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
38 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
43 * protects pnfs_modules_tbl.
45 static DEFINE_SPINLOCK(pnfs_spinlock
);
48 * pnfs_modules_tbl holds all pnfs modules
50 static LIST_HEAD(pnfs_modules_tbl
);
52 /* Return the registered pnfs layout driver module matching given id */
53 static struct pnfs_layoutdriver_type
*
54 find_pnfs_driver_locked(u32 id
)
56 struct pnfs_layoutdriver_type
*local
;
58 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
63 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
67 static struct pnfs_layoutdriver_type
*
68 find_pnfs_driver(u32 id
)
70 struct pnfs_layoutdriver_type
*local
;
72 spin_lock(&pnfs_spinlock
);
73 local
= find_pnfs_driver_locked(id
);
74 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
75 dprintk("%s: Could not grab reference on module\n", __func__
);
78 spin_unlock(&pnfs_spinlock
);
83 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
85 if (nfss
->pnfs_curr_ld
) {
86 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
87 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
88 /* Decrement the MDS count. Purge the deviceid cache if zero */
89 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
90 nfs4_deviceid_purge_client(nfss
->nfs_client
);
91 module_put(nfss
->pnfs_curr_ld
->owner
);
93 nfss
->pnfs_curr_ld
= NULL
;
97 * Try to set the server's pnfs module to the pnfs layout type specified by id.
98 * Currently only one pNFS layout driver per filesystem is supported.
100 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
103 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
106 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
110 if (!(server
->nfs_client
->cl_exchange_flags
&
111 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
112 printk(KERN_ERR
"NFS: %s: id %u cl_exchange_flags 0x%x\n",
113 __func__
, id
, server
->nfs_client
->cl_exchange_flags
);
116 ld_type
= find_pnfs_driver(id
);
118 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
, id
);
119 ld_type
= find_pnfs_driver(id
);
121 dprintk("%s: No pNFS module found for %u.\n",
126 server
->pnfs_curr_ld
= ld_type
;
127 if (ld_type
->set_layoutdriver
128 && ld_type
->set_layoutdriver(server
, mntfh
)) {
129 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
130 "driver %u.\n", __func__
, id
);
131 module_put(ld_type
->owner
);
134 /* Bump the MDS count */
135 atomic_inc(&server
->nfs_client
->cl_mds_count
);
137 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
141 dprintk("%s: Using NFSv4 I/O\n", __func__
);
142 server
->pnfs_curr_ld
= NULL
;
146 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
148 int status
= -EINVAL
;
149 struct pnfs_layoutdriver_type
*tmp
;
151 if (ld_type
->id
== 0) {
152 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
155 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
156 printk(KERN_ERR
"NFS: %s Layout driver must provide "
157 "alloc_lseg and free_lseg.\n", __func__
);
161 spin_lock(&pnfs_spinlock
);
162 tmp
= find_pnfs_driver_locked(ld_type
->id
);
164 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
166 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
169 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
170 __func__
, ld_type
->id
);
172 spin_unlock(&pnfs_spinlock
);
176 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
179 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
181 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
182 spin_lock(&pnfs_spinlock
);
183 list_del(&ld_type
->pnfs_tblid
);
184 spin_unlock(&pnfs_spinlock
);
186 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
189 * pNFS client layout cache
192 /* Need to hold i_lock if caller does not already hold reference */
194 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
196 atomic_inc(&lo
->plh_refcount
);
199 static struct pnfs_layout_hdr
*
200 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
202 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
203 return ld
->alloc_layout_hdr
? ld
->alloc_layout_hdr(ino
, gfp_flags
) :
204 kzalloc(sizeof(struct pnfs_layout_hdr
), gfp_flags
);
208 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
210 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
211 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
213 if (!list_empty(&lo
->plh_layouts
)) {
214 struct nfs_client
*clp
= server
->nfs_client
;
216 spin_lock(&clp
->cl_lock
);
217 list_del_init(&lo
->plh_layouts
);
218 spin_unlock(&clp
->cl_lock
);
220 put_rpccred(lo
->plh_lc_cred
);
221 return ld
->alloc_layout_hdr
? ld
->free_layout_hdr(lo
) : kfree(lo
);
225 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
227 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
228 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
230 /* Reset MDS Threshold I/O counters */
236 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
238 struct inode
*inode
= lo
->plh_inode
;
240 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
241 pnfs_detach_layout_hdr(lo
);
242 spin_unlock(&inode
->i_lock
);
243 pnfs_free_layout_hdr(lo
);
248 pnfs_iomode_to_fail_bit(u32 iomode
)
250 return iomode
== IOMODE_RW
?
