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>
36 #include "nfs4trace.h"
38 #define NFSDBG_FACILITY NFSDBG_PNFS
39 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
44 * protects pnfs_modules_tbl.
46 static DEFINE_SPINLOCK(pnfs_spinlock
);
49 * pnfs_modules_tbl holds all pnfs modules
51 static LIST_HEAD(pnfs_modules_tbl
);
53 /* Return the registered pnfs layout driver module matching given id */
54 static struct pnfs_layoutdriver_type
*
55 find_pnfs_driver_locked(u32 id
)
57 struct pnfs_layoutdriver_type
*local
;
59 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
64 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
68 static struct pnfs_layoutdriver_type
*
69 find_pnfs_driver(u32 id
)
71 struct pnfs_layoutdriver_type
*local
;
73 spin_lock(&pnfs_spinlock
);
74 local
= find_pnfs_driver_locked(id
);
75 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
76 dprintk("%s: Could not grab reference on module\n", __func__
);
79 spin_unlock(&pnfs_spinlock
);
84 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
86 if (nfss
->pnfs_curr_ld
) {
87 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
88 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
89 /* Decrement the MDS count. Purge the deviceid cache if zero */
90 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
91 nfs4_deviceid_purge_client(nfss
->nfs_client
);
92 module_put(nfss
->pnfs_curr_ld
->owner
);
94 nfss
->pnfs_curr_ld
= NULL
;
98 * Try to set the server's pnfs module to the pnfs layout type specified by id.
99 * Currently only one pNFS layout driver per filesystem is supported.
101 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
104 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
107 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
111 if (!(server
->nfs_client
->cl_exchange_flags
&
112 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
113 printk(KERN_ERR
"NFS: %s: id %u cl_exchange_flags 0x%x\n",
114 __func__
, id
, server
->nfs_client
->cl_exchange_flags
);
117 ld_type
= find_pnfs_driver(id
);
119 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
, id
);
120 ld_type
= find_pnfs_driver(id
);
122 dprintk("%s: No pNFS module found for %u.\n",
127 server
->pnfs_curr_ld
= ld_type
;
128 if (ld_type
->set_layoutdriver
129 && ld_type
->set_layoutdriver(server
, mntfh
)) {
130 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
131 "driver %u.\n", __func__
, id
);
132 module_put(ld_type
->owner
);
135 /* Bump the MDS count */
136 atomic_inc(&server
->nfs_client
->cl_mds_count
);
138 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
142 dprintk("%s: Using NFSv4 I/O\n", __func__
);
143 server
->pnfs_curr_ld
= NULL
;
147 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
149 int status
= -EINVAL
;
150 struct pnfs_layoutdriver_type
*tmp
;
152 if (ld_type
->id
== 0) {
153 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
156 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
157 printk(KERN_ERR
"NFS: %s Layout driver must provide "
158 "alloc_lseg and free_lseg.\n", __func__
);
162 spin_lock(&pnfs_spinlock
);
163 tmp
= find_pnfs_driver_locked(ld_type
->id
);
165 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
167 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
170 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
171 __func__
, ld_type
->id
);
173 spin_unlock(&pnfs_spinlock
);
177 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
180 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
182 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
183 spin_lock(&pnfs_spinlock
);
184 list_del(&ld_type
->pnfs_tblid
);
185 spin_unlock(&pnfs_spinlock
);
187 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
190 * pNFS client layout cache
193 /* Need to hold i_lock if caller does not already hold reference */
195 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
197 atomic_inc(&lo
->plh_refcount
);
200 static struct pnfs_layout_hdr
*
201 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
203 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
204 return ld
->alloc_layout_hdr(ino
, 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
->free_layout_hdr(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
);
326 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
327 struct pnfs_layout_segment
*lseg
)
329 struct inode
*inode
= lo
->plh_inode
;
331 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
332 list_del_init(&lseg
->pls_list
);
333 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
334 atomic_dec(&lo
->plh_refcount
);
335 if (list_empty(&lo
->plh_segs
))
336 clear_bit(NFS_LAYOUT_BULK_RECALL
, &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_get_layout_hdr(lo
);
356 pnfs_layout_remove_lseg(lo
, lseg
);
357 spin_unlock(&inode
->i_lock
);
358 pnfs_free_lseg(lseg
);
359 pnfs_put_layout_hdr(lo
);
362 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
364 static void pnfs_put_lseg_async_work(struct work_struct
*work
)
366 struct pnfs_layout_segment
*lseg
;
368 lseg
= container_of(work
, struct pnfs_layout_segment
, pls_work
);
374 pnfs_put_lseg_async(struct pnfs_layout_segment
*lseg
)
376 INIT_WORK(&lseg
->pls_work
, pnfs_put_lseg_async_work
);
377 schedule_work(&lseg
->pls_work
);
379 EXPORT_SYMBOL_GPL(pnfs_put_lseg_async
);
382 end_offset(u64 start
, u64 len
)
387 return end
>= start
? end
: NFS4_MAX_UINT64
;
391 * is l2 fully contained in l1?
393 * [----------------------------------)
398 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
399 const struct pnfs_layout_range
*l2
)
401 u64 start1
= l1
->offset
;
402 u64 end1
= end_offset(start1
, l1
->length
);
403 u64 start2
= l2
->offset
;
404 u64 end2
= end_offset(start2
, l2
->length
);
406 return (start1
<= start2
) && (end1
>= end2
);
410 * is l1 and l2 intersecting?
