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(ino
, gfp_flags
);
207 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
209 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
210 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
212 if (!list_empty(&lo
->plh_layouts
)) {
213 struct nfs_client
*clp
= server
->nfs_client
;
215 spin_lock(&clp
->cl_lock
);
216 list_del_init(&lo
->plh_layouts
);
217 spin_unlock(&clp
->cl_lock
);
219 put_rpccred(lo
->plh_lc_cred
);
220 return ld
->free_layout_hdr(lo
);
224 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
226 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
227 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
229 /* Reset MDS Threshold I/O counters */
235 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
237 struct inode
*inode
= lo
->plh_inode
;
239 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
240 pnfs_detach_layout_hdr(lo
);
241 spin_unlock(&inode
->i_lock
);
242 pnfs_free_layout_hdr(lo
);
247 pnfs_iomode_to_fail_bit(u32 iomode
)
249 return iomode
== IOMODE_RW
?
250 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
254 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
256 lo
->plh_retry_timestamp
= jiffies
;
257 if (!test_and_set_bit(fail_bit
, &lo
->plh_flags
))
258 atomic_inc(&lo
->plh_refcount
);
262 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
264 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
265 atomic_dec(&lo
->plh_refcount
);
269 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
271 struct inode
*inode
= lo
->plh_inode
;
272 struct pnfs_layout_range range
= {
275 .length
= NFS4_MAX_UINT64
,
279 spin_lock(&inode
->i_lock
);
280 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
281 pnfs_mark_matching_lsegs_invalid(lo
, &head
, &range
);
282 spin_unlock(&inode
->i_lock
);
283 pnfs_free_lseg_list(&head
);
284 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
285 iomode
== IOMODE_RW
? "RW" : "READ");
289 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
291 unsigned long start
, end
;
292 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
294 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
297 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
298 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
299 /* It is time to retry the failed layoutgets */
300 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
307 init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
)
309 INIT_LIST_HEAD(&lseg
->pls_list
);
310 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
311 atomic_set(&lseg
->pls_refcount
, 1);
313 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
314 lseg
->pls_layout
= lo
;
317 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
319 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
321 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
325 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
326 struct pnfs_layout_segment
*lseg
)
328 struct inode
*inode
= lo
->plh_inode
;
330 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
331 list_del_init(&lseg
->pls_list
);
332 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
333 atomic_dec(&lo
->plh_refcount
);
334 if (list_empty(&lo
->plh_segs
))
335 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
336 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
340 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
342 struct pnfs_layout_hdr
*lo
;
348 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
349 atomic_read(&lseg
->pls_refcount
),
350 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
351 lo
= lseg
->pls_layout
;
352 inode
= lo
->plh_inode
;
353 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
354 pnfs_get_layout_hdr(lo
);
355 pnfs_layout_remove_lseg(lo
, lseg
);
356 spin_unlock(&inode
->i_lock
);
357 pnfs_free_lseg(lseg
);
358 pnfs_put_layout_hdr(lo
);
361 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
364 end_offset(u64 start
, u64 len
)
369 return end
>= start
? end
: NFS4_MAX_UINT64
;
373 * is l2 fully contained in l1?
375 * [----------------------------------)
380 lo_seg_contained(struct pnfs_layout_range
*l1
,
381 struct pnfs_layout_range
*l2
)
383 u64 start1
= l1
->offset
;
384 u64 end1
= end_offset(start1
, l1
->length
);
385 u64 start2
= l2
->offset
;
386 u64 end2
= end_offset(start2
, l2
->length
);
388 return (start1
<= start2
) && (end1
>= end2
);
392 * is l1 and l2 intersecting?
394 * [----------------------------------)
399 lo_seg_intersecting(struct pnfs_layout_range
*l1
,
400 struct pnfs_layout_range
*l2
)
402 u64 start1
= l1
->offset
;
403 u64 end1
= end_offset(start1
, l1
->length
);
404 u64 start2
= l2
->offset
;
405 u64 end2
= end_offset(start2
, l2
->length
);
407 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
408 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
412 should_free_lseg(struct pnfs_layout_range
*lseg_range
,
413 struct pnfs_layout_range
*recall_range
)
415 return (recall_range
->iomode
== IOMODE_ANY
||
416 lseg_range
->iomode
== recall_range
->iomode
) &&
417 lo_seg_intersecting(lseg_range
, recall_range
);
420 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
421 struct list_head
*tmp_list
)
423 if (!atomic_dec_and_test(&lseg
->pls_refcount
))
425 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
426 list_add(&lseg
->pls_list
, tmp_list
);
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 (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
449 /* Returns count of number of matching invalid lsegs remaining in list
453 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
454 struct list_head
*tmp_list
,
455 struct pnfs_layout_range
*recall_range
)
457 struct pnfs_layout_segment
*lseg
, *next
;
458 int invalid
= 0, removed
= 0;
460 dprintk("%s:Begin lo %p\n", __func__
, lo
);
462 if (list_empty(&lo
->plh_segs
))
464 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
466 should_free_lseg(&lseg
->pls_range
, recall_range
)) {
467 dprintk("%s: freeing lseg %p iomode %d "
468 "offset %llu length %llu\n", __func__
,