251 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
255 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
257 lo
->plh_retry_timestamp
= jiffies
;
258 if (test_and_set_bit(fail_bit
, &lo
->plh_flags
))
259 atomic_inc(&lo
->plh_refcount
);
263 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
265 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
266 atomic_dec(&lo
->plh_refcount
);
270 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
272 struct inode
*inode
= lo
->plh_inode
;
273 struct pnfs_layout_range range
= {
276 .length
= NFS4_MAX_UINT64
,
280 spin_lock(&inode
->i_lock
);
281 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
282 pnfs_mark_matching_lsegs_invalid(lo
, &head
, &range
);
283 spin_unlock(&inode
->i_lock
);
284 pnfs_free_lseg_list(&head
);
285 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
286 iomode
== IOMODE_RW
? "RW" : "READ");
290 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
292 unsigned long start
, end
;
293 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
295 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
298 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
299 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
300 /* It is time to retry the failed layoutgets */
301 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
308 init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
)
310 INIT_LIST_HEAD(&lseg
->pls_list
);
311 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
312 atomic_set(&lseg
->pls_refcount
, 1);
314 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
315 lseg
->pls_layout
= lo
;
318 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
320 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
322 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
323 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
324 pnfs_put_layout_hdr(NFS_I(ino
)->layout
);
328 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
329 struct pnfs_layout_segment
*lseg
)
331 struct inode
*inode
= lo
->plh_inode
;
333 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
334 list_del_init(&lseg
->pls_list
);
335 if (list_empty(&lo
->plh_segs
))
336 set_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
);
337 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
341 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
343 struct pnfs_layout_hdr
*lo
;
349 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
350 atomic_read(&lseg
->pls_refcount
),
351 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
352 lo
= lseg
->pls_layout
;
353 inode
= lo
->plh_inode
;
354 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
355 pnfs_layout_remove_lseg(lo
, lseg
);
356 spin_unlock(&inode
->i_lock
);
357 pnfs_free_lseg(lseg
);
360 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
363 end_offset(u64 start
, u64 len
)
368 return end
>= start
? end
: NFS4_MAX_UINT64
;
371 /* last octet in a range */
373 last_byte_offset(u64 start
, u64 len
)
379 return end
> start
? end
- 1 : NFS4_MAX_UINT64
;
383 * is l2 fully contained in l1?
385 * [----------------------------------)
390 lo_seg_contained(struct pnfs_layout_range
*l1
,
391 struct pnfs_layout_range
*l2
)
393 u64 start1
= l1
->offset
;
394 u64 end1
= end_offset(start1
, l1
->length
);
395 u64 start2
= l2
->offset
;
396 u64 end2
= end_offset(start2
, l2
->length
);
398 return (start1
<= start2
) && (end1
>= end2
);
402 * is l1 and l2 intersecting?
404 * [----------------------------------)
409 lo_seg_intersecting(struct pnfs_layout_range
*l1
,
410 struct pnfs_layout_range
*l2
)
412 u64 start1
= l1
->offset
;
413 u64 end1
= end_offset(start1
, l1
->length
);
414 u64 start2
= l2
->offset
;
415 u64 end2
= end_offset(start2
, l2
->length
);
417 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
418 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
422 should_free_lseg(struct pnfs_layout_range
*lseg_range
,
423 struct pnfs_layout_range
*recall_range
)
425 return (recall_range
->iomode
== IOMODE_ANY
||
426 lseg_range
->iomode
== recall_range
->iomode
) &&
427 lo_seg_intersecting(lseg_range
, recall_range
);
430 /* Returns 1 if lseg is removed from list, 0 otherwise */
431 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
432 struct list_head
*tmp_list
)
436 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
437 /* Remove the reference keeping the lseg in the
438 * list. It will now be removed when all
439 * outstanding io is finished.
441 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
442 atomic_read(&lseg
->pls_refcount
));
443 if (atomic_dec_and_test(&lseg
->pls_refcount
)) {
444 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
445 list_add(&lseg
->pls_list
, tmp_list
);
452 /* Returns count of number of matching invalid lsegs remaining in list
456 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
457 struct list_head
*tmp_list
,
458 struct pnfs_layout_range
*recall_range
)
460 struct pnfs_layout_segment
*lseg
, *next
;
461 int invalid
= 0, removed
= 0;
463 dprintk("%s:Begin lo %p\n", __func__
, lo
);
465 if (list_empty(&lo
->plh_segs
)) {
466 set_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
);
469 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
471 should_free_lseg(&lseg
->pls_range
, recall_range
)) {
472 dprintk("%s: freeing lseg %p iomode %d "
473 "offset %llu length %llu\n", __func__
,
474 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_range
.offset
,
475 lseg
->pls_range
.length
);
477 removed
+= mark_lseg_invalid(lseg
, tmp_list
);
479 dprintk("%s:Return %i\n", __func__
, invalid
- removed
);
480 return invalid
- removed
;
483 /* note free_me must contain lsegs from a single layout_hdr */
485 pnfs_free_lseg_list(struct list_head
*free_me
)
487 struct pnfs_layout_segment
*lseg
, *tmp
;
489 if (list_empty(free_me
))
492 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
493 list_del(&lseg
->pls_list
);
494 pnfs_free_lseg(lseg
);
499 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
501 struct pnfs_layout_hdr
*lo
;
504 spin_lock(&nfsi
->vfs_inode
.i_lock
);
507 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
508 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
509 pnfs_get_layout_hdr(lo
);
510 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
511 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
512 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
513 pnfs_free_lseg_list(&tmp_list
);
514 pnfs_put_layout_hdr(lo
);
516 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
518 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
521 * Called by the state manger to remove all layouts established under an
525 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
527 struct nfs_server
*server
;
528 struct pnfs_layout_hdr
*lo
;
531 nfs4_deviceid_mark_client_invalid(clp
);
532 nfs4_deviceid_purge_client(clp
);
534 spin_lock(&clp
->cl_lock
);
536 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
537 if (!list_empty(&server
->layouts
))
538 list_splice_init(&server
->layouts
, &tmp_list
);
541 spin_unlock(&clp
->cl_lock
);
543 while (!list_empty(&tmp_list
)) {
544 lo
= list_entry(tmp_list
.next
, struct pnfs_layout_hdr
,
546 dprintk("%s freeing layout for inode %lu\n", __func__
,
547 lo
->plh_inode
->i_ino
);
548 list_del_init(&lo
->plh_layouts
);
549 pnfs_destroy_layout(NFS_I(lo
->plh_inode
));
553 /* update lo->plh_stateid with new if is more recent */
555 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
560 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
561 newseq
= be32_to_cpu(new->seqid
);
562 if ((int)(newseq
- oldseq
) > 0) {
563 nfs4_stateid_copy(&lo
->plh_stateid
, new);
564 if (update_barrier
) {
565 u32 new_barrier
= be32_to_cpu(new->seqid
);
567 if ((int)(new_barrier
- lo
->plh_barrier
))
568 lo
->plh_barrier
= new_barrier
;
570 /* Because of wraparound, we want to keep the barrier
571 * "close" to the current seqids. It needs to be
572 * within 2**31 to count as "behind", so if it
573 * gets too near that limit, give us a litle leeway
574 * and bring it to within 2**30.