412 * [----------------------------------)
417 pnfs_lseg_range_intersecting(const struct pnfs_layout_range
*l1
,
418 const struct pnfs_layout_range
*l2
)
420 u64 start1
= l1
->offset
;
421 u64 end1
= end_offset(start1
, l1
->length
);
422 u64 start2
= l2
->offset
;
423 u64 end2
= end_offset(start2
, l2
->length
);
425 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
426 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
430 should_free_lseg(const struct pnfs_layout_range
*lseg_range
,
431 const struct pnfs_layout_range
*recall_range
)
433 return (recall_range
->iomode
== IOMODE_ANY
||
434 lseg_range
->iomode
== recall_range
->iomode
) &&
435 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
438 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
439 struct list_head
*tmp_list
)
441 if (!atomic_dec_and_test(&lseg
->pls_refcount
))
443 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
444 list_add(&lseg
->pls_list
, tmp_list
);
448 /* Returns 1 if lseg is removed from list, 0 otherwise */
449 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
450 struct list_head
*tmp_list
)
454 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
455 /* Remove the reference keeping the lseg in the
456 * list. It will now be removed when all
457 * outstanding io is finished.
459 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
460 atomic_read(&lseg
->pls_refcount
));
461 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
467 /* Returns count of number of matching invalid lsegs remaining in list
471 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
472 struct list_head
*tmp_list
,
473 struct pnfs_layout_range
*recall_range
)
475 struct pnfs_layout_segment
*lseg
, *next
;
476 int invalid
= 0, removed
= 0;
478 dprintk("%s:Begin lo %p\n", __func__
, lo
);
480 if (list_empty(&lo
->plh_segs
))
482 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
484 should_free_lseg(&lseg
->pls_range
, recall_range
)) {
485 dprintk("%s: freeing lseg %p iomode %d "
486 "offset %llu length %llu\n", __func__
,
487 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_range
.offset
,
488 lseg
->pls_range
.length
);
490 removed
+= mark_lseg_invalid(lseg
, tmp_list
);
492 dprintk("%s:Return %i\n", __func__
, invalid
- removed
);
493 return invalid
- removed
;
496 /* note free_me must contain lsegs from a single layout_hdr */
498 pnfs_free_lseg_list(struct list_head
*free_me
)
500 struct pnfs_layout_segment
*lseg
, *tmp
;
502 if (list_empty(free_me
))
505 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
506 list_del(&lseg
->pls_list
);
507 pnfs_free_lseg(lseg
);
512 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
514 struct pnfs_layout_hdr
*lo
;
517 spin_lock(&nfsi
->vfs_inode
.i_lock
);
520 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
521 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
522 pnfs_get_layout_hdr(lo
);
523 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
524 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
525 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
526 pnfs_free_lseg_list(&tmp_list
);
527 pnfs_put_layout_hdr(lo
);
529 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
531 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
534 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
535 struct list_head
*layout_list
)
537 struct pnfs_layout_hdr
*lo
;
540 spin_lock(&inode
->i_lock
);
541 lo
= NFS_I(inode
)->layout
;
542 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
543 pnfs_get_layout_hdr(lo
);
544 list_add(&lo
->plh_bulk_destroy
, layout_list
);
547 spin_unlock(&inode
->i_lock
);
551 /* Caller must hold rcu_read_lock and clp->cl_lock */
553 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
554 struct nfs_server
*server
,
555 struct list_head
*layout_list
)
557 struct pnfs_layout_hdr
*lo
, *next
;
560 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
561 inode
= igrab(lo
->plh_inode
);
564 list_del_init(&lo
->plh_layouts
);
565 if (pnfs_layout_add_bulk_destroy_list(inode
, layout_list
))
568 spin_unlock(&clp
->cl_lock
);
570 spin_lock(&clp
->cl_lock
);
578 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
581 struct pnfs_layout_hdr
*lo
;
583 struct pnfs_layout_range range
= {
584 .iomode
= IOMODE_ANY
,
586 .length
= NFS4_MAX_UINT64
,
588 LIST_HEAD(lseg_list
);
591 while (!list_empty(layout_list
)) {
592 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
594 dprintk("%s freeing layout for inode %lu\n", __func__
,
595 lo
->plh_inode
->i_ino
);
596 inode
= lo
->plh_inode
;
597 spin_lock(&inode
->i_lock
);
598 list_del_init(&lo
->plh_bulk_destroy
);
599 lo
->plh_block_lgets
++; /* permanently block new LAYOUTGETs */
601 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
602 if (pnfs_mark_matching_lsegs_invalid(lo
, &lseg_list
, &range
))
604 spin_unlock(&inode
->i_lock
);
605 pnfs_free_lseg_list(&lseg_list
);
606 pnfs_put_layout_hdr(lo
);
613 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
614 struct nfs_fsid
*fsid
,
617 struct nfs_server
*server
;
618 LIST_HEAD(layout_list
);
620 spin_lock(&clp
->cl_lock
);
623 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
624 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
626 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
632 spin_unlock(&clp
->cl_lock
);
634 if (list_empty(&layout_list
))
636 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
640 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
643 struct nfs_server
*server
;
644 LIST_HEAD(layout_list
);
646 spin_lock(&clp
->cl_lock
);
649 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
650 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
656 spin_unlock(&clp
->cl_lock
);
658 if (list_empty(&layout_list
))
660 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
664 * Called by the state manger to remove all layouts established under an
668 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
670 nfs4_deviceid_mark_client_invalid(clp
);
671 nfs4_deviceid_purge_client(clp
);
673 pnfs_destroy_layouts_byclid(clp
, false);
677 * Compare 2 layout stateid sequence ids, to see which is newer,
678 * taking into account wraparound issues.