469 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_range
.offset
,
470 lseg
->pls_range
.length
);
472 removed
+= mark_lseg_invalid(lseg
, tmp_list
);
474 dprintk("%s:Return %i\n", __func__
, invalid
- removed
);
475 return invalid
- removed
;
478 /* note free_me must contain lsegs from a single layout_hdr */
480 pnfs_free_lseg_list(struct list_head
*free_me
)
482 struct pnfs_layout_segment
*lseg
, *tmp
;
484 if (list_empty(free_me
))
487 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
488 list_del(&lseg
->pls_list
);
489 pnfs_free_lseg(lseg
);
494 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
496 struct pnfs_layout_hdr
*lo
;
499 spin_lock(&nfsi
->vfs_inode
.i_lock
);
502 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
503 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
504 pnfs_get_layout_hdr(lo
);
505 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
506 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
507 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
508 pnfs_free_lseg_list(&tmp_list
);
509 pnfs_put_layout_hdr(lo
);
511 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
513 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
516 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
517 struct list_head
*layout_list
)
519 struct pnfs_layout_hdr
*lo
;
522 spin_lock(&inode
->i_lock
);
523 lo
= NFS_I(inode
)->layout
;
524 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
525 pnfs_get_layout_hdr(lo
);
526 list_add(&lo
->plh_bulk_destroy
, layout_list
);
529 spin_unlock(&inode
->i_lock
);
533 /* Caller must hold rcu_read_lock and clp->cl_lock */
535 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
536 struct nfs_server
*server
,
537 struct list_head
*layout_list
)
539 struct pnfs_layout_hdr
*lo
, *next
;
542 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
543 inode
= igrab(lo
->plh_inode
);
546 list_del_init(&lo
->plh_layouts
);
547 if (pnfs_layout_add_bulk_destroy_list(inode
, layout_list
))
550 spin_unlock(&clp
->cl_lock
);
552 spin_lock(&clp
->cl_lock
);
560 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
563 struct pnfs_layout_hdr
*lo
;
565 struct pnfs_layout_range range
= {
566 .iomode
= IOMODE_ANY
,
568 .length
= NFS4_MAX_UINT64
,
570 LIST_HEAD(lseg_list
);
573 while (!list_empty(layout_list
)) {
574 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
576 dprintk("%s freeing layout for inode %lu\n", __func__
,
577 lo
->plh_inode
->i_ino
);
578 inode
= lo
->plh_inode
;
579 spin_lock(&inode
->i_lock
);
580 list_del_init(&lo
->plh_bulk_destroy
);
581 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
583 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
584 if (pnfs_mark_matching_lsegs_invalid(lo
, &lseg_list
, &range
))
586 spin_unlock(&inode
->i_lock
);
587 pnfs_free_lseg_list(&lseg_list
);
588 pnfs_put_layout_hdr(lo
);
595 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
596 struct nfs_fsid
*fsid
,
599 struct nfs_server
*server
;
600 LIST_HEAD(layout_list
);
602 spin_lock(&clp
->cl_lock
);
605 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
606 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
608 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
614 spin_unlock(&clp
->cl_lock
);
616 if (list_empty(&layout_list
))
618 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
622 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
625 struct nfs_server
*server
;
626 LIST_HEAD(layout_list
);
628 spin_lock(&clp
->cl_lock
);
631 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
632 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
638 spin_unlock(&clp
->cl_lock
);
640 if (list_empty(&layout_list
))
642 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
646 * Called by the state manger to remove all layouts established under an
650 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
652 nfs4_deviceid_mark_client_invalid(clp
);
653 nfs4_deviceid_purge_client(clp
);
655 pnfs_destroy_layouts_byclid(clp
, false);
659 * Compare 2 layout stateid sequence ids, to see which is newer,
660 * taking into account wraparound issues.
662 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
664 return (s32
)s1
- (s32
)s2
> 0;
667 /* update lo->plh_stateid with new if is more recent */
669 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
672 u32 oldseq
, newseq
, new_barrier
;
673 int empty
= list_empty(&lo
->plh_segs
);
675 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
676 newseq
= be32_to_cpu(new->seqid
);
677 if (empty
|| pnfs_seqid_is_newer(newseq
, oldseq
)) {
678 nfs4_stateid_copy(&lo
->plh_stateid
, new);
679 if (update_barrier
) {
680 new_barrier
= be32_to_cpu(new->seqid
);
682 /* Because of wraparound, we want to keep the barrier
683 * "close" to the current seqids.
685 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
687 if (empty
|| pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
688 lo
->plh_barrier
= new_barrier
;
693 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
694 const nfs4_stateid
*stateid
)
696 u32 seqid
= be32_to_cpu(stateid
->seqid
);
698 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
701 /* lget is set to 1 if called from inside send_layoutget call chain */
703 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
, int lget
)
705 return lo
->plh_block_lgets
||
706 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
) ||
707 (list_empty(&lo
->plh_segs
) &&
708 (atomic_read(&lo
->plh_outstanding
) > lget
));
712 pnfs_choose_layoutget_stateid(nfs4_stateid
*dst
, struct pnfs_layout_hdr
*lo
,
713 struct nfs4_state
*open_state
)
717 dprintk("--> %s\n", __func__
);
718 spin_lock(&lo
->plh_inode
->i_lock
);
719 if (pnfs_layoutgets_blocked(lo
, 1)) {
721 } else if (list_empty(&lo
->plh_segs
)) {
725 seq
= read_seqbegin(&open_state
->seqlock
);
726 nfs4_stateid_copy(dst
, &open_state
->stateid
);
727 } while (read_seqretry(&open_state
->seqlock
, seq
));
729 nfs4_stateid_copy(dst
, &lo
->plh_stateid
);
730 spin_unlock(&lo
->plh_inode
->i_lock
);
731 dprintk("<-- %s\n", __func__
);
736 * Get layout from server.
