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"
37 #include "delegation.h"
40 #define NFSDBG_FACILITY NFSDBG_PNFS
41 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
46 * protects pnfs_modules_tbl.
48 static DEFINE_SPINLOCK(pnfs_spinlock
);
51 * pnfs_modules_tbl holds all pnfs modules
53 static LIST_HEAD(pnfs_modules_tbl
);
55 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
);
57 /* Return the registered pnfs layout driver module matching given id */
58 static struct pnfs_layoutdriver_type
*
59 find_pnfs_driver_locked(u32 id
)
61 struct pnfs_layoutdriver_type
*local
;
63 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
68 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
72 static struct pnfs_layoutdriver_type
*
73 find_pnfs_driver(u32 id
)
75 struct pnfs_layoutdriver_type
*local
;
77 spin_lock(&pnfs_spinlock
);
78 local
= find_pnfs_driver_locked(id
);
79 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
80 dprintk("%s: Could not grab reference on module\n", __func__
);
83 spin_unlock(&pnfs_spinlock
);
88 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
90 if (nfss
->pnfs_curr_ld
) {
91 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
92 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
93 /* Decrement the MDS count. Purge the deviceid cache if zero */
94 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
95 nfs4_deviceid_purge_client(nfss
->nfs_client
);
96 module_put(nfss
->pnfs_curr_ld
->owner
);
98 nfss
->pnfs_curr_ld
= NULL
;
102 * Try to set the server's pnfs module to the pnfs layout type specified by id.
103 * Currently only one pNFS layout driver per filesystem is supported.
105 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
108 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
111 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
115 if (!(server
->nfs_client
->cl_exchange_flags
&
116 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
117 printk(KERN_ERR
"NFS: %s: id %u cl_exchange_flags 0x%x\n",
118 __func__
, id
, server
->nfs_client
->cl_exchange_flags
);
121 ld_type
= find_pnfs_driver(id
);
123 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
, id
);
124 ld_type
= find_pnfs_driver(id
);
126 dprintk("%s: No pNFS module found for %u.\n",
131 server
->pnfs_curr_ld
= ld_type
;
132 if (ld_type
->set_layoutdriver
133 && ld_type
->set_layoutdriver(server
, mntfh
)) {
134 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
135 "driver %u.\n", __func__
, id
);
136 module_put(ld_type
->owner
);
139 /* Bump the MDS count */
140 atomic_inc(&server
->nfs_client
->cl_mds_count
);
142 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
146 dprintk("%s: Using NFSv4 I/O\n", __func__
);
147 server
->pnfs_curr_ld
= NULL
;
151 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
153 int status
= -EINVAL
;
154 struct pnfs_layoutdriver_type
*tmp
;
156 if (ld_type
->id
== 0) {
157 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
160 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
161 printk(KERN_ERR
"NFS: %s Layout driver must provide "
162 "alloc_lseg and free_lseg.\n", __func__
);
166 spin_lock(&pnfs_spinlock
);
167 tmp
= find_pnfs_driver_locked(ld_type
->id
);
169 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
171 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
174 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
175 __func__
, ld_type
->id
);
177 spin_unlock(&pnfs_spinlock
);
181 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
184 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
186 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
187 spin_lock(&pnfs_spinlock
);
188 list_del(&ld_type
->pnfs_tblid
);
189 spin_unlock(&pnfs_spinlock
);
191 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
194 * pNFS client layout cache
197 /* Need to hold i_lock if caller does not already hold reference */
199 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
201 atomic_inc(&lo
->plh_refcount
);
204 static struct pnfs_layout_hdr
*
205 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
207 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
208 return ld
->alloc_layout_hdr(ino
, gfp_flags
);
212 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
214 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
215 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
217 if (!list_empty(&lo
->plh_layouts
)) {
218 struct nfs_client
*clp
= server
->nfs_client
;
220 spin_lock(&clp
->cl_lock
);
221 list_del_init(&lo
->plh_layouts
);
222 spin_unlock(&clp
->cl_lock
);
224 put_rpccred(lo
->plh_lc_cred
);
225 return ld
->free_layout_hdr(lo
);
229 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
231 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
232 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
234 /* Reset MDS Threshold I/O counters */
240 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
242 struct inode
*inode
= lo
->plh_inode
;
244 pnfs_layoutreturn_before_put_layout_hdr(lo
);
246 if (atomic_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
247 if (!list_empty(&lo
->plh_segs
))
248 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
249 pnfs_detach_layout_hdr(lo
);
250 spin_unlock(&inode
->i_lock
);
251 pnfs_free_layout_hdr(lo
);
256 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
258 * In order to continue using the pnfs_layout_hdr, a full recovery
260 * Note that caller must hold inode->i_lock.
263 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr
*lo
,
264 struct list_head
*lseg_list
)
266 struct pnfs_layout_range range
= {
267 .iomode
= IOMODE_ANY
,
269 .length
= NFS4_MAX_UINT64
,
272 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
273 return pnfs_mark_matching_lsegs_invalid(lo
, lseg_list
, &range
, 0);
277 pnfs_iomode_to_fail_bit(u32 iomode
)
279 return iomode
== IOMODE_RW
?