680 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
682 return (s32
)(s1
- s2
) > 0;
686 pnfs_verify_layout_stateid(struct pnfs_layout_hdr
*lo
,
687 const nfs4_stateid
*new,
688 struct list_head
*free_me_list
)
690 if (nfs4_stateid_match_other(&lo
->plh_stateid
, new))
692 /* Layout is new! Kill existing layout segments */
693 pnfs_mark_matching_lsegs_invalid(lo
, free_me_list
, NULL
);
696 /* update lo->plh_stateid with new if is more recent */
698 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
701 u32 oldseq
, newseq
, new_barrier
;
702 int empty
= list_empty(&lo
->plh_segs
);
704 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
705 newseq
= be32_to_cpu(new->seqid
);
706 if (empty
|| pnfs_seqid_is_newer(newseq
, oldseq
)) {
707 nfs4_stateid_copy(&lo
->plh_stateid
, new);
708 if (update_barrier
) {
709 new_barrier
= be32_to_cpu(new->seqid
);
711 /* Because of wraparound, we want to keep the barrier
712 * "close" to the current seqids.
714 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
716 if (empty
|| pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
717 lo
->plh_barrier
= new_barrier
;
722 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
723 const nfs4_stateid
*stateid
)
725 u32 seqid
= be32_to_cpu(stateid
->seqid
);
727 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
730 /* lget is set to 1 if called from inside send_layoutget call chain */
732 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
, int lget
)
734 return lo
->plh_block_lgets
||
735 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
) ||
736 (list_empty(&lo
->plh_segs
) &&
737 (atomic_read(&lo
->plh_outstanding
) > lget
));
741 pnfs_choose_layoutget_stateid(nfs4_stateid
*dst
, struct pnfs_layout_hdr
*lo
,
742 struct nfs4_state
*open_state
)
746 dprintk("--> %s\n", __func__
);
747 spin_lock(&lo
->plh_inode
->i_lock
);
748 if (pnfs_layoutgets_blocked(lo
, 1)) {
750 } else if (!nfs4_valid_open_stateid(open_state
)) {
752 } else if (list_empty(&lo
->plh_segs
)) {
756 seq
= read_seqbegin(&open_state
->seqlock
);
757 nfs4_stateid_copy(dst
, &open_state
->stateid
);
758 } while (read_seqretry(&open_state
->seqlock
, seq
));
760 nfs4_stateid_copy(dst
, &lo
->plh_stateid
);
761 spin_unlock(&lo
->plh_inode
->i_lock
);
762 dprintk("<-- %s\n", __func__
);
767 * Get layout from server.
768 * for now, assume that whole file layouts are requested.
770 * arg->length: all ones
772 static struct pnfs_layout_segment
*
773 send_layoutget(struct pnfs_layout_hdr
*lo
,
774 struct nfs_open_context
*ctx
,
775 struct pnfs_layout_range
*range
,
778 struct inode
*ino
= lo
->plh_inode
;
779 struct nfs_server
*server
= NFS_SERVER(ino
);
780 struct nfs4_layoutget
*lgp
;
781 struct pnfs_layout_segment
*lseg
;
783 dprintk("--> %s\n", __func__
);
785 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
789 lgp
->args
.minlength
= PAGE_CACHE_SIZE
;
790 if (lgp
->args
.minlength
> range
->length
)
791 lgp
->args
.minlength
= range
->length
;
792 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
793 lgp
->args
.range
= *range
;
794 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
795 lgp
->args
.inode
= ino
;
796 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
797 lgp
->gfp_flags
= gfp_flags
;
798 lgp
->cred
= lo
->plh_lc_cred
;
800 /* Synchronously retrieve layout information from server and
803 lseg
= nfs4_proc_layoutget(lgp
, gfp_flags
);
805 switch (PTR_ERR(lseg
)) {
810 /* remember that LAYOUTGET failed and suspend trying */
811 pnfs_layout_io_set_failed(lo
, range
->iomode
);
819 static void pnfs_clear_layoutcommit(struct inode
*inode
,
820 struct list_head
*head
)
822 struct nfs_inode
*nfsi
= NFS_I(inode
);
823 struct pnfs_layout_segment
*lseg
, *tmp
;
825 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
827 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
828 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
830 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
835 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
836 * when the layout segment list is empty.
838 * Note that a pnfs_layout_hdr can exist with an empty layout segment
839 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
840 * deviceid is marked invalid.