737 * for now, assume that whole file layouts are requested.
739 * arg->length: all ones
741 static struct pnfs_layout_segment
*
742 send_layoutget(struct pnfs_layout_hdr
*lo
,
743 struct nfs_open_context
*ctx
,
744 struct pnfs_layout_range
*range
,
747 struct inode
*ino
= lo
->plh_inode
;
748 struct nfs_server
*server
= NFS_SERVER(ino
);
749 struct nfs4_layoutget
*lgp
;
750 struct pnfs_layout_segment
*lseg
;
752 dprintk("--> %s\n", __func__
);
754 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
758 lgp
->args
.minlength
= PAGE_CACHE_SIZE
;
759 if (lgp
->args
.minlength
> range
->length
)
760 lgp
->args
.minlength
= range
->length
;
761 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
762 lgp
->args
.range
= *range
;
763 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
764 lgp
->args
.inode
= ino
;
765 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
766 lgp
->gfp_flags
= gfp_flags
;
768 /* Synchronously retrieve layout information from server and
771 lseg
= nfs4_proc_layoutget(lgp
, gfp_flags
);
773 switch (PTR_ERR(lseg
)) {
778 /* remember that LAYOUTGET failed and suspend trying */
779 pnfs_layout_io_set_failed(lo
, range
->iomode
);
787 static void pnfs_clear_layoutcommit(struct inode
*inode
,
788 struct list_head
*head
)
790 struct nfs_inode
*nfsi
= NFS_I(inode
);
791 struct pnfs_layout_segment
*lseg
, *tmp
;
793 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
795 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
796 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
798 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
803 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
804 * when the layout segment list is empty.
806 * Note that a pnfs_layout_hdr can exist with an empty layout segment
807 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
808 * deviceid is marked invalid.
811 _pnfs_return_layout(struct inode
*ino
)
813 struct pnfs_layout_hdr
*lo
= NULL
;
814 struct nfs_inode
*nfsi
= NFS_I(ino
);
816 struct nfs4_layoutreturn
*lrp
;
817 nfs4_stateid stateid
;
818 int status
= 0, empty
;
820 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
822 spin_lock(&ino
->i_lock
);
825 spin_unlock(&ino
->i_lock
);
826 dprintk("NFS: %s no layout to return\n", __func__
);
829 stateid
= nfsi
->layout
->plh_stateid
;
830 /* Reference matched in nfs4_layoutreturn_release */
831 pnfs_get_layout_hdr(lo
);
832 empty
= list_empty(&lo
->plh_segs
);
833 pnfs_clear_layoutcommit(ino
, &tmp_list
);
834 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
835 /* Don't send a LAYOUTRETURN if list was initially empty */
837 spin_unlock(&ino
->i_lock
);
838 pnfs_put_layout_hdr(lo
);
839 dprintk("NFS: %s no layout segments to return\n", __func__
);
842 lo
->plh_block_lgets
++;
843 spin_unlock(&ino
->i_lock
);
844 pnfs_free_lseg_list(&tmp_list
);
846 lrp
= kzalloc(sizeof(*lrp
), GFP_KERNEL
);
847 if (unlikely(lrp
== NULL
)) {
849 spin_lock(&ino
->i_lock
);
850 lo
->plh_block_lgets
--;
851 spin_unlock(&ino
->i_lock
);
852 pnfs_put_layout_hdr(lo
);
856 lrp
->args
.stateid
= stateid
;
857 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
858 lrp
->args
.inode
= ino
;
859 lrp
->args
.layout
= lo
;
860 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
862 status
= nfs4_proc_layoutreturn(lrp
);
864 dprintk("<-- %s status: %d\n", __func__
, status
);
867 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
870 pnfs_commit_and_return_layout(struct inode
*inode
)
872 struct pnfs_layout_hdr
*lo
;
875 spin_lock(&inode
->i_lock
);
876 lo
= NFS_I(inode
)->layout
;
878 spin_unlock(&inode
->i_lock
);
881 pnfs_get_layout_hdr(lo
);
882 /* Block new layoutgets and read/write to ds */
883 lo
->plh_block_lgets
++;
884 spin_unlock(&inode
->i_lock
);
885 filemap_fdatawait(inode
->i_mapping
);
886 ret
= pnfs_layoutcommit_inode(inode
, true);
888 ret
= _pnfs_return_layout(inode
);
889 spin_lock(&inode
->i_lock
);
890 lo
->plh_block_lgets
--;
891 spin_unlock(&inode
->i_lock
);
892 pnfs_put_layout_hdr(lo
);
896 bool pnfs_roc(struct inode
*ino
)
898 struct pnfs_layout_hdr
*lo
;
899 struct pnfs_layout_segment
*lseg
, *tmp
;
903 spin_lock(&ino
->i_lock
);
904 lo
= NFS_I(ino
)->layout
;
905 if (!lo
|| !test_and_clear_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
) ||
906 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
908 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
909 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
910 mark_lseg_invalid(lseg
, &tmp_list
);
915 lo
->plh_block_lgets
++;
916 pnfs_get_layout_hdr(lo
); /* matched in pnfs_roc_release */
917 spin_unlock(&ino
->i_lock
);
918 pnfs_free_lseg_list(&tmp_list
);
922 spin_unlock(&ino
->i_lock
);
926 void pnfs_roc_release(struct inode
*ino
)
928 struct pnfs_layout_hdr
*lo
;
930 spin_lock(&ino
->i_lock
);
931 lo
= NFS_I(ino
)->layout
;
932 lo
->plh_block_lgets
--;
933 if (atomic_dec_and_test(&lo
->plh_refcount
)) {
934 pnfs_detach_layout_hdr(lo
);
935 spin_unlock(&ino
->i_lock
);
936 pnfs_free_layout_hdr(lo
);
938 spin_unlock(&ino
->i_lock
);
941 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
943 struct pnfs_layout_hdr
*lo
;
945 spin_lock(&ino
->i_lock
);
946 lo
= NFS_I(ino
)->layout
;
947 if (pnfs_seqid_is_newer(barrier
, lo
->plh_barrier
))
948 lo
->plh_barrier
= barrier
;
949 spin_unlock(&ino
->i_lock
);
952 bool pnfs_roc_drain(struct inode
*ino
, u32
*barrier
, struct rpc_task
*task
)
954 struct nfs_inode
*nfsi
= NFS_I(ino
);
955 struct pnfs_layout_hdr
*lo
;
956 struct pnfs_layout_segment
*lseg
;
960 spin_lock(&ino
->i_lock
);
961 list_for_each_entry(lseg
, &nfsi
->layout
->plh_segs
, pls_list
)
962 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
963 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
968 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
970 /* Since close does not return a layout stateid for use as
971 * a barrier, we choose the worst-case barrier.