575 * NOTE - and yes, this is all unsigned arithmetic.
577 if (unlikely((newseq
- lo
->plh_barrier
) > (3 << 29)))
578 lo
->plh_barrier
= newseq
- (1 << 30);
583 /* lget is set to 1 if called from inside send_layoutget call chain */
585 pnfs_layoutgets_blocked(struct pnfs_layout_hdr
*lo
, nfs4_stateid
*stateid
,
589 (int)(lo
->plh_barrier
- be32_to_cpu(stateid
->seqid
)) >= 0)
591 return lo
->plh_block_lgets
||
592 test_bit(NFS_LAYOUT_DESTROYED
, &lo
->plh_flags
) ||
593 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
) ||
594 (list_empty(&lo
->plh_segs
) &&
595 (atomic_read(&lo
->plh_outstanding
) > lget
));
599 pnfs_choose_layoutget_stateid(nfs4_stateid
*dst
, struct pnfs_layout_hdr
*lo
,
600 struct nfs4_state
*open_state
)
604 dprintk("--> %s\n", __func__
);
605 spin_lock(&lo
->plh_inode
->i_lock
);
606 if (pnfs_layoutgets_blocked(lo
, NULL
, 1)) {
608 } else if (list_empty(&lo
->plh_segs
)) {
612 seq
= read_seqbegin(&open_state
->seqlock
);
613 nfs4_stateid_copy(dst
, &open_state
->stateid
);
614 } while (read_seqretry(&open_state
->seqlock
, seq
));
616 nfs4_stateid_copy(dst
, &lo
->plh_stateid
);
617 spin_unlock(&lo
->plh_inode
->i_lock
);
618 dprintk("<-- %s\n", __func__
);
623 * Get layout from server.
624 * for now, assume that whole file layouts are requested.
626 * arg->length: all ones
628 static struct pnfs_layout_segment
*
629 send_layoutget(struct pnfs_layout_hdr
*lo
,
630 struct nfs_open_context
*ctx
,
631 struct pnfs_layout_range
*range
,
634 struct inode
*ino
= lo
->plh_inode
;
635 struct nfs_server
*server
= NFS_SERVER(ino
);
636 struct nfs4_layoutget
*lgp
;
637 struct pnfs_layout_segment
*lseg
;
639 dprintk("--> %s\n", __func__
);
642 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
646 lgp
->args
.minlength
= PAGE_CACHE_SIZE
;
647 if (lgp
->args
.minlength
> range
->length
)
648 lgp
->args
.minlength
= range
->length
;
649 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
650 lgp
->args
.range
= *range
;
651 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
652 lgp
->args
.inode
= ino
;
653 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
654 lgp
->gfp_flags
= gfp_flags
;
656 /* Synchronously retrieve layout information from server and
659 lseg
= nfs4_proc_layoutget(lgp
, gfp_flags
);
661 switch (PTR_ERR(lseg
)) {
666 /* remember that LAYOUTGET failed and suspend trying */
667 pnfs_layout_io_set_failed(lo
, range
->iomode
);
676 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
677 * when the layout segment list is empty.
679 * Note that a pnfs_layout_hdr can exist with an empty layout segment
680 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
681 * deviceid is marked invalid.