843 _pnfs_return_layout(struct inode
*ino
)
845 struct pnfs_layout_hdr
*lo
= NULL
;
846 struct nfs_inode
*nfsi
= NFS_I(ino
);
848 struct nfs4_layoutreturn
*lrp
;
849 nfs4_stateid stateid
;
850 int status
= 0, empty
;
852 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
854 spin_lock(&ino
->i_lock
);
857 spin_unlock(&ino
->i_lock
);
858 dprintk("NFS: %s no layout to return\n", __func__
);
861 stateid
= nfsi
->layout
->plh_stateid
;
862 /* Reference matched in nfs4_layoutreturn_release */
863 pnfs_get_layout_hdr(lo
);
864 empty
= list_empty(&lo
->plh_segs
);
865 pnfs_clear_layoutcommit(ino
, &tmp_list
);
866 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
);
867 /* Don't send a LAYOUTRETURN if list was initially empty */
869 spin_unlock(&ino
->i_lock
);
870 pnfs_put_layout_hdr(lo
);
871 dprintk("NFS: %s no layout segments to return\n", __func__
);
874 lo
->plh_block_lgets
++;
875 spin_unlock(&ino
->i_lock
);
876 pnfs_free_lseg_list(&tmp_list
);
878 lrp
= kzalloc(sizeof(*lrp
), GFP_KERNEL
);
879 if (unlikely(lrp
== NULL
)) {
881 spin_lock(&ino
->i_lock
);
882 lo
->plh_block_lgets
--;
883 spin_unlock(&ino
->i_lock
);
884 pnfs_put_layout_hdr(lo
);
888 lrp
->args
.stateid
= stateid
;
889 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
890 lrp
->args
.inode
= ino
;
891 lrp
->args
.layout
= lo
;
892 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
893 lrp
->cred
= lo
->plh_lc_cred
;
895 status
= nfs4_proc_layoutreturn(lrp
);
897 dprintk("<-- %s status: %d\n", __func__
, status
);
900 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
903 pnfs_commit_and_return_layout(struct inode
*inode
)
905 struct pnfs_layout_hdr
*lo
;
908 spin_lock(&inode
->i_lock
);
909 lo
= NFS_I(inode
)->layout
;
911 spin_unlock(&inode
->i_lock
);
914 pnfs_get_layout_hdr(lo
);
915 /* Block new layoutgets and read/write to ds */
916 lo
->plh_block_lgets
++;
917 spin_unlock(&inode
->i_lock
);
918 filemap_fdatawait(inode
->i_mapping
);
919 ret
= pnfs_layoutcommit_inode(inode
, true);
921 ret
= _pnfs_return_layout(inode
);
922 spin_lock(&inode
->i_lock
);
923 lo
->plh_block_lgets
--;
924 spin_unlock(&inode
->i_lock
);
925 pnfs_put_layout_hdr(lo
);
929 bool pnfs_roc(struct inode
*ino
)
931 struct pnfs_layout_hdr
*lo
;
932 struct pnfs_layout_segment
*lseg
, *tmp
;
936 spin_lock(&ino
->i_lock
);
937 lo
= NFS_I(ino
)->layout
;
938 if (!lo
|| !test_and_clear_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
) ||
939 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
941 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
942 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
943 mark_lseg_invalid(lseg
, &tmp_list
);
948 lo
->plh_block_lgets
++;
949 pnfs_get_layout_hdr(lo
); /* matched in pnfs_roc_release */
950 spin_unlock(&ino
->i_lock
);
951 pnfs_free_lseg_list(&tmp_list
);
955 spin_unlock(&ino
->i_lock
);
959 void pnfs_roc_release(struct inode
*ino
)
961 struct pnfs_layout_hdr
*lo
;
963 spin_lock(&ino
->i_lock
);
964 lo
= NFS_I(ino
)->layout
;
965 lo
->plh_block_lgets
--;
966 if (atomic_dec_and_test(&lo
->plh_refcount
)) {
967 pnfs_detach_layout_hdr(lo
);
968 spin_unlock(&ino
->i_lock
);
969 pnfs_free_layout_hdr(lo
);
971 spin_unlock(&ino
->i_lock
);
974 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
976 struct pnfs_layout_hdr
*lo
;
978 spin_lock(&ino
->i_lock
);
979 lo
= NFS_I(ino
)->layout
;
980 if (pnfs_seqid_is_newer(barrier
, lo
->plh_barrier
))
981 lo
->plh_barrier
= barrier
;
982 spin_unlock(&ino
->i_lock
);
985 bool pnfs_roc_drain(struct inode
*ino
, u32
*barrier
, struct rpc_task
*task
)
987 struct nfs_inode
*nfsi
= NFS_I(ino
);
988 struct pnfs_layout_hdr
*lo
;
989 struct pnfs_layout_segment
*lseg
;
993 spin_lock(&ino
->i_lock
);
994 list_for_each_entry(lseg
, &nfsi
->layout
->plh_segs
, pls_list
)
995 if (test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
996 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1001 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
1003 /* Since close does not return a layout stateid for use as
1004 * a barrier, we choose the worst-case barrier.
1006 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
1008 spin_unlock(&ino
->i_lock
);
1013 * Compare two layout segments for sorting into layout cache.