973 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
975 spin_unlock(&ino
->i_lock
);
980 * Compare two layout segments for sorting into layout cache.
981 * We want to preferentially return RW over RO layouts, so ensure those
985 cmp_layout(struct pnfs_layout_range
*l1
,
986 struct pnfs_layout_range
*l2
)
990 /* high offset > low offset */
991 d
= l1
->offset
- l2
->offset
;
995 /* short length > long length */
996 d
= l2
->length
- l1
->length
;
1000 /* read > read/write */
1001 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1005 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1006 struct pnfs_layout_segment
*lseg
)
1008 struct pnfs_layout_segment
*lp
;
1010 dprintk("%s:Begin\n", __func__
);
1012 list_for_each_entry(lp
, &lo
->plh_segs
, pls_list
) {
1013 if (cmp_layout(&lseg
->pls_range
, &lp
->pls_range
) > 0)
1015 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1016 dprintk("%s: inserted lseg %p "
1017 "iomode %d offset %llu length %llu before "
1018 "lp %p iomode %d offset %llu length %llu\n",
1019 __func__
, lseg
, lseg
->pls_range
.iomode
,
1020 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1021 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1022 lp
->pls_range
.length
);
1025 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1026 dprintk("%s: inserted lseg %p "
1027 "iomode %d offset %llu length %llu at tail\n",
1028 __func__
, lseg
, lseg
->pls_range
.iomode
,
1029 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1031 pnfs_get_layout_hdr(lo
);
1033 dprintk("%s:Return\n", __func__
);
1036 static struct pnfs_layout_hdr
*
1037 alloc_init_layout_hdr(struct inode
*ino
,
1038 struct nfs_open_context
*ctx
,
1041 struct pnfs_layout_hdr
*lo
;
1043 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1046 atomic_set(&lo
->plh_refcount
, 1);
1047 INIT_LIST_HEAD(&lo
->plh_layouts
);
1048 INIT_LIST_HEAD(&lo
->plh_segs
);
1049 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1050 lo
->plh_inode
= ino
;
1051 lo
->plh_lc_cred
= get_rpccred(ctx
->state
->owner
->so_cred
);
1055 static struct pnfs_layout_hdr
*
1056 pnfs_find_alloc_layout(struct inode
*ino
,
1057 struct nfs_open_context
*ctx
,
1060 struct nfs_inode
*nfsi
= NFS_I(ino
);
1061 struct pnfs_layout_hdr
*new = NULL
;
1063 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1065 if (nfsi
->layout
!= NULL
)
1067 spin_unlock(&ino
->i_lock
);
1068 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1069 spin_lock(&ino
->i_lock
);
1071 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1074 } else if (new != NULL
)
1075 pnfs_free_layout_hdr(new);
1077 pnfs_get_layout_hdr(nfsi
->layout
);
1078 return nfsi
->layout
;
1082 * iomode matching rules:
1093 is_matching_lseg(struct pnfs_layout_range
*ls_range
,
1094 struct pnfs_layout_range
*range
)
1096 struct pnfs_layout_range range1
;
1098 if ((range
->iomode
== IOMODE_RW
&&
1099 ls_range
->iomode
!= IOMODE_RW
) ||
1100 !lo_seg_intersecting(ls_range
, range
))
1103 /* range1 covers only the first byte in the range */
1106 return lo_seg_contained(ls_range
, &range1
);
1110 * lookup range in layout
1112 static struct pnfs_layout_segment
*
1113 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1114 struct pnfs_layout_range
*range
)
1116 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1118 dprintk("%s:Begin\n", __func__
);
1120 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1121 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1122 is_matching_lseg(&lseg
->pls_range
, range
)) {
1123 ret
= pnfs_get_lseg(lseg
);
1126 if (lseg
->pls_range
.offset
> range
->offset
)
1130 dprintk("%s:Return lseg %p ref %d\n",
1131 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
1136 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1137 * to the MDS or over pNFS
1139 * The nfs_inode read_io and write_io fields are cumulative counters reset
1140 * when there are no layout segments. Note that in pnfs_update_layout iomode
1141 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1144 * A return of true means use MDS I/O.