280 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
284 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
286 lo
->plh_retry_timestamp
= jiffies
;
287 if (!test_and_set_bit(fail_bit
, &lo
->plh_flags
))
288 atomic_inc(&lo
->plh_refcount
);
292 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
294 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
295 atomic_dec(&lo
->plh_refcount
);
299 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
301 struct inode
*inode
= lo
->plh_inode
;
302 struct pnfs_layout_range range
= {
305 .length
= NFS4_MAX_UINT64
,
309 spin_lock(&inode
->i_lock
);
310 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
311 pnfs_mark_matching_lsegs_invalid(lo
, &head
, &range
, 0);
312 spin_unlock(&inode
->i_lock
);
313 pnfs_free_lseg_list(&head
);
314 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
315 iomode
== IOMODE_RW
? "RW" : "READ");
319 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
321 unsigned long start
, end
;
322 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
324 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
327 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
328 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
329 /* It is time to retry the failed layoutgets */
330 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
337 pnfs_init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
,
338 const struct pnfs_layout_range
*range
,
339 const nfs4_stateid
*stateid
)
341 INIT_LIST_HEAD(&lseg
->pls_list
);
342 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
343 atomic_set(&lseg
->pls_refcount
, 1);
344 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
345 lseg
->pls_layout
= lo
;
346 lseg
->pls_range
= *range
;
347 lseg
->pls_seq
= be32_to_cpu(stateid
->seqid
);
350 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
352 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
354 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
358 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
359 struct pnfs_layout_segment
*lseg
)
361 struct inode
*inode
= lo
->plh_inode
;
363 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
364 list_del_init(&lseg
->pls_list
);
365 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
366 atomic_dec(&lo
->plh_refcount
);
367 if (list_empty(&lo
->plh_segs
)) {
368 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
369 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
371 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
375 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
377 struct pnfs_layout_hdr
*lo
;
383 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
384 atomic_read(&lseg
->pls_refcount
),
385 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
387 lo
= lseg
->pls_layout
;
388 inode
= lo
->plh_inode
;
390 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
391 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
392 spin_unlock(&inode
->i_lock
);
395 pnfs_get_layout_hdr(lo
);
396 pnfs_layout_remove_lseg(lo
, lseg
);
397 spin_unlock(&inode
->i_lock
);
398 pnfs_free_lseg(lseg
);
399 pnfs_put_layout_hdr(lo
);
402 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
404 static void pnfs_free_lseg_async_work(struct work_struct
*work
)
406 struct pnfs_layout_segment
*lseg
;
407 struct pnfs_layout_hdr
*lo
;
409 lseg
= container_of(work
, struct pnfs_layout_segment
, pls_work
);
410 lo
= lseg
->pls_layout
;
412 pnfs_free_lseg(lseg
);
413 pnfs_put_layout_hdr(lo
);
416 static void pnfs_free_lseg_async(struct pnfs_layout_segment
*lseg
)
418 INIT_WORK(&lseg
->pls_work
, pnfs_free_lseg_async_work
);
419 schedule_work(&lseg
->pls_work
);
423 pnfs_put_lseg_locked(struct pnfs_layout_segment
*lseg
)
428 assert_spin_locked(&lseg
->pls_layout
->plh_inode
->i_lock
);
430 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
431 atomic_read(&lseg
->pls_refcount
),
432 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
433 if (atomic_dec_and_test(&lseg
->pls_refcount
)) {
434 struct pnfs_layout_hdr
*lo
= lseg
->pls_layout
;
435 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
437 pnfs_get_layout_hdr(lo
);
438 pnfs_layout_remove_lseg(lo
, lseg
);
439 pnfs_free_lseg_async(lseg
);
442 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked
);
445 end_offset(u64 start
, u64 len
)
450 return end
>= start
? end
: NFS4_MAX_UINT64
;
454 * is l2 fully contained in l1?
456 * [----------------------------------)
461 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
462 const struct pnfs_layout_range
*l2
)
464 u64 start1
= l1
->offset
;
465 u64 end1
= end_offset(start1
, l1
->length
);
466 u64 start2
= l2
->offset
;
467 u64 end2
= end_offset(start2
, l2
->length
);
469 return (start1
<= start2
) && (end1
>= end2
);
473 * is l1 and l2 intersecting?
475 * [----------------------------------)
480 pnfs_lseg_range_intersecting(const struct pnfs_layout_range
*l1
,
481 const struct pnfs_layout_range
*l2
)
483 u64 start1
= l1
->offset
;
484 u64 end1
= end_offset(start1
, l1
->length
);
485 u64 start2
= l2
->offset
;
486 u64 end2
= end_offset(start2
, l2
->length
);
488 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
489 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
492 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
493 struct list_head
*tmp_list
)
495 if (!atomic_dec_and_test(&lseg
->pls_refcount
))
497 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
498 list_add(&lseg
->pls_list
, tmp_list
);
502 /* Returns 1 if lseg is removed from list, 0 otherwise */
503 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
504 struct list_head
*tmp_list
)
508 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
509 /* Remove the reference keeping the lseg in the
510 * list. It will now be removed when all
511 * outstanding io is finished.
513 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
514 atomic_read(&lseg
->pls_refcount
));
515 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
522 * Compare 2 layout stateid sequence ids, to see which is newer,
523 * taking into account wraparound issues.
525 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
527 return (s32
)(s1
- s2
) > 0;
531 pnfs_should_free_range(const struct pnfs_layout_range
*lseg_range
,
532 const struct pnfs_layout_range
*recall_range
)
534 return (recall_range
->iomode
== IOMODE_ANY
||
535 lseg_range
->iomode
== recall_range
->iomode
) &&
536 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
540 pnfs_match_lseg_recall(const struct pnfs_layout_segment
*lseg
,
541 const struct pnfs_layout_range
*recall_range
,
544 if (seq
!= 0 && pnfs_seqid_is_newer(lseg
->pls_seq
, seq
))
546 if (recall_range
== NULL
)
548 return pnfs_should_free_range(&lseg
->pls_range
, recall_range
);
552 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
553 * @lo: layout header containing the lsegs
554 * @tmp_list: list head where doomed lsegs should go
555 * @recall_range: optional recall range argument to match (may be NULL)
556 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
558 * Walk the list of lsegs in the layout header, and tear down any that should
559 * be destroyed. If "recall_range" is specified then the segment must match
560 * that range. If "seq" is non-zero, then only match segments that were handed
561 * out at or before that sequence.
563 * Returns number of matching invalid lsegs remaining in list after scanning
564 * it and purging them.