1014 * We want to preferentially return RW over RO layouts, so ensure those
1018 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1019 const struct pnfs_layout_range
*l2
)
1023 /* high offset > low offset */
1024 d
= l1
->offset
- l2
->offset
;
1028 /* short length > long length */
1029 d
= l2
->length
- l1
->length
;
1033 /* read > read/write */
1034 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1038 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1039 struct pnfs_layout_segment
*lseg
)
1041 struct pnfs_layout_segment
*lp
;
1043 dprintk("%s:Begin\n", __func__
);
1045 list_for_each_entry(lp
, &lo
->plh_segs
, pls_list
) {
1046 if (pnfs_lseg_range_cmp(&lseg
->pls_range
, &lp
->pls_range
) > 0)
1048 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1049 dprintk("%s: inserted lseg %p "
1050 "iomode %d offset %llu length %llu before "
1051 "lp %p iomode %d offset %llu length %llu\n",
1052 __func__
, lseg
, lseg
->pls_range
.iomode
,
1053 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1054 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1055 lp
->pls_range
.length
);
1058 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1059 dprintk("%s: inserted lseg %p "
1060 "iomode %d offset %llu length %llu at tail\n",
1061 __func__
, lseg
, lseg
->pls_range
.iomode
,
1062 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1064 pnfs_get_layout_hdr(lo
);
1066 dprintk("%s:Return\n", __func__
);
1069 static struct pnfs_layout_hdr
*
1070 alloc_init_layout_hdr(struct inode
*ino
,
1071 struct nfs_open_context
*ctx
,
1074 struct pnfs_layout_hdr
*lo
;
1076 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1079 atomic_set(&lo
->plh_refcount
, 1);
1080 INIT_LIST_HEAD(&lo
->plh_layouts
);
1081 INIT_LIST_HEAD(&lo
->plh_segs
);
1082 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1083 lo
->plh_inode
= ino
;
1084 lo
->plh_lc_cred
= get_rpccred(ctx
->cred
);
1088 static struct pnfs_layout_hdr
*
1089 pnfs_find_alloc_layout(struct inode
*ino
,
1090 struct nfs_open_context
*ctx
,
1093 struct nfs_inode
*nfsi
= NFS_I(ino
);
1094 struct pnfs_layout_hdr
*new = NULL
;
1096 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1098 if (nfsi
->layout
!= NULL
)
1100 spin_unlock(&ino
->i_lock
);
1101 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1102 spin_lock(&ino
->i_lock
);
1104 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1107 } else if (new != NULL
)
1108 pnfs_free_layout_hdr(new);
1110 pnfs_get_layout_hdr(nfsi
->layout
);
1111 return nfsi
->layout
;
1115 * iomode matching rules:
1126 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1127 const struct pnfs_layout_range
*range
)
1129 struct pnfs_layout_range range1
;
1131 if ((range
->iomode
== IOMODE_RW
&&
1132 ls_range
->iomode
!= IOMODE_RW
) ||
1133 !pnfs_lseg_range_intersecting(ls_range
, range
))
1136 /* range1 covers only the first byte in the range */
1139 return pnfs_lseg_range_contained(ls_range
, &range1
);
1143 * lookup range in layout
1145 static struct pnfs_layout_segment
*
1146 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1147 struct pnfs_layout_range
*range
)
1149 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1151 dprintk("%s:Begin\n", __func__
);
1153 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1154 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1155 pnfs_lseg_range_match(&lseg
->pls_range
, range
)) {
1156 ret
= pnfs_get_lseg(lseg
);
1159 if (lseg
->pls_range
.offset
> range
->offset
)
1163 dprintk("%s:Return lseg %p ref %d\n",
1164 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
1169 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1170 * to the MDS or over pNFS
1172 * The nfs_inode read_io and write_io fields are cumulative counters reset
1173 * when there are no layout segments. Note that in pnfs_update_layout iomode
1174 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1177 * A return of true means use MDS I/O.
1180 * If a file's size is smaller than the file size threshold, data accesses
1181 * SHOULD be sent to the metadata server. If an I/O request has a length that
1182 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1183 * server. If both file size and I/O size are provided, the client SHOULD
1184 * reach or exceed both thresholds before sending its read or write
1185 * requests to the data server.
1187 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1188 struct inode
*ino
, int iomode
)
1190 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1191 struct nfs_inode
*nfsi
= NFS_I(ino
);
1192 loff_t fsize
= i_size_read(ino
);
1193 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1198 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1199 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1203 if (t
->bm
& THRESHOLD_RD
) {
1204 dprintk("%s fsize %llu\n", __func__
, fsize
);
1206 if (fsize
< t
->rd_sz
)
1209 if (t
->bm
& THRESHOLD_RD_IO
) {
1210 dprintk("%s nfsi->read_io %llu\n", __func__
,
1213 if (nfsi
->read_io
< t
->rd_io_sz
)
1218 if (t
->bm
& THRESHOLD_WR
) {
1219 dprintk("%s fsize %llu\n", __func__
, fsize
);
1221 if (fsize
< t
->wr_sz
)
1224 if (t
->bm
& THRESHOLD_WR_IO
) {
1225 dprintk("%s nfsi->write_io %llu\n", __func__
,
1228 if (nfsi
->write_io
< t
->wr_io_sz
)
1233 if (size_set
&& io_set
) {
1236 } else if (size
|| io
)
1239 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1244 * Layout segment is retreived from the server if not cached.
1245 * The appropriate layout segment is referenced and returned to the caller.