1147 * If a file's size is smaller than the file size threshold, data accesses
1148 * SHOULD be sent to the metadata server. If an I/O request has a length that
1149 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1150 * server. If both file size and I/O size are provided, the client SHOULD
1151 * reach or exceed both thresholds before sending its read or write
1152 * requests to the data server.
1154 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1155 struct inode
*ino
, int iomode
)
1157 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1158 struct nfs_inode
*nfsi
= NFS_I(ino
);
1159 loff_t fsize
= i_size_read(ino
);
1160 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1165 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1166 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1170 if (t
->bm
& THRESHOLD_RD
) {
1171 dprintk("%s fsize %llu\n", __func__
, fsize
);
1173 if (fsize
< t
->rd_sz
)
1176 if (t
->bm
& THRESHOLD_RD_IO
) {
1177 dprintk("%s nfsi->read_io %llu\n", __func__
,
1180 if (nfsi
->read_io
< t
->rd_io_sz
)
1185 if (t
->bm
& THRESHOLD_WR
) {
1186 dprintk("%s fsize %llu\n", __func__
, fsize
);
1188 if (fsize
< t
->wr_sz
)
1191 if (t
->bm
& THRESHOLD_WR_IO
) {
1192 dprintk("%s nfsi->write_io %llu\n", __func__
,
1195 if (nfsi
->write_io
< t
->wr_io_sz
)
1200 if (size_set
&& io_set
) {
1203 } else if (size
|| io
)
1206 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1211 * Layout segment is retreived from the server if not cached.
1212 * The appropriate layout segment is referenced and returned to the caller.
1214 struct pnfs_layout_segment
*
1215 pnfs_update_layout(struct inode
*ino
,
1216 struct nfs_open_context
*ctx
,
1219 enum pnfs_iomode iomode
,
1222 struct pnfs_layout_range arg
= {
1228 struct nfs_server
*server
= NFS_SERVER(ino
);
1229 struct nfs_client
*clp
= server
->nfs_client
;
1230 struct pnfs_layout_hdr
*lo
;
1231 struct pnfs_layout_segment
*lseg
= NULL
;
1234 if (!pnfs_enabled_sb(NFS_SERVER(ino
)))
1237 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
))
1240 spin_lock(&ino
->i_lock
);
1241 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1243 spin_unlock(&ino
->i_lock
);
1247 /* Do we even need to bother with this? */
1248 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1249 dprintk("%s matches recall, use MDS\n", __func__
);
1253 /* if LAYOUTGET already failed once we don't try again */
1254 if (pnfs_layout_io_test_failed(lo
, iomode
))
1257 /* Check to see if the layout for the given range already exists */
1258 lseg
= pnfs_find_lseg(lo
, &arg
);
1262 if (pnfs_layoutgets_blocked(lo
, 0))
1264 atomic_inc(&lo
->plh_outstanding
);
1266 first
= list_empty(&lo
->plh_layouts
) ? true : false;
1267 spin_unlock(&ino
->i_lock
);
1270 /* The lo must be on the clp list if there is any
1271 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1273 spin_lock(&clp
->cl_lock
);
1274 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1275 spin_unlock(&clp
->cl_lock
);
1278 pg_offset
= arg
.offset
& ~PAGE_CACHE_MASK
;
1280 arg
.offset
-= pg_offset
;
1281 arg
.length
+= pg_offset
;
1283 if (arg
.length
!= NFS4_MAX_UINT64
)
1284 arg
.length
= PAGE_CACHE_ALIGN(arg
.length
);
1286 lseg
= send_layoutget(lo
, ctx
, &arg
, gfp_flags
);
1287 atomic_dec(&lo
->plh_outstanding
);
1289 pnfs_put_layout_hdr(lo
);
1291 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1292 "(%s, offset: %llu, length: %llu)\n",
1293 __func__
, ino
->i_sb
->s_id
,
1294 (unsigned long long)NFS_FILEID(ino
),
1295 lseg
== NULL
? "not found" : "found",
1296 iomode
==IOMODE_RW
? "read/write" : "read-only",
1297 (unsigned long long)pos
,
1298 (unsigned long long)count
);
1301 spin_unlock(&ino
->i_lock
);
1302 goto out_put_layout_hdr
;
1304 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1306 struct pnfs_layout_segment
*
1307 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1309 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1310 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1311 struct pnfs_layout_segment
*lseg
;
1312 struct inode
*ino
= lo
->plh_inode
;
1315 /* Inject layout blob into I/O device driver */
1316 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1317 if (!lseg
|| IS_ERR(lseg
)) {
1321 status
= PTR_ERR(lseg
);
1322 dprintk("%s: Could not allocate layout: error %d\n",
1327 spin_lock(&ino
->i_lock
);
1328 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1329 dprintk("%s forget reply due to recall\n", __func__
);
1330 goto out_forget_reply
;