567 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
568 struct list_head
*tmp_list
,
569 const struct pnfs_layout_range
*recall_range
,
572 struct pnfs_layout_segment
*lseg
, *next
;
575 dprintk("%s:Begin lo %p\n", __func__
, lo
);
577 if (list_empty(&lo
->plh_segs
))
579 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
580 if (pnfs_match_lseg_recall(lseg
, recall_range
, seq
)) {
581 dprintk("%s: freeing lseg %p iomode %d seq %u"
582 "offset %llu length %llu\n", __func__
,
583 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_seq
,
584 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
585 if (!mark_lseg_invalid(lseg
, tmp_list
))
588 dprintk("%s:Return %i\n", __func__
, remaining
);
592 /* note free_me must contain lsegs from a single layout_hdr */
594 pnfs_free_lseg_list(struct list_head
*free_me
)
596 struct pnfs_layout_segment
*lseg
, *tmp
;
598 if (list_empty(free_me
))
601 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
602 list_del(&lseg
->pls_list
);
603 pnfs_free_lseg(lseg
);
608 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
610 struct pnfs_layout_hdr
*lo
;
613 spin_lock(&nfsi
->vfs_inode
.i_lock
);
616 pnfs_get_layout_hdr(lo
);
617 pnfs_mark_layout_stateid_invalid(lo
, &tmp_list
);
618 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
619 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
620 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
621 pnfs_free_lseg_list(&tmp_list
);
622 pnfs_put_layout_hdr(lo
);
624 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
626 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
629 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
630 struct list_head
*layout_list
)
632 struct pnfs_layout_hdr
*lo
;
635 spin_lock(&inode
->i_lock
);
636 lo
= NFS_I(inode
)->layout
;
637 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
638 pnfs_get_layout_hdr(lo
);
639 list_add(&lo
->plh_bulk_destroy
, layout_list
);
642 spin_unlock(&inode
->i_lock
);
646 /* Caller must hold rcu_read_lock and clp->cl_lock */
648 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
649 struct nfs_server
*server
,
650 struct list_head
*layout_list
)
652 struct pnfs_layout_hdr
*lo
, *next
;
655 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
656 inode
= igrab(lo
->plh_inode
);
659 list_del_init(&lo
->plh_layouts
);
660 if (pnfs_layout_add_bulk_destroy_list(inode
, layout_list
))
663 spin_unlock(&clp
->cl_lock
);
665 spin_lock(&clp
->cl_lock
);
673 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
676 struct pnfs_layout_hdr
*lo
;
678 LIST_HEAD(lseg_list
);
681 while (!list_empty(layout_list
)) {
682 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
684 dprintk("%s freeing layout for inode %lu\n", __func__
,
685 lo
->plh_inode
->i_ino
);
686 inode
= lo
->plh_inode
;
688 pnfs_layoutcommit_inode(inode
, false);
690 spin_lock(&inode
->i_lock
);
691 list_del_init(&lo
->plh_bulk_destroy
);
692 if (pnfs_mark_layout_stateid_invalid(lo
, &lseg_list
)) {
694 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
697 spin_unlock(&inode
->i_lock
);
698 pnfs_free_lseg_list(&lseg_list
);
699 /* Free all lsegs that are attached to commit buckets */
700 nfs_commit_inode(inode
, 0);
701 pnfs_put_layout_hdr(lo
);
708 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
709 struct nfs_fsid
*fsid
,
712 struct nfs_server
*server
;
713 LIST_HEAD(layout_list
);
715 spin_lock(&clp
->cl_lock
);
718 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
719 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
721 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
727 spin_unlock(&clp
->cl_lock
);
729 if (list_empty(&layout_list
))
731 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
735 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
738 struct nfs_server
*server
;
739 LIST_HEAD(layout_list
);
741 spin_lock(&clp
->cl_lock
);
744 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
745 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
751 spin_unlock(&clp
->cl_lock
);
753 if (list_empty(&layout_list
))
755 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
759 * Called by the state manger to remove all layouts established under an
763 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
765 nfs4_deviceid_mark_client_invalid(clp
);
766 nfs4_deviceid_purge_client(clp
);
768 pnfs_destroy_layouts_byclid(clp
, false);
771 /* update lo->plh_stateid with new if is more recent */
773 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
776 u32 oldseq
, newseq
, new_barrier
= 0;
777 bool invalid
= !pnfs_layout_is_valid(lo
);
779 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
780 newseq
= be32_to_cpu(new->seqid
);
781 if (invalid
|| pnfs_seqid_is_newer(newseq
, oldseq
)) {
782 nfs4_stateid_copy(&lo
->plh_stateid
, new);
784 * Because of wraparound, we want to keep the barrier
785 * "close" to the current seqids.
787 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
790 new_barrier
= be32_to_cpu(new->seqid
);
791 else if (new_barrier
== 0)
793 if (invalid
|| pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
794 lo
->plh_barrier
= new_barrier
;
798 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
799 const nfs4_stateid
*stateid
)
801 u32 seqid
= be32_to_cpu(stateid
->seqid
);
803 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
806 /* lget is set to 1 if called from inside send_layoutget call chain */
808 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
)
810 return lo
->plh_block_lgets
||
811 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
815 * Get layout from server.
816 * for now, assume that whole file layouts are requested.
818 * arg->length: all ones
820 static struct pnfs_layout_segment
*
821 send_layoutget(struct pnfs_layout_hdr
*lo
,
822 struct nfs_open_context
*ctx
,
823 nfs4_stateid
*stateid
,
824 const struct pnfs_layout_range
*range
,
825 long *timeout
, gfp_t gfp_flags
)
827 struct inode
*ino
= lo
->plh_inode
;
828 struct nfs_server
*server
= NFS_SERVER(ino
);
829 struct nfs4_layoutget
*lgp
;
832 dprintk("--> %s\n", __func__
);
835 * Synchronously retrieve layout information from server and
836 * store in lseg. If we race with a concurrent seqid morphing
837 * op, then re-send the LAYOUTGET.