1247 struct pnfs_layout_segment
*
1248 pnfs_update_layout(struct inode
*ino
,
1249 struct nfs_open_context
*ctx
,
1252 enum pnfs_iomode iomode
,
1255 struct pnfs_layout_range arg
= {
1261 struct nfs_server
*server
= NFS_SERVER(ino
);
1262 struct nfs_client
*clp
= server
->nfs_client
;
1263 struct pnfs_layout_hdr
*lo
;
1264 struct pnfs_layout_segment
*lseg
= NULL
;
1267 if (!pnfs_enabled_sb(NFS_SERVER(ino
)))
1270 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
))
1273 spin_lock(&ino
->i_lock
);
1274 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1276 spin_unlock(&ino
->i_lock
);
1280 /* Do we even need to bother with this? */
1281 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1282 dprintk("%s matches recall, use MDS\n", __func__
);
1286 /* if LAYOUTGET already failed once we don't try again */
1287 if (pnfs_layout_io_test_failed(lo
, iomode
))
1290 /* Check to see if the layout for the given range already exists */
1291 lseg
= pnfs_find_lseg(lo
, &arg
);
1295 if (pnfs_layoutgets_blocked(lo
, 0))
1297 atomic_inc(&lo
->plh_outstanding
);
1299 first
= list_empty(&lo
->plh_layouts
) ? true : false;
1300 spin_unlock(&ino
->i_lock
);
1303 /* The lo must be on the clp list if there is any
1304 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1306 spin_lock(&clp
->cl_lock
);
1307 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1308 spin_unlock(&clp
->cl_lock
);
1311 pg_offset
= arg
.offset
& ~PAGE_CACHE_MASK
;
1313 arg
.offset
-= pg_offset
;
1314 arg
.length
+= pg_offset
;
1316 if (arg
.length
!= NFS4_MAX_UINT64
)
1317 arg
.length
= PAGE_CACHE_ALIGN(arg
.length
);
1319 lseg
= send_layoutget(lo
, ctx
, &arg
, gfp_flags
);
1320 atomic_dec(&lo
->plh_outstanding
);
1322 pnfs_put_layout_hdr(lo
);
1324 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1325 "(%s, offset: %llu, length: %llu)\n",
1326 __func__
, ino
->i_sb
->s_id
,
1327 (unsigned long long)NFS_FILEID(ino
),
1328 lseg
== NULL
? "not found" : "found",
1329 iomode
==IOMODE_RW
? "read/write" : "read-only",
1330 (unsigned long long)pos
,
1331 (unsigned long long)count
);
1334 spin_unlock(&ino
->i_lock
);
1335 goto out_put_layout_hdr
;
1337 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1339 struct pnfs_layout_segment
*
1340 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1342 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1343 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1344 struct pnfs_layout_segment
*lseg
;
1345 struct inode
*ino
= lo
->plh_inode
;
1349 /* Inject layout blob into I/O device driver */
1350 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1351 if (!lseg
|| IS_ERR(lseg
)) {
1355 status
= PTR_ERR(lseg
);
1356 dprintk("%s: Could not allocate layout: error %d\n",
1361 spin_lock(&ino
->i_lock
);
1362 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1363 dprintk("%s forget reply due to recall\n", __func__
);
1364 goto out_forget_reply
;
1367 if (pnfs_layoutgets_blocked(lo
, 1) ||
1368 pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
1369 dprintk("%s forget reply due to state\n", __func__
);
1370 goto out_forget_reply
;
1373 /* Check that the new stateid matches the old stateid */
1374 pnfs_verify_layout_stateid(lo
, &res
->stateid
, &free_me
);
1375 /* Done processing layoutget. Set the layout stateid */
1376 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1378 init_lseg(lo
, lseg
);
1379 lseg
->pls_range
= res
->range
;
1380 pnfs_get_lseg(lseg
);
1381 pnfs_layout_insert_lseg(lo
, lseg
);
1383 if (res
->return_on_close
) {
1384 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1385 set_bit(NFS_LAYOUT_ROC
, &lo
->plh_flags
);
1388 spin_unlock(&ino
->i_lock
);
1389 pnfs_free_lseg_list(&free_me
);
1392 return ERR_PTR(status
);
1395 spin_unlock(&ino
->i_lock
);
1396 lseg
->pls_layout
= lo
;
1397 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1402 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1404 u64 rd_size
= req
->wb_bytes
;
1406 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
1408 if (pgio
->pg_dreq
== NULL
)
1409 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
1411 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
1413 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1419 /* If no lseg, fall back to read through mds */
1420 if (pgio
->pg_lseg
== NULL
)
1421 nfs_pageio_reset_read_mds(pgio
);
1424 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1427 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
1428 struct nfs_page
*req
, u64 wb_size
)
1430 WARN_ON_ONCE(pgio
->pg_lseg
!= NULL
);
1432 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1438 /* If no lseg, fall back to write through mds */
1439 if (pgio
->pg_lseg
== NULL
)
1440 nfs_pageio_reset_write_mds(pgio
);
1442 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1445 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1446 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1449 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*prev
,
1450 struct nfs_page
*req
)
1453 u64 seg_end
, req_start
, seg_left
;
1455 size
= nfs_generic_pg_test(pgio
, prev
, req
);
1460 * 'size' contains the number of bytes left in the current page (up
1461 * to the original size asked for in @req->wb_bytes).
1463 * Calculate how many bytes are left in the layout segment
1464 * and if there are less bytes than 'size', return that instead.
1466 * Please also note that 'end_offset' is actually the offset of the
1467 * first byte that lies outside the pnfs_layout_range. FIXME?