1333 if (pnfs_layoutgets_blocked(lo
, 1) ||
1334 pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
1335 dprintk("%s forget reply due to state\n", __func__
);
1336 goto out_forget_reply
;
1339 /* Done processing layoutget. Set the layout stateid */
1340 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1342 init_lseg(lo
, lseg
);
1343 lseg
->pls_range
= res
->range
;
1344 pnfs_get_lseg(lseg
);
1345 pnfs_layout_insert_lseg(lo
, lseg
);
1347 if (res
->return_on_close
) {
1348 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1349 set_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
);
1352 spin_unlock(&ino
->i_lock
);
1355 return ERR_PTR(status
);
1358 spin_unlock(&ino
->i_lock
);
1359 lseg
->pls_layout
= lo
;
1360 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1365 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1367 u64 rd_size
= req
->wb_bytes
;
1369 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
1371 if (req
->wb_offset
!= req
->wb_pgbase
) {
1372 nfs_pageio_reset_read_mds(pgio
);
1376 if (pgio
->pg_dreq
== NULL
)
1377 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
1379 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
1381 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1387 /* If no lseg, fall back to read through mds */
1388 if (pgio
->pg_lseg
== NULL
)
1389 nfs_pageio_reset_read_mds(pgio
);
1392 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1395 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
1396 struct nfs_page
*req
, u64 wb_size
)
1398 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
1400 if (req
->wb_offset
!= req
->wb_pgbase
) {
1401 nfs_pageio_reset_write_mds(pgio
);
1405 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1411 /* If no lseg, fall back to write through mds */
1412 if (pgio
->pg_lseg
== NULL
)
1413 nfs_pageio_reset_write_mds(pgio
);
1415 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1418 pnfs_pageio_init_read(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1419 const struct nfs_pgio_completion_ops
*compl_ops
)
1421 struct nfs_server
*server
= NFS_SERVER(inode
);
1422 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1425 nfs_pageio_init_read(pgio
, inode
, compl_ops
);
1427 nfs_pageio_init(pgio
, inode
, ld
->pg_read_ops
, compl_ops
, server
->rsize
, 0);
1431 pnfs_pageio_init_write(struct nfs_pageio_descriptor
*pgio
, struct inode
*inode
,
1433 const struct nfs_pgio_completion_ops
*compl_ops
)
1435 struct nfs_server
*server
= NFS_SERVER(inode
);
1436 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
1439 nfs_pageio_init_write(pgio
, inode
, ioflags
, compl_ops
);
1441 nfs_pageio_init(pgio
, inode
, ld
->pg_write_ops
, compl_ops
, server
->wsize
, ioflags
);
1445 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
1446 struct nfs_page
*req
)
1448 if (pgio
->pg_lseg
== NULL
)
1449 return nfs_generic_pg_test(pgio
, prev
, req
);
1452 * Test if a nfs_page is fully contained in the pnfs_layout_range.
1453 * Note that this test makes several assumptions:
1454 * - that the previous nfs_page in the struct nfs_pageio_descriptor
1455 * is known to lie within the range.
1456 * - that the nfs_page being tested is known to be contiguous with the
1457 * previous nfs_page.
1458 * - Layout ranges are page aligned, so we only have to test the
1459 * start offset of the request.
1461 * Please also note that 'end_offset' is actually the offset of the
1462 * first byte that lies outside the pnfs_layout_range. FIXME?
1465 return req_offset(req
) < end_offset(pgio
->pg_lseg
->pls_range
.offset
,
1466 pgio
->pg_lseg
->pls_range
.length
);
1468 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
1470 int pnfs_write_done_resend_to_mds(struct inode
*inode
,
1471 struct list_head
*head
,
1472 const struct nfs_pgio_completion_ops
*compl_ops
,
1473 struct nfs_direct_req
*dreq
)
1475 struct nfs_pageio_descriptor pgio
;
1478 /* Resend all requests through the MDS */
1479 nfs_pageio_init_write(&pgio
, inode
, FLUSH_STABLE
, compl_ops
);
1480 pgio
.pg_dreq
= dreq
;
1481 while (!list_empty(head
)) {
1482 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1484 nfs_list_remove_request(req
);
1485 if (!nfs_pageio_add_request(&pgio
, req
))
1486 nfs_list_add_request(req
, &failed
);
1488 nfs_pageio_complete(&pgio
);
1490 if (!list_empty(&failed
)) {
1491 /* For some reason our attempt to resend pages. Mark the
1492 * overall send request as having failed, and let
1493 * nfs_writeback_release_full deal with the error.