684 _pnfs_return_layout(struct inode
*ino
)
686 struct pnfs_layout_hdr
*lo
= NULL
;
687 struct nfs_inode
*nfsi
= NFS_I(ino
);
689 struct nfs4_layoutreturn
*lrp
;
690 nfs4_stateid stateid
;
691 int status
= 0, empty
;
693 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
695 spin_lock(&ino
->i_lock
);
697 if (!lo
|| pnfs_test_layout_returned(lo
)) {
698 spin_unlock(&ino
->i_lock
);
699 dprintk("NFS: %s no layout to return\n", __func__
);
702 stateid
= nfsi
->layout
->plh_stateid
;
703 /* Reference matched in nfs4_layoutreturn_release */
704 pnfs_get_layout_hdr(lo
);
705 empty
= list_empty(&lo
->plh_segs
);
706 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
707 /* Don't send a LAYOUTRETURN if list was initially empty */
709 spin_unlock(&ino
->i_lock
);
710 pnfs_put_layout_hdr(lo
);
711 dprintk("NFS: %s no layout segments to return\n", __func__
);
714 lo
->plh_block_lgets
++;
715 pnfs_mark_layout_returned(lo
);
716 spin_unlock(&ino
->i_lock
);
717 pnfs_free_lseg_list(&tmp_list
);
719 WARN_ON(test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
));
721 lrp
= kzalloc(sizeof(*lrp
), GFP_KERNEL
);
722 if (unlikely(lrp
== NULL
)) {
724 pnfs_layout_io_set_failed(lo
, IOMODE_RW
);
725 pnfs_layout_io_set_failed(lo
, IOMODE_READ
);
726 pnfs_clear_layout_returned(lo
);
727 pnfs_put_layout_hdr(lo
);
731 lrp
->args
.stateid
= stateid
;
732 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
733 lrp
->args
.inode
= ino
;
734 lrp
->args
.layout
= lo
;
735 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
737 status
= nfs4_proc_layoutreturn(lrp
);
739 dprintk("<-- %s status: %d\n", __func__
, status
);
742 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
744 bool pnfs_roc(struct inode
*ino
)
746 struct pnfs_layout_hdr
*lo
;
747 struct pnfs_layout_segment
*lseg
, *tmp
;
751 spin_lock(&ino
->i_lock
);
752 lo
= NFS_I(ino
)->layout
;
753 if (!lo
|| !test_and_clear_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
) ||
754 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
756 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
757 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
758 mark_lseg_invalid(lseg
, &tmp_list
);
763 lo
->plh_block_lgets
++;
764 pnfs_get_layout_hdr(lo
); /* matched in pnfs_roc_release */
765 spin_unlock(&ino
->i_lock
);
766 pnfs_free_lseg_list(&tmp_list
);
770 spin_unlock(&ino
->i_lock
);
774 void pnfs_roc_release(struct inode
*ino
)
776 struct pnfs_layout_hdr
*lo
;
778 spin_lock(&ino
->i_lock
);
779 lo
= NFS_I(ino
)->layout
;
780 lo
->plh_block_lgets
--;
781 if (atomic_dec_and_test(&lo
->plh_refcount
)) {
782 pnfs_detach_layout_hdr(lo
);
783 spin_unlock(&ino
->i_lock
);
784 pnfs_free_layout_hdr(lo
);
786 spin_unlock(&ino
->i_lock
);
789 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
791 struct pnfs_layout_hdr
*lo
;
793 spin_lock(&ino
->i_lock
);
794 lo
= NFS_I(ino
)->layout
;
795 if ((int)(barrier
- lo
->plh_barrier
) > 0)
796 lo
->plh_barrier
= barrier
;
797 spin_unlock(&ino
->i_lock
);
800 bool pnfs_roc_drain(struct inode
*ino
, u32
*barrier
, struct rpc_task
*task
)
802 struct nfs_inode
*nfsi
= NFS_I(ino
);
803 struct pnfs_layout_hdr
*lo
;
804 struct pnfs_layout_segment
*lseg
;
808 spin_lock(&ino
->i_lock
);
809 list_for_each_entry(lseg
, &nfsi
->layout
->plh_segs
, pls_list
)
810 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
811 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
816 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
818 /* Since close does not return a layout stateid for use as
819 * a barrier, we choose the worst-case barrier.
821 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
823 spin_unlock(&ino
->i_lock
);
828 * Compare two layout segments for sorting into layout cache.
829 * We want to preferentially return RW over RO layouts, so ensure those
833 cmp_layout(struct pnfs_layout_range
*l1
,
834 struct pnfs_layout_range
*l2
)
838 /* high offset > low offset */
839 d
= l1
->offset
- l2
->offset
;
843 /* short length > long length */
844 d
= l2
->length
- l1
->length
;
848 /* read > read/write */
849 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
853 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
854 struct pnfs_layout_segment
*lseg
)
856 struct pnfs_layout_segment
*lp
;
858 dprintk("%s:Begin\n", __func__
);
860 assert_spin_locked(&lo
->plh_inode
->i_lock
);
861 list_for_each_entry(lp
, &lo
->plh_segs
, pls_list
) {
862 if (cmp_layout(&lseg
->pls_range
, &lp
->pls_range
) > 0)
864 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
865 dprintk("%s: inserted lseg %p "
866 "iomode %d offset %llu length %llu before "
867 "lp %p iomode %d offset %llu length %llu\n",
868 __func__
, lseg
, lseg
->pls_range
.iomode
,
869 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
870 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
871 lp
->pls_range
.length
);
874 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
875 dprintk("%s: inserted lseg %p "
876 "iomode %d offset %llu length %llu at tail\n",
877 __func__
, lseg
, lseg
->pls_range
.iomode
,
878 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
880 pnfs_get_layout_hdr(lo
);
882 dprintk("%s:Return\n", __func__
);
885 static struct pnfs_layout_hdr
*
886 alloc_init_layout_hdr(struct inode
*ino
,
887 struct nfs_open_context
*ctx
,
890 struct pnfs_layout_hdr
*lo
;
892 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
895 atomic_set(&lo
->plh_refcount
, 1);
896 INIT_LIST_HEAD(&lo
->plh_layouts
);
897 INIT_LIST_HEAD(&lo
->plh_segs
);
898 INIT_LIST_HEAD(&lo
->plh_bulk_recall
);
900 lo
->plh_lc_cred
= get_rpccred(ctx
->state
->owner
->so_cred
);
904 static struct pnfs_layout_hdr
*
905 pnfs_find_alloc_layout(struct inode
*ino
,
906 struct nfs_open_context
*ctx
,
909 struct nfs_inode
*nfsi
= NFS_I(ino
);
910 struct pnfs_layout_hdr
*new = NULL
;
912 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
914 assert_spin_locked(&ino
->i_lock
);
916 if (test_bit(NFS_LAYOUT_DESTROYED
, &nfsi
->layout
->plh_flags
))
918 pnfs_get_layout_hdr(nfsi
->layout
);
921 spin_unlock(&ino
->i_lock
);
922 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
923 spin_lock(&ino
->i_lock
);
925 if (likely(nfsi
->layout
== NULL
)) /* Won the race? */
928 pnfs_free_layout_hdr(new);
933 * iomode matching rules:
944 is_matching_lseg(struct pnfs_layout_range
*ls_range
,
945 struct pnfs_layout_range
*range
)
947 struct pnfs_layout_range range1
;
949 if ((range
->iomode
== IOMODE_RW
&&
950 ls_range
->iomode
!= IOMODE_RW
) ||
951 !lo_seg_intersecting(ls_range
, range
))
954 /* range1 covers only the first byte in the range */
957 return lo_seg_contained(ls_range
, &range1
);
961 * lookup range in layout
963 static struct pnfs_layout_segment
*
964 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
965 struct pnfs_layout_range
*range
)
967 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
969 dprintk("%s:Begin\n", __func__
);
971 assert_spin_locked(&lo
->plh_inode
->i_lock
);
972 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
973 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
974 is_matching_lseg(&lseg
->pls_range
, range
)) {
975 ret
= pnfs_get_lseg(lseg
);
978 if (lseg
->pls_range
.offset
> range
->offset
)
982 dprintk("%s:Return lseg %p ref %d\n",
983 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
988 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
989 * to the MDS or over pNFS
991 * The nfs_inode read_io and write_io fields are cumulative counters reset
992 * when there are no layout segments. Note that in pnfs_update_layout iomode
993 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
996 * A return of true means use MDS I/O.