839 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
841 return ERR_PTR(-ENOMEM
);
843 i_size
= i_size_read(ino
);
845 lgp
->args
.minlength
= PAGE_SIZE
;
846 if (lgp
->args
.minlength
> range
->length
)
847 lgp
->args
.minlength
= range
->length
;
848 if (range
->iomode
== IOMODE_READ
) {
849 if (range
->offset
>= i_size
)
850 lgp
->args
.minlength
= 0;
851 else if (i_size
- range
->offset
< lgp
->args
.minlength
)
852 lgp
->args
.minlength
= i_size
- range
->offset
;
854 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
855 pnfs_copy_range(&lgp
->args
.range
, range
);
856 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
857 lgp
->args
.inode
= ino
;
858 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
859 nfs4_stateid_copy(&lgp
->args
.stateid
, stateid
);
860 lgp
->gfp_flags
= gfp_flags
;
861 lgp
->cred
= lo
->plh_lc_cred
;
863 return nfs4_proc_layoutget(lgp
, timeout
, gfp_flags
);
866 static void pnfs_clear_layoutcommit(struct inode
*inode
,
867 struct list_head
*head
)
869 struct nfs_inode
*nfsi
= NFS_I(inode
);
870 struct pnfs_layout_segment
*lseg
, *tmp
;
872 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
874 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
875 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
877 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
881 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr
*lo
)
883 clear_bit_unlock(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
884 smp_mb__after_atomic();
885 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
);
886 rpc_wake_up(&NFS_SERVER(lo
->plh_inode
)->roc_rpcwaitq
);
890 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr
*lo
)
892 lo
->plh_return_iomode
= 0;
893 lo
->plh_return_seq
= 0;
894 clear_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
898 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr
*lo
,
899 nfs4_stateid
*stateid
,
900 enum pnfs_iomode
*iomode
)
902 if (test_and_set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
904 pnfs_get_layout_hdr(lo
);
905 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
)) {
906 if (stateid
!= NULL
) {
907 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
908 if (lo
->plh_return_seq
!= 0)
909 stateid
->seqid
= cpu_to_be32(lo
->plh_return_seq
);
912 *iomode
= lo
->plh_return_iomode
;
913 pnfs_clear_layoutreturn_info(lo
);
917 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
919 *iomode
= IOMODE_ANY
;
924 pnfs_send_layoutreturn(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*stateid
,
925 enum pnfs_iomode iomode
, bool sync
)
927 struct inode
*ino
= lo
->plh_inode
;
928 struct nfs4_layoutreturn
*lrp
;
931 lrp
= kzalloc(sizeof(*lrp
), GFP_NOFS
);
932 if (unlikely(lrp
== NULL
)) {
934 spin_lock(&ino
->i_lock
);
935 pnfs_clear_layoutreturn_waitbit(lo
);
936 spin_unlock(&ino
->i_lock
);
937 pnfs_put_layout_hdr(lo
);
941 nfs4_stateid_copy(&lrp
->args
.stateid
, stateid
);
942 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
943 lrp
->args
.inode
= ino
;
944 lrp
->args
.range
.iomode
= iomode
;
945 lrp
->args
.range
.offset
= 0;
946 lrp
->args
.range
.length
= NFS4_MAX_UINT64
;
947 lrp
->args
.layout
= lo
;
948 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
949 lrp
->cred
= lo
->plh_lc_cred
;
951 status
= nfs4_proc_layoutreturn(lrp
, sync
);
953 dprintk("<-- %s status: %d\n", __func__
, status
);
957 /* Return true if layoutreturn is needed */
959 pnfs_layout_need_return(struct pnfs_layout_hdr
*lo
)
961 struct pnfs_layout_segment
*s
;
963 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
966 /* Defer layoutreturn until all lsegs are done */
967 list_for_each_entry(s
, &lo
->plh_segs
, pls_list
) {
968 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &s
->pls_flags
))
975 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
977 struct inode
*inode
= lo
->plh_inode
;
979 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
981 spin_lock(&inode
->i_lock
);
982 if (pnfs_layout_need_return(lo
)) {
983 nfs4_stateid stateid
;
984 enum pnfs_iomode iomode
;
987 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
988 spin_unlock(&inode
->i_lock
);
990 /* Send an async layoutreturn so we dont deadlock */
991 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
994 spin_unlock(&inode
->i_lock
);
998 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
999 * when the layout segment list is empty.
1001 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1002 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1003 * deviceid is marked invalid.
1006 _pnfs_return_layout(struct inode
*ino
)
1008 struct pnfs_layout_hdr
*lo
= NULL
;
1009 struct nfs_inode
*nfsi
= NFS_I(ino
);
1010 LIST_HEAD(tmp_list
);
1011 nfs4_stateid stateid
;
1012 int status
= 0, empty
;
1015 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
1017 spin_lock(&ino
->i_lock
);
1020 spin_unlock(&ino
->i_lock
);
1021 dprintk("NFS: %s no layout to return\n", __func__
);
1024 /* Reference matched in nfs4_layoutreturn_release */
1025 pnfs_get_layout_hdr(lo
);
1026 empty
= list_empty(&lo
->plh_segs
);
1027 pnfs_clear_layoutcommit(ino
, &tmp_list
);
1028 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
, 0);
1030 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
1031 struct pnfs_layout_range range
= {
1032 .iomode
= IOMODE_ANY
,
1034 .length
= NFS4_MAX_UINT64
,
1036 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
, &range
);
1039 /* Don't send a LAYOUTRETURN if list was initially empty */
1041 spin_unlock(&ino
->i_lock
);
1042 dprintk("NFS: %s no layout segments to return\n", __func__
);
1043 goto out_put_layout_hdr
;
1046 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, NULL
);
1047 spin_unlock(&ino
->i_lock
);
1048 pnfs_free_lseg_list(&tmp_list