1495 list_move(&failed
, head
);
1500 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
1502 static void pnfs_ld_handle_write_error(struct nfs_write_data
*data
)
1504 struct nfs_pgio_header
*hdr
= data
->header
;
1506 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
1507 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1508 PNFS_LAYOUTRET_ON_ERROR
) {
1509 pnfs_return_layout(hdr
->inode
);
1511 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1512 data
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
->inode
,
1514 hdr
->completion_ops
,
1519 * Called by non rpc-based layout drivers
1521 void pnfs_ld_write_done(struct nfs_write_data
*data
)
1523 struct nfs_pgio_header
*hdr
= data
->header
;
1525 if (!hdr
->pnfs_error
) {
1526 pnfs_set_layoutcommit(data
);
1527 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1529 pnfs_ld_handle_write_error(data
);
1530 hdr
->mds_ops
->rpc_release(data
);
1532 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
1535 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
1536 struct nfs_write_data
*data
)
1538 struct nfs_pgio_header
*hdr
= data
->header
;
1540 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1541 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1542 nfs_pageio_reset_write_mds(desc
);
1543 desc
->pg_recoalesce
= 1;
1545 nfs_writedata_release(data
);
1548 static enum pnfs_try_status
1549 pnfs_try_to_write_data(struct nfs_write_data
*wdata
,
1550 const struct rpc_call_ops
*call_ops
,
1551 struct pnfs_layout_segment
*lseg
,
1554 struct nfs_pgio_header
*hdr
= wdata
->header
;
1555 struct inode
*inode
= hdr
->inode
;
1556 enum pnfs_try_status trypnfs
;
1557 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1559 hdr
->mds_ops
= call_ops
;
1561 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
1562 inode
->i_ino
, wdata
->args
.count
, wdata
->args
.offset
, how
);
1563 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(wdata
, how
);
1564 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1565 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
1566 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1571 pnfs_do_multiple_writes(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
, int how
)
1573 struct nfs_write_data
*data
;
1574 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1575 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1577 desc
->pg_lseg
= NULL
;
1578 while (!list_empty(head
)) {
1579 enum pnfs_try_status trypnfs
;
1581 data
= list_first_entry(head
, struct nfs_write_data
, list
);
1582 list_del_init(&data
->list
);
1584 trypnfs
= pnfs_try_to_write_data(data
, call_ops
, lseg
, how
);
1585 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1586 pnfs_write_through_mds(desc
, data
);
1588 pnfs_put_lseg(lseg
);
1591 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
1593 pnfs_put_lseg(hdr
->lseg
);
1594 nfs_writehdr_free(hdr
);
1596 EXPORT_SYMBOL_GPL(pnfs_writehdr_free
);
1599 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1601 struct nfs_write_header
*whdr
;
1602 struct nfs_pgio_header
*hdr
;
1605 whdr
= nfs_writehdr_alloc();
1607 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1608 pnfs_put_lseg(desc
->pg_lseg
);
1609 desc
->pg_lseg
= NULL
;
1612 hdr
= &whdr
->header
;
1613 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
1614 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1615 atomic_inc(&hdr
->refcnt
);
1616 ret
= nfs_generic_flush(desc
, hdr
);
1618 pnfs_put_lseg(desc
->pg_lseg
);
1619 desc
->pg_lseg
= NULL
;
1621 pnfs_do_multiple_writes(desc
, &hdr
->rpc_list
, desc
->pg_ioflags
);
1622 if (atomic_dec_and_test(&hdr
->refcnt
))
1623 hdr
->completion_ops
->completion(hdr
);
1626 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
1628 int pnfs_read_done_resend_to_mds(struct inode
*inode
,
1629 struct list_head
*head
,
1630 const struct nfs_pgio_completion_ops
*compl_ops
,
1631 struct nfs_direct_req
*dreq
)
1633 struct nfs_pageio_descriptor pgio
;
1636 /* Resend all requests through the MDS */
1637 nfs_pageio_init_read(&pgio
, inode
, compl_ops
);
1638 pgio
.pg_dreq
= dreq
;
1639 while (!list_empty(head
)) {
1640 struct nfs_page
*req
= nfs_list_entry(head
->next
);
1642 nfs_list_remove_request(req
);
1643 if (!nfs_pageio_add_request(&pgio
, req
))
1644 nfs_list_add_request(req
, &failed
);
1646 nfs_pageio_complete(&pgio
);
1648 if (!list_empty(&failed
)) {
1649 list_move(&failed
, head
);
1654 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
1656 static void pnfs_ld_handle_read_error(struct nfs_read_data
*data
)
1658 struct nfs_pgio_header
*hdr
= data
->header
;
1660 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
1661 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1662 PNFS_LAYOUTRET_ON_ERROR
) {
1663 pnfs_return_layout(hdr
->inode
);
1665 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1666 data
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
->inode
,
1668 hdr
->completion_ops
,
1673 * Called by non rpc-based layout drivers
1675 void pnfs_ld_read_done(struct nfs_read_data
*data
)
1677 struct nfs_pgio_header
*hdr
= data
->header
;
1679 if (likely(!hdr
->pnfs_error
)) {
1680 __nfs4_read_done_cb(data
);
1681 hdr
->mds_ops
->rpc_call_done(&data
->task
, data
);
1683 pnfs_ld_handle_read_error(data
);
1684 hdr
->mds_ops
->rpc_release(data
);
1686 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
1689 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
1690 struct nfs_read_data
*data
)
1692 struct nfs_pgio_header
*hdr
= data
->header
;
1694 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1695 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1696 nfs_pageio_reset_read_mds(desc
);
1697 desc
->pg_recoalesce
= 1;
1699 nfs_readdata_release(data
);
1703 * Call the appropriate parallel I/O subsystem read function.