999 * If a file's size is smaller than the file size threshold, data accesses
1000 * SHOULD be sent to the metadata server. If an I/O request has a length that
1001 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1002 * server. If both file size and I/O size are provided, the client SHOULD
1003 * reach or exceed both thresholds before sending its read or write
1004 * requests to the data server.
1006 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1007 struct inode
*ino
, int iomode
)
1009 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1010 struct nfs_inode
*nfsi
= NFS_I(ino
);
1011 loff_t fsize
= i_size_read(ino
);
1012 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1017 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1018 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1022 if (t
->bm
& THRESHOLD_RD
) {
1023 dprintk("%s fsize %llu\n", __func__
, fsize
);
1025 if (fsize
< t
->rd_sz
)
1028 if (t
->bm
& THRESHOLD_RD_IO
) {
1029 dprintk("%s nfsi->read_io %llu\n", __func__
,
1032 if (nfsi
->read_io
< t
->rd_io_sz
)
1037 if (t
->bm
& THRESHOLD_WR
) {
1038 dprintk("%s fsize %llu\n", __func__
, fsize
);
1040 if (fsize
< t
->wr_sz
)
1043 if (t
->bm
& THRESHOLD_WR_IO
) {
1044 dprintk("%s nfsi->write_io %llu\n", __func__
,
1047 if (nfsi
->write_io
< t
->wr_io_sz
)
1052 if (size_set
&& io_set
) {
1055 } else if (size
|| io
)
1058 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1063 * Layout segment is retreived from the server if not cached.
1064 * The appropriate layout segment is referenced and returned to the caller.
1066 struct pnfs_layout_segment
*
1067 pnfs_update_layout(struct inode
*ino
,
1068 struct nfs_open_context
*ctx
,
1071 enum pnfs_iomode iomode
,
1074 struct pnfs_layout_range arg
= {
1080 struct nfs_server
*server
= NFS_SERVER(ino
);
1081 struct nfs_client
*clp
= server
->nfs_client
;
1082 struct pnfs_layout_hdr
*lo
;
1083 struct pnfs_layout_segment
*lseg
= NULL
;
1086 if (!pnfs_enabled_sb(NFS_SERVER(ino
)))
1089 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
))
1092 spin_lock(&ino
->i_lock
);
1093 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1095 spin_unlock(&ino
->i_lock
);
1099 /* Do we even need to bother with this? */
1100 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1101 dprintk("%s matches recall, use MDS\n", __func__
);
1105 /* if LAYOUTGET already failed once we don't try again */
1106 if (pnfs_layout_io_test_failed(lo
, iomode
))
1109 /* Check to see if the layout for the given range already exists */
1110 lseg
= pnfs_find_lseg(lo
, &arg
);
1114 if (pnfs_layoutgets_blocked(lo
, NULL
, 0))
1116 atomic_inc(&lo
->plh_outstanding
);
1118 if (list_empty(&lo
->plh_segs
))
1121 /* Enable LAYOUTRETURNs */
1122 pnfs_clear_layout_returned(lo
);
1124 spin_unlock(&ino
->i_lock
);
1126 /* The lo must be on the clp list if there is any
1127 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1129 spin_lock(&clp
->cl_lock
);
1130 BUG_ON(!list_empty(&lo
->plh_layouts
));
1131 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1132 spin_unlock(&clp
->cl_lock
);
1135 pg_offset
= arg
.offset
& ~PAGE_CACHE_MASK
;
1137 arg
.offset
-= pg_offset
;
1138 arg
.length
+= pg_offset
;
1140 if (arg
.length
!= NFS4_MAX_UINT64
)
1141 arg
.length
= PAGE_CACHE_ALIGN(arg
.length
);
1143 lseg
= send_layoutget(lo
, ctx
, &arg
, gfp_flags
);
1144 atomic_dec(&lo
->plh_outstanding
);
1146 pnfs_put_layout_hdr(lo
);
1148 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1149 "(%s, offset: %llu, length: %llu)\n",
1150 __func__
, ino
->i_sb
->s_id
,
1151 (unsigned long long)NFS_FILEID(ino
),
1152 lseg
== NULL
? "not found" : "found",
1153 iomode
==IOMODE_RW
? "read/write" : "read-only",
1154 (unsigned long long)pos
,
1155 (unsigned long long)count
);
1158 spin_unlock(&ino
->i_lock
);
1159 goto out_put_layout_hdr
;
1161 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1163 struct pnfs_layout_segment
*
1164 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1166 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1167 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1168 struct pnfs_layout_segment
*lseg
;
1169 struct inode
*ino
= lo
->plh_inode
;
1172 /* Inject layout blob into I/O device driver */
1173 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1174 if (!lseg
|| IS_ERR(lseg