);
1050 status
= pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1052 pnfs_put_layout_hdr(lo
);
1054 dprintk("<-- %s status: %d\n", __func__
, status
);
1057 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
1060 pnfs_commit_and_return_layout(struct inode
*inode
)
1062 struct pnfs_layout_hdr
*lo
;
1065 spin_lock(&inode
->i_lock
);
1066 lo
= NFS_I(inode
)->layout
;
1068 spin_unlock(&inode
->i_lock
);
1071 pnfs_get_layout_hdr(lo
);
1072 /* Block new layoutgets and read/write to ds */
1073 lo
->plh_block_lgets
++;
1074 spin_unlock(&inode
->i_lock
);
1075 filemap_fdatawait(inode
->i_mapping
);
1076 ret
= pnfs_layoutcommit_inode(inode
, true);
1078 ret
= _pnfs_return_layout(inode
);
1079 spin_lock(&inode
->i_lock
);
1080 lo
->plh_block_lgets
--;
1081 spin_unlock(&inode
->i_lock
);
1082 pnfs_put_layout_hdr(lo
);
1086 bool pnfs_roc(struct inode
*ino
)
1088 struct nfs_inode
*nfsi
= NFS_I(ino
);
1089 struct nfs_open_context
*ctx
;
1090 struct nfs4_state
*state
;
1091 struct pnfs_layout_hdr
*lo
;
1092 struct pnfs_layout_segment
*lseg
, *tmp
;
1093 nfs4_stateid stateid
;
1094 LIST_HEAD(tmp_list
);
1095 bool found
= false, layoutreturn
= false, roc
= false;
1097 spin_lock(&ino
->i_lock
);
1099 if (!lo
|| test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
1102 /* no roc if we hold a delegation */
1103 if (nfs4_check_delegation(ino
, FMODE_READ
))
1106 list_for_each_entry(ctx
, &nfsi
->open_files
, list
) {
1108 /* Don't return layout if there is open file state */
1109 if (state
!= NULL
&& state
->state
!= 0)
1113 /* always send layoutreturn if being marked so */
1114 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1115 layoutreturn
= pnfs_prepare_layoutreturn(lo
,
1118 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
1119 /* If we are sending layoutreturn, invalidate all valid lsegs */
1120 if (layoutreturn
|| test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
1121 mark_lseg_invalid(lseg
, &tmp_list
);
1124 /* ROC in two conditions:
1125 * 1. there are ROC lsegs
1126 * 2. we don't send layoutreturn
1128 if (found
&& !layoutreturn
) {
1129 /* lo ref dropped in pnfs_roc_release() */
1130 pnfs_get_layout_hdr(lo
);
1135 spin_unlock(&ino
->i_lock
);
1136 pnfs_free_lseg_list(&tmp_list
);
1137 pnfs_layoutcommit_inode(ino
, true);
1139 pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1143 void pnfs_roc_release(struct inode
*ino
)
1145 struct pnfs_layout_hdr
*lo
;
1147 spin_lock(&ino
->i_lock
);
1148 lo
= NFS_I(ino
)->layout
;
1149 pnfs_clear_layoutreturn_waitbit(lo
);
1150 if (atomic_dec_and_test(&lo
->plh_refcount
)) {
1151 pnfs_detach_layout_hdr(lo
);
1152 spin_unlock(&ino
->i_lock
);
1153 pnfs_free_layout_hdr(lo
);
1155 spin_unlock(&ino
->i_lock
);
1158 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
1160 struct pnfs_layout_hdr
*lo
;
1162 spin_lock(&ino
->i_lock
);
1163 lo
= NFS_I(ino
)->layout
;
1164 if (pnfs_seqid_is_newer(barrier
, lo
->plh_barrier
))
1165 lo
->plh_barrier
= barrier
;
1166 spin_unlock(&ino
->i_lock
);
1167 trace_nfs4_layoutreturn_on_close(ino
, 0);
1170 void pnfs_roc_get_barrier(struct inode
*ino
, u32
*barrier
)
1172 struct nfs_inode
*nfsi
= NFS_I(ino
);
1173 struct pnfs_layout_hdr
*lo
;
1176 spin_lock(&ino
->i_lock
);
1178 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
1180 /* Since close does not return a layout stateid for use as
1181 * a barrier, we choose the worst-case barrier.
1183 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
1184 spin_unlock(&ino
->i_lock
);
1187 bool pnfs_wait_on_layoutreturn(struct inode
*ino
, struct rpc_task
*task
)
1189 struct nfs_inode
*nfsi
= NFS_I(ino
);
1190 struct pnfs_layout_hdr
*lo
;
1193 /* we might not have grabbed lo reference. so need to check under
1195 spin_lock(&ino
->i_lock
);
1197 if (lo
&& test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
1199 spin_unlock(&ino
->i_lock
);
1202 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1208 * Compare two layout segments for sorting into layout cache.
1209 * We want to preferentially return RW over RO layouts, so ensure those
1213 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1214 const struct pnfs_layout_range
*l2
)
1218 /* high offset > low offset */
1219 d
= l1
->offset
- l2
->offset
;
1223 /* short length > long length */
1224 d
= l2
->length
- l1
->length
;
1228 /* read > read/write */
1229 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1233 pnfs_lseg_range_is_after(const struct pnfs_layout_range
*l1
,
1234 const struct pnfs_layout_range
*l2
)
1236 return pnfs_lseg_range_cmp(l1
, l2
) > 0;
1240 pnfs_lseg_no_merge(struct pnfs_layout_segment
*lseg
,
1241 struct pnfs_layout_segment
*old
)
1247 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1248 struct pnfs_layout_segment
*lseg
,
1249 bool (*is_after
)(const struct pnfs_layout_range
*,
1250 const struct pnfs_layout_range
*),
1251 bool (*do_merge
)(struct pnfs_layout_segment
*,
1252 struct pnfs_layout_segment
*),
1253 struct list_head
*free_me
)
1255 struct pnfs_layout_segment
*lp
, *tmp
;
1257 dprintk("%s:Begin\n", __func__
);
1259 list_for_each_entry_safe(lp
, tmp
, &lo
->plh_segs
, pls_list
) {
1260 if (test_bit(NFS_LSEG_VALID
, &lp
->pls_flags
) == 0)
1262 if (do_merge(lseg
, lp
)) {
1263 mark_lseg_invalid(lp
, free_me
);
1266 if (is_after(&lseg
->pls_range
, &lp
->pls_range
))
1268 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1269 dprintk("%s: inserted lseg %p "
1270 "iomode %d offset %llu length %llu before "
1271 "lp %p iomode %d offset %llu length %llu\n",
1272 __func__
, lseg
, lseg
->pls_range
.iomode
,
1273 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1274 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1275 lp
->pls_range
.length
);
