_ops
= call_ops
;
954 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
955 inode
->i_ino
, wdata
->args
.count
, wdata
->args
.offset
, how
);
957 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(wdata
, how
);
958 if (trypnfs
== PNFS_NOT_ATTEMPTED
) {
959 put_lseg(wdata
->lseg
);
962 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
964 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
969 * Call the appropriate parallel I/O subsystem read function.
972 pnfs_try_to_read_data(struct nfs_read_data
*rdata
,
973 const struct rpc_call_ops
*call_ops
)
975 struct inode
*inode
= rdata
->inode
;
976 struct nfs_server
*nfss
= NFS_SERVER(inode
);
977 enum pnfs_try_status trypnfs
;
979 rdata
->mds_ops
= call_ops
;
981 dprintk("%s: Reading ino:%lu %u@%llu\n",
982 __func__
, inode
->i_ino
, rdata
->args
.count
, rdata
->args
.offset
);
984 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(rdata
);
985 if (trypnfs
== PNFS_NOT_ATTEMPTED
) {
986 put_lseg(rdata
->lseg
);
989 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
991 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
996 * Currently there is only one (whole file) write lseg.
998 static struct pnfs_layout_segment
*pnfs_list_write_lseg(struct inode
*inode
)
1000 struct pnfs_layout_segment
*lseg
, *rv
= NULL
;
1002 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
)
1003 if (lseg
->pls_range
.iomode
== IOMODE_RW
)
1009 pnfs_set_layoutcommit(struct nfs_write_data
*wdata
)
1011 struct nfs_inode
*nfsi
= NFS_I(wdata
->inode
);
1012 loff_t end_pos
= wdata
->mds_offset
+ wdata
->res
.count
;
1013 bool mark_as_dirty
= false;
1015 spin_lock(&nfsi
->vfs_inode
.i_lock
);
1016 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1017 /* references matched in nfs4_layoutcommit_release */
1018 get_lseg(wdata
->lseg
);
1019 wdata
->lseg
->pls_lc_cred
=
1020 get_rpccred(wdata
->args
.context
->state
->owner
->so_cred
);
1021 mark_as_dirty
= true;
1022 dprintk("%s: Set layoutcommit for inode %lu ",
1023 __func__
, wdata
->inode
->i_ino
);
1025 if (end_pos
> wdata
->lseg
->pls_end_pos
)
1026 wdata
->lseg
->pls_end_pos
= end_pos
;
1027 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
1029 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1030 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1032 mark_inode_dirty_sync(wdata
->inode
);
1034 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
1037 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1038 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1039 * data to disk to allow the server to recover the data if it crashes.
1040 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1041 * is off, and a COMMIT is sent to a data server, or
1042 * if WRITEs to a data server return NFS_DATA_SYNC.
1045 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
1047 struct nfs4_layoutcommit_data
*data
;
1048 struct nfs_inode
*nfsi
= NFS_I(inode
);
1049 struct pnfs_layout_segment
*lseg
;
1050 struct rpc_cred
*cred
;
1054 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
1056 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1059 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1060 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
1062 mark_inode_dirty_sync(inode
);
1067 spin_lock(&inode
->i_lock
);
1068 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1069 spin_unlock(&inode
->i_lock
);
1074 * Currently only one (whole file) write lseg which is referenced
1075 * in pnfs_set_layoutcommit and will be found.
1077 lseg
= pnfs_list_write_lseg(inode
);
1079 end_pos
= lseg
->pls_end_pos
;
1080 cred
= lseg
->pls_lc_cred
;
1081 lseg
->pls_end_pos
= 0;
1082 lseg
->pls_lc_cred
= NULL
;
1084 memcpy(&data
->args
.stateid
.data
, nfsi
->layout
->plh_stateid
.data
,
1085 sizeof(nfsi
->layout
->plh_stateid
.data
));
1086 spin_unlock(&inode
->i_lock
);
1088 data
->args
.inode
= inode
;
1091 nfs_fattr_init(&data
->fattr
);
1092 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
1093 data
->res
.fattr
= &data
->fattr
;
1094 data
->args
.lastbytewritten
= end_pos
- 1;
1095 data
->res
.server
= NFS_SERVER(inode
);
1097 status
= nfs4_proc_layoutcommit(data
, sync
);
1099 dprintk("<-- %s status %d\n", __func__
, status
);