)) {
1178 status
= PTR_ERR(lseg
);
1179 dprintk("%s: Could not allocate layout: error %d\n",
1184 spin_lock(&ino
->i_lock
);
1185 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1186 dprintk("%s forget reply due to recall\n", __func__
);
1187 goto out_forget_reply
;
1190 if (pnfs_layoutgets_blocked(lo
, &res
->stateid
, 1)) {
1191 dprintk("%s forget reply due to state\n", __func__
);
1192 goto out_forget_reply
;
1194 init_lseg(lo
, lseg
);
1195 lseg
->pls_range
= res
->range
;
1196 pnfs_get_lseg(lseg
);
1197 pnfs_layout_insert_lseg(lo
, lseg
);
1199 if (res
->return_on_close
) {
1200 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1201 set_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
);
1204 /* Done processing layoutget. Set the layout stateid */
1205 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1206 spin_unlock(&ino
->i_lock
);
1209 return ERR_PTR(status
);
1212 spin_unlock(&ino
->i_lock
);
1213 lseg
->pls_layout
= lo
;
1214 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1219 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1221 BUG_ON(pgio
->pg_lseg
!= NULL
);
1223 if (req
->wb_offset
!= req
->wb_pgbase
) {
1224 nfs_pageio_reset_read_mds(pgio
);
1227 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1233 /* If no lseg, fall back to read through mds */
1234 if (pgio
->pg_lseg
== NULL
)
1235 nfs_pageio_reset_read_mds(pgio
);
1238 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1241 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1243 BUG_ON(pgio
->pg_lseg
!= NULL
);
1245 if (req
->wb_offset
!= req
->wb_pgbase
) {
1246 nfs_pageio_reset_write_mds(pgio
);
1249 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1255 /* If no lseg, fall back to write through mds */
1256 if (pgio
->pg_lseg
== NULL
)
1257 nfs_pageio_reset_write_mds(pgio
);
1259 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1262 pnfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1263 const struct nfs_pgio_completion_ops
*compl_ops
)
1265 struct nfs_server
*server
= NFS_SERVER(inode
);
1266 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1269 nfs_pageio_init_read(pgio
, inode
, compl_ops
);
1271 nfs_pageio_init(pgio
, inode
, ld
->pg_read_ops
, compl_ops
, server
->rsize
, 0);
1275 pnfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1277 const struct nfs_pgio_completion_ops
*compl_ops
)
1279 struct nfs_server
*server
= NFS_SERVER(inode
);
1280 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1283 nfs_pageio_init_write(pgio
, inode
, ioflags
, compl_ops
);
1285 nfs_pageio_init(pgio
, inode
, ld
->pg_write_ops
, compl_ops
, server
->wsize
, ioflags
);
1289 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
1290 struct nfs_page
*req
)
1292 if (pgio
->pg_lseg
== NULL
)
1293 return nfs_generic_pg_test(pgio
, prev
, req
);
1296 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1297 * Note that this test makes several assumptions:
1298 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1299 * is known to lie within the range.
1300 * - that the nfs_page being tested is known to be contiguous with the
1301 * previous nfs_page.
1302 * - Layout ranges are page aligned, so we only have to test the
1303 * start offset of the request.
1305 * Please also note that 'end_offset' is actually the offset of the
1306 * first byte that lies outside the pnfs_layout_range. FIXME?
1309 return req_offset(req
) < end_offset(pgio
->pg_lseg
->pls_range
.offset
,
1310 pgio
->pg_lseg
->pls_range
.length
);
1312 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
1314 int pnfs_write_done_resend_to_mds(struct inode
*inode
,
1315 struct list_head
*head
,
1316 const struct nfs_pgio_completion_ops
*compl_ops
)
1318 struct nfs_pageio_descriptor pgio
;
1321 /* Resend all requests through the MDS */
1322 nfs_pageio_init_write(&pgio
, inode
, FLUSH_STABLE
, compl_ops
);
1323 while (!list_empty(head
)) {
1324 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1326 nfs_list_remove_request(req
);
1327 if (!nfs_pageio_add_request(&pgio
, req
))
1328 nfs_list_add_request(req
, &failed
);
1330 nfs_pageio_complete(&pgio
);
1332 if (!list_empty(&failed
)) {
1333 /* For some reason our attempt to resend pages. Mark the
1334 * overall send request as having failed, and let
1335 * nfs_writeback_release_full deal with the error.