1278 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1279 dprintk("%s: inserted lseg %p "
1280 "iomode %d offset %llu length %llu at tail\n",
1281 __func__
, lseg
, lseg
->pls_range
.iomode
,
1282 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1284 pnfs_get_layout_hdr(lo
);
1286 dprintk("%s:Return\n", __func__
);
1288 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg
);
1291 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1292 struct pnfs_layout_segment
*lseg
,
1293 struct list_head
*free_me
)
1295 struct inode
*inode
= lo
->plh_inode
;
1296 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
1298 if (ld
->add_lseg
!= NULL
)
1299 ld
->add_lseg(lo
, lseg
, free_me
);
1301 pnfs_generic_layout_insert_lseg(lo
, lseg
,
1302 pnfs_lseg_range_is_after
,
1307 static struct pnfs_layout_hdr
*
1308 alloc_init_layout_hdr(struct inode
*ino
,
1309 struct nfs_open_context
*ctx
,
1312 struct pnfs_layout_hdr
*lo
;
1314 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1317 atomic_set(&lo
->plh_refcount
, 1);
1318 INIT_LIST_HEAD(&lo
->plh_layouts
);
1319 INIT_LIST_HEAD(&lo
->plh_segs
);
1320 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1321 lo
->plh_inode
= ino
;
1322 lo
->plh_lc_cred
= get_rpccred(ctx
->cred
);
1323 lo
->plh_flags
|= 1 << NFS_LAYOUT_INVALID_STID
;
1327 static struct pnfs_layout_hdr
*
1328 pnfs_find_alloc_layout(struct inode
*ino
,
1329 struct nfs_open_context
*ctx
,
1331 __releases(&ino
->i_lock
)
1332 __acquires(&ino
->i_lock
)
1334 struct nfs_inode
*nfsi
= NFS_I(ino
);
1335 struct pnfs_layout_hdr
*new = NULL
;
1337 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1339 if (nfsi
->layout
!= NULL
)
1341 spin_unlock(&ino
->i_lock
);
1342 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1343 spin_lock(&ino
->i_lock
);
1345 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1348 } else if (new != NULL
)
1349 pnfs_free_layout_hdr(new);
1351 pnfs_get_layout_hdr(nfsi
->layout
);
1352 return nfsi
->layout
;
1356 * iomode matching rules:
1357 * iomode lseg strict match
1359 * ----- ----- ------ -----
1364 * READ READ N/A true
1365 * READ RW true false
1366 * READ RW false true
1369 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1370 const struct pnfs_layout_range
*range
,
1373 struct pnfs_layout_range range1
;
1375 if ((range
->iomode
== IOMODE_RW
&&
1376 ls_range
->iomode
!= IOMODE_RW
) ||
1377 (range
->iomode
!= ls_range
->iomode
&&
1378 strict_iomode
== true) ||
1379 !pnfs_lseg_range_intersecting(ls_range
, range
))
1382 /* range1 covers only the first byte in the range */
1385 return pnfs_lseg_range_contained(ls_range
, &range1
);
1389 * lookup range in layout
1391 static struct pnfs_layout_segment
*
1392 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1393 struct pnfs_layout_range
*range
,
1396 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1398 dprintk("%s:Begin\n", __func__
);
1400 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1401 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1402 !test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
1403 pnfs_lseg_range_match(&lseg
->pls_range
, range
,
1405 ret
= pnfs_get_lseg(lseg
);
1410 dprintk("%s:Return lseg %p ref %d\n",
1411 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
1416 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1417 * to the MDS or over pNFS
1419 * The nfs_inode read_io and write_io fields are cumulative counters reset
1420 * when there are no layout segments. Note that in pnfs_update_layout iomode
1421 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1424 * A return of true means use MDS I/O.
1427 * If a file's size is smaller than the file size threshold, data accesses
1428 * SHOULD be sent to the metadata server. If an I/O request has a length that
1429 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1430 * server. If both file size and I/O size are provided, the client SHOULD
1431 * reach or exceed both thresholds before sending its read or write
1432 * requests to the data server.
1434 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1435 struct inode
*ino
, int iomode
)
1437 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1438 struct nfs_inode
*nfsi
= NFS_I(ino
);
1439 loff_t fsize
= i_size_read(ino
);
1440 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1445 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1446 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1450 if (t
->bm
& THRESHOLD_RD
) {
1451 dprintk("%s fsize %llu\n", __func__
, fsize
);
1453 if (fsize
< t
->rd_sz
)
1456 if (t
->bm
& THRESHOLD_RD_IO
) {
1457 dprintk("%s nfsi->read_io %llu\n", __func__
,
1460 if (nfsi
->read_io
< t
->rd_io_sz
)
1465 if (t
->bm
& THRESHOLD_WR
) {
1466 dprintk("%s fsize %llu\n", __func__
, fsize
);
1468 if (fsize
< t
->wr_sz
)
1471 if (t
->bm
& THRESHOLD_WR_IO
) {
1472 dprintk("%s nfsi->write_io %llu\n", __func__
,
1475 if (nfsi
->write_io
< t
->wr_io_sz
)
1480 if (size_set
&& io_set
) {
1483 } else if (size
|| io
)
1486 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1490 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr
*lo
)
1493 * send layoutcommit as it can hold up layoutreturn due to lseg
1496 pnfs_layoutcommit_inode(lo
->plh_inode
, false);
1497 return !wait_on_bit_action(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1498 nfs_wait_bit_killable
,
1499 TASK_UNINTERRUPTIBLE
);
1502 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr
*lo
)
1504 unsigned long *bitlock
= &lo
->plh_flags
;
1506 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET
, bitlock
);
1507 smp_mb__after_atomic();
1508 wake_up_bit(bitlock
, NFS_LAYOUT_FIRST_LAYOUTGET
);
1512 * Layout segment is retreived from the server if not cached.
1513 * The appropriate layout segment is referenced and returned to the caller.