1470 if (pgio
->pg_lseg
) {
1471 seg_end
= end_offset(pgio
->pg_lseg
->pls_range
.offset
,
1472 pgio
->pg_lseg
->pls_range
.length
);
1473 req_start
= req_offset(req
);
1474 WARN_ON_ONCE(req_start
> seg_end
);
1475 /* start of request is past the last byte of this segment */
1476 if (req_start
>= seg_end
)
1479 /* adjust 'size' iff there are fewer bytes left in the
1480 * segment than what nfs_generic_pg_test returned */
1481 seg_left
= seg_end
- req_start
;
1482 if (seg_left
< size
)
1483 size
= (unsigned int)seg_left
;
1488 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
1490 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
1492 struct nfs_pageio_descriptor pgio
;
1494 /* Resend all requests through the MDS */
1495 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
1496 hdr
->completion_ops
);
1497 return nfs_pageio_resend(&pgio
, hdr
);
1499 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
1501 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
1504 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
1505 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1506 PNFS_LAYOUTRET_ON_ERROR
) {
1507 pnfs_return_layout(hdr
->inode
);
1509 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1510 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
1514 * Called by non rpc-based layout drivers
1516 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
1518 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
1519 if (!hdr
->pnfs_error
) {
1520 pnfs_set_layoutcommit(hdr
);
1521 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
1523 pnfs_ld_handle_write_error(hdr
);
1524 hdr
->mds_ops
->rpc_release(hdr
);
1526 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
1529 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
1530 struct nfs_pgio_header
*hdr
)
1532 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1533 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1534 nfs_pageio_reset_write_mds(desc
);
1535 desc
->pg_recoalesce
= 1;
1537 nfs_pgio_data_destroy(hdr
);
1540 static enum pnfs_try_status
1541 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
1542 const struct rpc_call_ops
*call_ops
,
1543 struct pnfs_layout_segment
*lseg
,
1546 struct inode
*inode
= hdr
->inode
;
1547 enum pnfs_try_status trypnfs
;
1548 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1550 hdr
->mds_ops
= call_ops
;
1552 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
1553 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
1554 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
1555 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1556 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
1557 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1562 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
1563 struct nfs_pgio_header
*hdr
, int how
)
1565 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1566 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1567 enum pnfs_try_status trypnfs
;
1569 desc
->pg_lseg
= NULL
;
1570 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
1571 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1572 pnfs_write_through_mds(desc
, hdr
);
1573 pnfs_put_lseg(lseg
);
1576 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
1578 pnfs_put_lseg(hdr
->lseg
);
1579 nfs_pgio_header_free(hdr
);
1581 EXPORT_SYMBOL_GPL(pnfs_writehdr_free
);
1584 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
1586 struct nfs_pgio_header
*hdr
;
1589 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
1591 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1592 pnfs_put_lseg(desc
->pg_lseg
);
1593 desc
->pg_lseg
= NULL
;
1596 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
1597 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1598 ret
= nfs_generic_pgio(desc
, hdr
);
1600 pnfs_put_lseg(desc
->pg_lseg
);
1601 desc
->pg_lseg
= NULL
;
1603 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
1606 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
1608 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
1610 struct nfs_pageio_descriptor pgio
;
1612 /* Resend all requests through the MDS */
1613 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
1614 return nfs_pageio_resend(&pgio
, hdr
);
1616 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
1618 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
1620 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
1621 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
1622 PNFS_LAYOUTRET_ON_ERROR
) {
1623 pnfs_return_layout(hdr
->inode
);
1625 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
1626 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
1630 * Called by non rpc-based layout drivers
1632 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
1634 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
1635 if (likely(!hdr
->pnfs_error
)) {
1636 __nfs4_read_done_cb(hdr
);
1637 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
1639 pnfs_ld_handle_read_error(hdr
);
1640 hdr
->mds_ops
->rpc_release(hdr
);
1642 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
1645 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
1646 struct nfs_pgio_header
*hdr
)
1648 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
1649 list_splice_tail_init(&hdr
->pages
, &desc
->pg_list
);
1650 nfs_pageio_reset_read_mds(desc
);
1651 desc
->pg_recoalesce
= 1;
1653 nfs_pgio_data_destroy(hdr
);
1657 * Call the appropriate parallel I/O subsystem read function.
1659 static enum pnfs_try_status
1660 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
1661 const struct rpc_call_ops
*call_ops
,
1662 struct pnfs_layout_segment
*lseg
)
1664 struct inode
*inode
= hdr
->inode
;
1665 struct nfs_server
*nfss
= NFS_SERVER(inode
);
1666 enum pnfs_try_status trypnfs
;
1668 hdr
->mds_ops
= call_ops
;
1670 dprintk("%s: Reading ino:%lu %u@%llu\n",
1671 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
1673 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
1674 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
1675 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
1676 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
1681 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
1683 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
1684 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
1685 enum pnfs_try_status trypnfs
;
1687 desc
->pg_lseg
= NULL
;
1688 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
1689 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
1690 pnfs_read_through_mds(desc
, hdr
);
1691 pnfs_put_lseg(lseg
);
1694 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
1696 pnfs_put_lseg(hdr
->lseg
);
1697 nfs_pgio_header_free(hdr
);
1699 EXPORT_SYMBOL_GPL(pnfs_readhdr_free
);
1702 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
1704 struct nfs_pgio_header
*hdr
;
1707 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
1709 desc
->pg_completion_ops
->error_cleanup(&desc
->pg_list
);
1711 pnfs_put_lseg(desc
->pg_lseg
);
1712 desc
->pg_lseg
= NULL
;
1715 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
1716 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
1717 ret
= nfs_generic_pgio(desc
, hdr
);
1719 pnfs_put_lseg(desc
->pg_lseg
);
1720 desc
->pg_lseg
= NULL
;
1722 pnfs_do_read(desc
, hdr
);
1725 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
1727 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
1729 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
1731 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
1732 smp_mb__after_atomic();
1733 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
1737 * There can be multiple RW segments.