1337 list_move(&failed
, head
);
1342 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
1344 static void pnfs_ld_handle_write_error(struct nfs_write_data
*data
)
1346 struct nfs_pgio_header
*hdr
= data
->header
;
1348 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
1349 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1350 PNFS_LAYOUTRET_ON_ERROR
) {
1351 clear_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(hdr
->inode
)->flags
);
1352 pnfs_return_layout(hdr
->inode
);
1354 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1355 data
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
->inode
,
1357 hdr
->completion_ops
);
1361 * Called by non rpc-based layout drivers
1363 void pnfs_ld_write_done(struct nfs_write_data
*data
)
1365 struct nfs_pgio_header
*hdr
= data
->header
;
1367 if (!hdr
->pnfs_error
) {
1368 pnfs_set_layoutcommit(data
);
1369 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1371 pnfs_ld_handle_write_error(data
);
1372 hdr
->mds_ops
->rpc_release(data
);
1374 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
1377 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
1378 struct nfs_write_data
*data
)
1380 struct nfs_pgio_header
*hdr
= data
->header
;
1382 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1383 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1384 nfs_pageio_reset_write_mds(desc
);
1385 desc
->pg_recoalesce
= 1;
1387 nfs_writedata_release(data
);
1390 static enum pnfs_try_status
1391 pnfs_try_to_write_data(struct nfs_write_data
*wdata
,
1392 const struct rpc_call_ops
*call_ops
,
1393 struct pnfs_layout_segment
*lseg
,
1396 struct nfs_pgio_header
*hdr
= wdata
->header
;
1397 struct inode
*inode
= hdr
->inode
;
1398 enum pnfs_try_status trypnfs
;
1399 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1401 hdr
->mds_ops
= call_ops
;
1403 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
1404 inode
->i_ino
, wdata
->args
.count
, wdata
->args
.offset
, how
);
1405 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(wdata
, how
);
1406 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1407 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
1408 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1413 pnfs_do_multiple_writes(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
, int how
)
1415 struct nfs_write_data
*data
;
1416 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1417 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1419 desc
->pg_lseg
= NULL
;
1420 while (!list_empty(head
)) {
1421 enum pnfs_try_status trypnfs
;
1423 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1424 list_del_init(&data
->list
);
1426 trypnfs
= pnfs_try_to_write_data(data
, call_ops
, lseg
, how
);
1427 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1428 pnfs_write_through_mds(desc
, data
);
1430 pnfs_put_lseg(lseg
);
1433 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
1435 pnfs_put_lseg(hdr
->lseg
);
1436 nfs_writehdr_free(hdr
);
1438 EXPORT_SYMBOL_GPL(pnfs_writehdr_free
);
1441 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1443 struct nfs_write_header
*whdr
;
1444 struct nfs_pgio_header
*hdr
;
1447 whdr
= nfs_writehdr_alloc();
1449 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1450 pnfs_put_lseg(desc
->pg_lseg
);
1451 desc
->pg_lseg
= NULL
;
1454 hdr
= &whdr
->header
;
1455 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
1456 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1457 atomic_inc(&hdr
->refcnt
);
1458 ret
= nfs_generic_flush(desc
, hdr
);
1460 pnfs_put_lseg(desc
->pg_lseg
);
1461 desc
->pg_lseg
= NULL
;
1463 pnfs_do_multiple_writes(desc
, &hdr
->rpc_list
, desc
->pg_ioflags
);
1464 if (atomic_dec_and_test(&hdr
->refcnt
))
1465 hdr
->completion_ops
->completion(hdr
);
1468 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
1470 int pnfs_read_done_resend_to_mds(struct inode
*inode
,
1471 struct list_head
*head
,
1472 const struct nfs_pgio_completion_ops
*compl_ops
)
1474 struct nfs_pageio_descriptor pgio
;
1477 /* Resend all requests through the MDS */
1478 nfs_pageio_init_read(&pgio
, inode
, compl_ops
);
1479 while (!list_empty(head
)) {
1480 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1482 nfs_list_remove_request(req
);
1483 if (!nfs_pageio_add_request(&pgio
, req
))
1484 nfs_list_add_request(req
, &failed
);
1486 nfs_pageio_complete(&pgio
);
1488 if (!list_empty(&failed
)) {
1489 list_move(&failed
, head
);
1494 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
1496 static void pnfs_ld_handle_read_error(struct nfs_read_data
*data
)
1498 struct nfs_pgio_header
*hdr
= data
->header
;
1500 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
1501 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1502 PNFS_LAYOUTRET_ON_ERROR
) {
1503 clear_bit(NFS_INO_LAYOUTCOMMIT
, &NFS_I(hdr
->inode
)->flags
);
1504 pnfs_return_layout(hdr
->inode
);
1506 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1507 data
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
->inode
,
1509 hdr
->completion_ops
);
1513 * Called by non rpc-based layout drivers
1515 void pnfs_ld_read_done(struct nfs_read_data
*data
)
1517 struct nfs_pgio_header
*hdr
= data
->header
;
1519 if (likely(!hdr
->pnfs_error
)) {
1520 __nfs4_read_done_cb(data
);
1521 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1523 pnfs_ld_handle_read_error(data
);
1524 hdr
->mds_ops
->rpc_release(data
);
1526 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
1529 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
1530 struct nfs_read_data
*data
)
1532 struct nfs_pgio_header
*hdr
= data
->header
;
1534 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1535 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1536 nfs_pageio_reset_read_mds(desc
);
1537 desc
->pg_recoalesce
= 1;
1539 nfs_readdata_release(data
);
1543 * Call the appropriate parallel I/O subsystem read function.