1515 struct pnfs_layout_segment
*
1516 pnfs_update_layout(struct inode
*ino
,
1517 struct nfs_open_context
*ctx
,
1520 enum pnfs_iomode iomode
,
1524 struct pnfs_layout_range arg
= {
1529 unsigned pg_offset
, seq
;
1530 struct nfs_server
*server
= NFS_SERVER(ino
);
1531 struct nfs_client
*clp
= server
->nfs_client
;
1532 struct pnfs_layout_hdr
*lo
= NULL
;
1533 struct pnfs_layout_segment
*lseg
= NULL
;
1534 nfs4_stateid stateid
;
1536 unsigned long giveup
= jiffies
+ (clp
->cl_lease_time
<< 1);
1539 if (!pnfs_enabled_sb(NFS_SERVER(ino
))) {
1540 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1541 PNFS_UPDATE_LAYOUT_NO_PNFS
);
1545 if (iomode
== IOMODE_READ
&& i_size_read(ino
) == 0) {
1546 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1547 PNFS_UPDATE_LAYOUT_RD_ZEROLEN
);
1551 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
)) {
1552 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1553 PNFS_UPDATE_LAYOUT_MDSTHRESH
);
1558 nfs4_client_recover_expired_lease(clp
);
1560 spin_lock(&ino
->i_lock
);
1561 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1563 spin_unlock(&ino
->i_lock
);
1564 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1565 PNFS_UPDATE_LAYOUT_NOMEM
);
1569 /* Do we even need to bother with this? */
1570 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1571 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1572 PNFS_UPDATE_LAYOUT_BULK_RECALL
);
1573 dprintk("%s matches recall, use MDS\n", __func__
);
1577 /* if LAYOUTGET already failed once we don't try again */
1578 if (pnfs_layout_io_test_failed(lo
, iomode
)) {
1579 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1580 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL
);
1584 lseg
= pnfs_find_lseg(lo
, &arg
, strict_iomode
);
1586 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1587 PNFS_UPDATE_LAYOUT_FOUND_CACHED
);
1591 if (!nfs4_valid_open_stateid(ctx
->state
)) {
1592 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1593 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1598 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1599 * stateid, or it has been invalidated, then we must use the open
1602 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
)) {
1605 * The first layoutget for the file. Need to serialize per
1606 * RFC 5661 Errata 3208.
1608 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
,
1610 spin_unlock(&ino
->i_lock
);
1611 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_FIRST_LAYOUTGET
,
1612 TASK_UNINTERRUPTIBLE
);
1613 pnfs_put_layout_hdr(lo
);
1614 dprintk("%s retrying\n", __func__
);
1620 seq
= read_seqbegin(&ctx
->state
->seqlock
);
1621 nfs4_stateid_copy(&stateid
, &ctx
->state
->stateid
);
1622 } while (read_seqretry(&ctx
->state
->seqlock
, seq
));
1624 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1628 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1629 * for LAYOUTRETURN even if first is true.
1631 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1632 spin_unlock(&ino
->i_lock
);
1633 dprintk("%s wait for layoutreturn\n", __func__
);
1634 if (pnfs_prepare_to_retry_layoutget(lo
)) {
1636 pnfs_clear_first_layoutget(lo
);
1637 pnfs_put_layout_hdr(lo
);
1638 dprintk("%s retrying\n", __func__
);
1639 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
,
1640 lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1643 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1644 PNFS_UPDATE_LAYOUT_RETURN
);
1645 goto out_put_layout_hdr
;
1648 if (pnfs_layoutgets_blocked(lo
)) {
1649 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1650 PNFS_UPDATE_LAYOUT_BLOCKED
);
1653 atomic_inc(&lo
->plh_outstanding
);
1654 spin_unlock(&ino
->i_lock
);
1656 if (list_empty(&lo
->plh_layouts
)) {
1657 /* The lo must be on the clp list if there is any
1658 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1660 spin_lock(&clp
->cl_lock
);
1661 if (list_empty(&lo
->plh_layouts
))
1662 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1663 spin_unlock(&clp
->cl_lock
);
1666 pg_offset
= arg
.offset
& ~PAGE_MASK
;
1668 arg
.offset
-= pg_offset
;
1669 arg
.length
+= pg_offset
;
1671 if (arg
.length
!= NFS4_MAX_UINT64
)
1672 arg
.length
= PAGE_ALIGN(arg
.length
);
1674 lseg
= send_layoutget(lo
, ctx
, &stateid
, &arg
, &timeout
, gfp_flags
);
1675 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1676 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET
);
1677 atomic_dec(&lo
->plh_outstanding
);
1679 switch(PTR_ERR(lseg
)) {
1681 if (time_after(jiffies
, giveup
))
1684 case -ERECALLCONFLICT
:
1685 /* Huh? We hold no layouts, how is there a recall? */
1690 /* Destroy the existing layout and start over */
1691 if (time_after(jiffies
, giveup
))
1692 pnfs_destroy_layout(NFS_I(ino
));
1697 if (!nfs_error_is_fatal(PTR_ERR(lseg
))) {
1698 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1701 goto out_put_layout_hdr
;
1705 pnfs_clear_first_layoutget(lo
);
1706 trace_pnfs_update_layout(ino
, pos
, count
,
1707 iomode
, lo
, lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1708 pnfs_put_layout_hdr(lo
);
1712 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1717 pnfs_clear_first_layoutget(lo
);
1718 pnfs_put_layout_hdr(lo
);
1720 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1721 "(%s, offset: %llu, length: %llu)\n",
1722 __func__
, ino
->i_sb
->s_id
,
1723 (unsigned long long)NFS_FILEID(ino
),
1724 IS_ERR_OR_NULL(lseg
) ? "not found" : "found",
1725 iomode
==IOMODE_RW
? "read/write" : "read-only",
1726 (unsigned long long)pos
,
1727 (unsigned long long)count
);
1730 spin_unlock(&ino
->i_lock
);
1731 goto out_put_layout_hdr
;
1733 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1736 pnfs_sanity_check_layout_range(struct pnfs_layout_range
*range
)
1738 switch (range
->iomode
) {
1745 if (range
->offset
== NFS4_MAX_UINT64
)
1747 if (range
->length
== 0)
1749 if (range
->length
!= NFS4_MAX_UINT64
&&
1750 range
->length
> NFS4_MAX_UINT64
- range
->offset
)
1755 struct pnfs_layout_segment
*
1756 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1758 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1759 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1760 struct pnfs_layout_segment
*lseg
;
1761 struct inode
*ino
= lo
->plh_inode
;
1764 if (!pnfs_sanity_check_layout_range(&res
->range
))
1765 return ERR_PTR(-EINVAL
);