1739 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
1741 struct pnfs_layout_segment
*lseg
;
1743 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
1744 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
1745 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1746 list_add(&lseg
->pls_lc_list
, listp
);
1750 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
1752 struct pnfs_layout_segment
*lseg
, *tmp
;
1754 /* Matched by references in pnfs_set_layoutcommit */
1755 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
1756 list_del_init(&lseg
->pls_lc_list
);
1757 pnfs_put_lseg(lseg
);
1760 pnfs_clear_layoutcommitting(inode
);
1763 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
1765 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
1767 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
1770 pnfs_set_layoutcommit(struct nfs_pgio_header
*hdr
)
1772 struct inode
*inode
= hdr
->inode
;
1773 struct nfs_inode
*nfsi
= NFS_I(inode
);
1774 loff_t end_pos
= hdr
->mds_offset
+ hdr
->res
.count
;
1775 bool mark_as_dirty
= false;
1777 spin_lock(&inode
->i_lock
);
1778 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1779 mark_as_dirty
= true;
1780 dprintk("%s: Set layoutcommit for inode %lu ",
1781 __func__
, inode
->i_ino
);
1783 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &hdr
->lseg
->pls_flags
)) {
1784 /* references matched in nfs4_layoutcommit_release */
1785 pnfs_get_lseg(hdr
->lseg
);
1787 if (end_pos
> nfsi
->layout
->plh_lwb
)
1788 nfsi
->layout
->plh_lwb
= end_pos
;
1789 spin_unlock(&inode
->i_lock
);
1790 dprintk("%s: lseg %p end_pos %llu\n",
1791 __func__
, hdr
->lseg
, nfsi
->layout
->plh_lwb
);
1793 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1794 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1796 mark_inode_dirty_sync(inode
);
1798 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
1800 void pnfs_commit_set_layoutcommit(struct nfs_commit_data
*data
)
1802 struct inode
*inode
= data
->inode
;
1803 struct nfs_inode
*nfsi
= NFS_I(inode
);
1804 bool mark_as_dirty
= false;
1806 spin_lock(&inode
->i_lock
);
1807 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
1808 mark_as_dirty
= true;
1809 dprintk("%s: Set layoutcommit for inode %lu ",
1810 __func__
, inode
->i_ino
);
1812 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &data
->lseg
->pls_flags
)) {
1813 /* references matched in nfs4_layoutcommit_release */
1814 pnfs_get_lseg(data
->lseg
);
1816 if (data
->lwb
> nfsi
->layout
->plh_lwb
)
1817 nfsi
->layout
->plh_lwb
= data
->lwb
;
1818 spin_unlock(&inode
->i_lock
);
1819 dprintk("%s: lseg %p end_pos %llu\n",
1820 __func__
, data
->lseg
, nfsi
->layout
->plh_lwb
);
1822 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
1823 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
1825 mark_inode_dirty_sync(inode
);
1827 EXPORT_SYMBOL_GPL(pnfs_commit_set_layoutcommit
);
1829 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
1831 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
1833 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
1834 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
1835 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
1839 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
1840 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
1841 * data to disk to allow the server to recover the data if it crashes.
1842 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
1843 * is off, and a COMMIT is sent to a data server, or
1844 * if WRITEs to a data server return NFS_DATA_SYNC.
1847 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
1849 struct nfs4_layoutcommit_data
*data
;
1850 struct nfs_inode
*nfsi
= NFS_I(inode
);
1854 if (!pnfs_layoutcommit_outstanding(inode
))
1857 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
1860 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
1863 status
= wait_on_bit_lock_action(&nfsi
->flags
,
1864 NFS_INO_LAYOUTCOMMITTING
,
1865 nfs_wait_bit_killable
,
1872 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1873 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
1875 goto clear_layoutcommitting
;
1878 spin_lock(&inode
->i_lock
);
1879 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1882 INIT_LIST_HEAD(&data
->lseg_list
);
1883 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
1885 end_pos
= nfsi
->layout
->plh_lwb
;
1886 nfsi
->layout
->plh_lwb
= 0;
1888 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
1889 spin_unlock(&inode
->i_lock
);
1891 data
->args
.inode
= inode
;
1892 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
1893 nfs_fattr_init(&data
->fattr
);
1894 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
1895 data
->res
.fattr
= &data
->fattr
;
1896 data
->args
.lastbytewritten
= end_pos
- 1;
1897 data
->res
.server
= NFS_SERVER(inode
);
1899 status
= nfs4_proc_layoutcommit(data
, sync
);
1902 mark_inode_dirty_sync(inode
);
1903 dprintk("<-- %s status %d\n", __func__
, status
);
1906 spin_unlock(&inode
->i_lock
);
1908 clear_layoutcommitting
:
1909 pnfs_clear_layoutcommitting(inode
);
1913 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
1915 struct nfs4_threshold
*thp
;
1917 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
1919 dprintk("%s mdsthreshold allocation failed\n", __func__
);