1545 static enum pnfs_try_status
1546 pnfs_try_to_read_data(struct nfs_read_data
*rdata
,
1547 const struct rpc_call_ops
*call_ops
,
1548 struct pnfs_layout_segment
*lseg
)
1550 struct nfs_pgio_header
*hdr
= rdata
->header
;
1551 struct inode
*inode
= hdr
->inode
;
1552 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1553 enum pnfs_try_status trypnfs
;
1555 hdr
->mds_ops
= call_ops
;
1557 dprintk("%s: Reading ino:%lu %u@%llu\n",
1558 __func__
, inode
->i_ino
, rdata
->args
.count
, rdata
->args
.offset
);
1560 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(rdata
);
1561 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1562 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
1563 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1568 pnfs_do_multiple_reads(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
)
1570 struct nfs_read_data
*data
;
1571 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1572 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1574 desc
->pg_lseg
= NULL
;
1575 while (!list_empty(head
)) {
1576 enum pnfs_try_status trypnfs
;
1578 data
= list_first_entry(head
, struct nfs_read_data
, list
);
1579 list_del_init(&data
->list
);
1581 trypnfs
= pnfs_try_to_read_data(data
, call_ops
, lseg
);
1582 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1583 pnfs_read_through_mds(desc
, data
);
1585 pnfs_put_lseg(lseg
);
1588 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
1590 pnfs_put_lseg(hdr
->lseg
);
1591 nfs_readhdr_free(hdr
);
1593 EXPORT_SYMBOL_GPL(pnfs_readhdr_free
);
1596 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
1598 struct nfs_read_header
*rhdr
;
1599 struct nfs_pgio_header
*hdr
;
1602 rhdr
= nfs_readhdr_alloc();
1604 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1606 pnfs_put_lseg(desc
->pg_lseg
);
1607 desc
->pg_lseg
= NULL
;
1610 hdr
= &rhdr
->header
;
1611 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
1612 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1613 atomic_inc(&hdr
->refcnt
);
1614 ret
= nfs_generic_pagein(desc
, hdr
);
1616 pnfs_put_lseg(desc
->pg_lseg
);
1617 desc
->pg_lseg
= NULL
;
1619 pnfs_do_multiple_reads(desc
, &hdr
->rpc_list
);
1620 if (atomic_dec_and_test(&hdr
->refcnt
))
1621 hdr
->completion_ops
->completion(hdr
);
1624 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
1627 * There can be multiple RW segments.
1629 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
1631 struct pnfs_layout_segment
*lseg
;
1633 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
1634 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
1635 test_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1636 list_add(&lseg
->pls_lc_list
, listp
);
1640 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
1642 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
1644 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
1647 pnfs_set_layoutcommit(struct nfs_write_data
*wdata
)
1649 struct nfs_pgio_header
*hdr
= wdata
->header
;
1650 struct inode
*inode
= hdr
->inode
;
1651 struct nfs_inode
*nfsi
= NFS_I(inode
);
1652 loff_t end_pos
= wdata
->mds_offset
+ wdata
->res
.count
;
1653 bool mark_as_dirty
= false;
1655 spin_lock(&inode
->i_lock
);
1656 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1657 mark_as_dirty
= true;
1658 dprintk("%s: Set layoutcommit for inode %lu ",
1659 __func__
, inode
->i_ino
);
1661 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &hdr
->lseg
->pls_flags
)) {
1662 /* references matched in nfs4_layoutcommit_release */
1663 pnfs_get_lseg(hdr
->lseg
);
1665 if (end_pos
> nfsi
->layout
->plh_lwb
)
1666 nfsi
->layout
->plh_lwb
= end_pos
;
1667 spin_unlock(&inode
->i_lock
);
1668 dprintk("%s: lseg %p end_pos %llu\n",
1669 __func__
, hdr
->lseg
, nfsi
->layout
->plh_lwb
);
1671 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1672 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1674 mark_inode_dirty_sync(inode
);
1676 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
1678 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
1680 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
1682 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
1683 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
1687 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1688 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1689 * data to disk to allow the server to recover the data if it crashes.
1690 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1691 * is off, and a COMMIT is sent to a data server, or
1692 * if WRITEs to a data server return NFS_DATA_SYNC.
1695 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
1697 struct nfs4_layoutcommit_data
*data
;
1698 struct nfs_inode
*nfsi
= NFS_I(inode
);
1702 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
1704 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1707 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1708 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
1714 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1717 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
1722 status
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
,
1723 nfs_wait_bit_killable
, TASK_KILLABLE
);
1728 INIT_LIST_HEAD(&data
->lseg_list
);
1729 spin_lock(&inode
->i_lock
);
1730 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1731 clear_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
);
1732 spin_unlock(&inode
->i_lock
);
1733 wake_up_bit(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
);
1737 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
1739 end_pos
= nfsi
->layout
->plh_lwb
;
1740 nfsi
->layout
->plh_lwb
= 0;
1742 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
1743 spin_unlock(&inode
->i_lock
);
1745 data
->args
.inode
= inode
;
1746 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
1747 nfs_fattr_init(&data
->fattr
);
1748 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
1749 data
->res
.fattr
= &data
->fattr
;
1750 data
->args
.lastbytewritten
= end_pos
- 1;
1751 data
->res
.server
= NFS_SERVER(inode
);
1753 status
= nfs4_proc_layoutcommit(data
, sync
);
1756 mark_inode_dirty_sync(inode
);
1757 dprintk("<-- %s status %d\n", __func__
, status
);
1764 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
1766 struct nfs4_threshold
*thp
;
1768 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
1770 dprintk("%s mdsthreshold allocation failed\n", __func__
);