1705 static enum pnfs_try_status
1706 pnfs_try_to_read_data(struct nfs_read_data
*rdata
,
1707 const struct rpc_call_ops
*call_ops
,
1708 struct pnfs_layout_segment
*lseg
)
1710 struct nfs_pgio_header
*hdr
= rdata
->header
;
1711 struct inode
*inode
= hdr
->inode
;
1712 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1713 enum pnfs_try_status trypnfs
;
1715 hdr
->mds_ops
= call_ops
;
1717 dprintk("%s: Reading ino:%lu %u@%llu\n",
1718 __func__
, inode
->i_ino
, rdata
->args
.count
, rdata
->args
.offset
);
1720 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(rdata
);
1721 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1722 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
1723 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1728 pnfs_do_multiple_reads(struct nfs_pageio_descriptor
*desc
, struct list_head
*head
)
1730 struct nfs_read_data
*data
;
1731 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1732 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1734 desc
->pg_lseg
= NULL
;
1735 while (!list_empty(head
)) {
1736 enum pnfs_try_status trypnfs
;
1738 data
= list_first_entry(head
, struct nfs_read_data
, list
);
1739 list_del_init(&data
->list
);
1741 trypnfs
= pnfs_try_to_read_data(data
, call_ops
, lseg
);
1742 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1743 pnfs_read_through_mds(desc
, data
);
1745 pnfs_put_lseg(lseg
);
1748 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
1750 pnfs_put_lseg(hdr
->lseg
);
1751 nfs_readhdr_free(hdr
);
1753 EXPORT_SYMBOL_GPL(pnfs_readhdr_free
);
1756 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
1758 struct nfs_read_header
*rhdr
;
1759 struct nfs_pgio_header
*hdr
;
1762 rhdr
= nfs_readhdr_alloc();
1764 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1766 pnfs_put_lseg(desc
->pg_lseg
);
1767 desc
->pg_lseg
= NULL
;
1770 hdr
= &rhdr
->header
;
1771 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
1772 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1773 atomic_inc(&hdr
->refcnt
);
1774 ret
= nfs_generic_pagein(desc
, hdr
);
1776 pnfs_put_lseg(desc
->pg_lseg
);
1777 desc
->pg_lseg
= NULL
;
1779 pnfs_do_multiple_reads(desc
, &hdr
->rpc_list
);
1780 if (atomic_dec_and_test(&hdr
->refcnt
))
1781 hdr
->completion_ops
->completion(hdr
);
1784 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
1787 * There can be multiple RW segments.
1789 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
1791 struct pnfs_layout_segment
*lseg
;
1793 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
1794 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
1795 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1796 list_add(&lseg
->pls_lc_list
, listp
);
1800 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
1802 struct pnfs_layout_segment
*lseg
, *tmp
;
1803 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
1805 /* Matched by references in pnfs_set_layoutcommit */
1806 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
1807 list_del_init(&lseg
->pls_lc_list
);
1808 pnfs_put_lseg(lseg
);
1811 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
1812 smp_mb__after_clear_bit();
1813 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
1816 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
1818 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
1820 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
1823 pnfs_set_layoutcommit(struct nfs_write_data
*wdata
)
1825 struct nfs_pgio_header
*hdr
= wdata
->header
;
1826 struct inode
*inode
= hdr
->inode
;
1827 struct nfs_inode
*nfsi
= NFS_I(inode
);
1828 loff_t end_pos
= wdata
->mds_offset
+ wdata
->res
.count
;
1829 bool mark_as_dirty
= false;
1831 spin_lock(&inode
->i_lock
);
1832 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1833 mark_as_dirty
= true;
1834 dprintk("%s: Set layoutcommit for inode %lu ",
1835 __func__
, inode
->i_ino
);
1837 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &hdr
->lseg
->pls_flags
)) {
1838 /* references matched in nfs4_layoutcommit_release */
1839 pnfs_get_lseg(hdr
->lseg
);
1841 if (end_pos
> nfsi
->layout
->plh_lwb
)
1842 nfsi
->layout
->plh_lwb
= end_pos
;
1843 spin_unlock(&inode
->i_lock
);
1844 dprintk("%s: lseg %p end_pos %llu\n",
1845 __func__
, hdr
->lseg
, nfsi
->layout
->plh_lwb
);
1847 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1848 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1850 mark_inode_dirty_sync(inode
);
1852 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
1854 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
1856 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
1858 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
1859 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
1860 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
1864 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1865 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1866 * data to disk to allow the server to recover the data if it crashes.
1867 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1868 * is off, and a COMMIT is sent to a data server, or
1869 * if WRITEs to a data server return NFS_DATA_SYNC.
1872 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
1874 struct nfs4_layoutcommit_data
*data
;
1875 struct nfs_inode
*nfsi
= NFS_I(inode
);
1879 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
1881 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1884 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1885 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
1891 if (!test_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1894 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
1899 status
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
,
1900 nfs_wait_bit_killable
, TASK_KILLABLE
);
1905 INIT_LIST_HEAD(&data
->lseg_list
);
1906 spin_lock(&inode
->i_lock
);
1907 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1908 clear_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
);
1909 spin_unlock(&inode
->i_lock
);
1910 wake_up_bit(&nfsi
->flags
, NFS_INO_LAYOUTCOMMITTING
);
1914 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
1916 end_pos
= nfsi
->layout
->plh_lwb
;
1917 nfsi
->layout
->plh_lwb
= 0;
1919 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
1920 spin_unlock(&inode
->i_lock
);
1922 data
->args
.inode
= inode
;
1923 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
1924 nfs_fattr_init(&data
->fattr
);
1925 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
1926 data
->res
.fattr
= &data
->fattr
;
1927 data
->args
.lastbytewritten
= end_pos
- 1;
1928 data
->res
.server
= NFS_SERVER(inode
);
1930 status
= nfs4_proc_layoutcommit(data
, sync
);
1933 mark_inode_dirty_sync(inode
);
1934 dprintk("<-- %s status %d\n", __func__
, status
);
1941 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
1943 struct nfs4_threshold
*thp
;
1945 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
1947 dprintk("%s mdsthreshold allocation failed\n", __func__
);