1767 /* Inject layout blob into I/O device driver */
1768 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1769 if (IS_ERR_OR_NULL(lseg
)) {
1771 lseg
= ERR_PTR(-ENOMEM
);
1773 dprintk("%s: Could not allocate layout: error %ld\n",
1774 __func__
, PTR_ERR(lseg
));
1778 pnfs_init_lseg(lo
, lseg
, &res
->range
, &res
->stateid
);
1780 spin_lock(&ino
->i_lock
);
1781 if (pnfs_layoutgets_blocked(lo
)) {
1782 dprintk("%s forget reply due to state\n", __func__
);
1786 if (nfs4_stateid_match_other(&lo
->plh_stateid
, &res
->stateid
)) {
1787 /* existing state ID, make sure the sequence number matches. */
1788 if (pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
1789 dprintk("%s forget reply due to sequence\n", __func__
);
1792 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1795 * We got an entirely new state ID. Mark all segments for the
1796 * inode invalid, and don't bother validating the stateid
1799 pnfs_mark_layout_stateid_invalid(lo
, &free_me
);
1801 nfs4_stateid_copy(&lo
->plh_stateid
, &res
->stateid
);
1802 lo
->plh_barrier
= be32_to_cpu(res
->stateid
.seqid
);
1805 pnfs_get_lseg(lseg
);
1806 pnfs_layout_insert_lseg(lo
, lseg
, &free_me
);
1807 if (!pnfs_layout_is_valid(lo
)) {
1808 pnfs_clear_layoutreturn_info(lo
);
1809 clear_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
1813 if (res
->return_on_close
)
1814 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1816 spin_unlock(&ino
->i_lock
);
1817 pnfs_free_lseg_list(&free_me
);
1821 spin_unlock(&ino
->i_lock
);
1822 lseg
->pls_layout
= lo
;
1823 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1824 return ERR_PTR(-EAGAIN
);
1828 pnfs_set_plh_return_info(struct pnfs_layout_hdr
*lo
, enum pnfs_iomode iomode
,
1831 if (lo
->plh_return_iomode
!= 0 && lo
->plh_return_iomode
!= iomode
)
1832 iomode
= IOMODE_ANY
;
1833 lo
->plh_return_iomode
= iomode
;
1834 set_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
1836 WARN_ON_ONCE(lo
->plh_return_seq
!= 0 && lo
->plh_return_seq
!= seq
);
1837 lo
->plh_return_seq
= seq
;
1842 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
1843 * @lo: pointer to layout header
1844 * @tmp_list: list header to be used with pnfs_free_lseg_list()
1845 * @return_range: describe layout segment ranges to be returned
1847 * This function is mainly intended for use by layoutrecall. It attempts
1848 * to free the layout segment immediately, or else to mark it for return
1849 * as soon as its reference count drops to zero.
1852 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr
*lo
,
1853 struct list_head
*tmp_list
,
1854 const struct pnfs_layout_range
*return_range
,
1857 struct pnfs_layout_segment
*lseg
, *next
;
1860 dprintk("%s:Begin lo %p\n", __func__
, lo
);
1862 if (list_empty(&lo
->plh_segs
))
1865 assert_spin_locked(&lo
->plh_inode
->i_lock
);
1867 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
1868 if (pnfs_match_lseg_recall(lseg
, return_range
, seq
)) {
1869 dprintk("%s: marking lseg %p iomode %d "
1870 "offset %llu length %llu\n", __func__
,
1871 lseg
, lseg
->pls_range
.iomode
,
1872 lseg
->pls_range
.offset
,
1873 lseg
->pls_range
.length
);
1874 if (mark_lseg_invalid(lseg
, tmp_list
))
1877 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
1881 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
1886 void pnfs_error_mark_layout_for_return(struct inode
*inode
,
1887 struct pnfs_layout_segment
*lseg
)
1889 struct pnfs_layout_hdr
*lo
= NFS_I(inode
)->layout
;
1890 struct pnfs_layout_range range
= {
1891 .iomode
= lseg
->pls_range
.iomode
,
1893 .length
= NFS4_MAX_UINT64
,
1896 bool return_now
= false;
1898 spin_lock(&inode
->i_lock
);
1899 pnfs_set_plh_return_info(lo
, range
.iomode
, 0);
1901 * mark all matching lsegs so that we are sure to have no live
1902 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1905 if (!pnfs_mark_matching_lsegs_return(lo
, &free_me
, &range
, 0)) {
1906 nfs4_stateid stateid
;
1907 enum pnfs_iomode iomode
;
1909 return_now
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
1910 spin_unlock(&inode
->i_lock
);
1912 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
1914 spin_unlock(&inode
->i_lock
);
1915 nfs_commit_inode(inode
, 0);
1917 pnfs_free_lseg_list(&free_me
);
1919 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return
);
1922 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1924 u64 rd_size
= req
->wb_bytes
;
1926 if (pgio
->pg_lseg
== NULL
) {
1927 if (pgio
->pg_dreq
== NULL
)
1928 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
1930 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
1932 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1939 if (IS_ERR(pgio
->pg_lseg
)) {
1940 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
1941 pgio
->pg_lseg
= NULL
;
1945 /* If no lseg, fall back to read through mds */
1946 if (pgio
->pg_lseg
== NULL
)
1947 nfs_pageio_reset_read_mds(pgio
);
1950 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1953 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
1954 struct nfs_page
*req
, u64 wb_size
)
1956 if (pgio
->pg_lseg
== NULL
) {
1957 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1964 if (IS_ERR(pgio
->pg_lseg
)) {
1965 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
1966 pgio
->pg_lseg
= NULL
;
1970 /* If no lseg, fall back to write through mds */
1971 if (pgio
->pg_lseg
== NULL
)
1972 nfs_pageio_reset_write_mds(pgio
);
1974 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1977 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor
*desc
)
1979 if (desc
->pg_lseg
) {
1980 pnfs_put_lseg(desc
->pg_lseg
);
1981 desc
->pg_lseg
= NULL
;
1984 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup
);
1987 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1988 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1991 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
,
1992 struct nfs_page
*prev
, struct nfs_page
*req
)
1995 u64 seg_end
, req_start
, seg_left
;
1997 size
= nfs_generic_pg_test(pgio
, prev
, req
);
2002 * 'size' contains the number of bytes left in the current page (up
2003 * to the original size asked for in @req->wb_bytes).
2005 * Calculate how many bytes are left in the layout segment
2006 * and if there are less bytes than 'size', return that instead.
2008 * Please also note that 'end_offset' is actually the offset of the
2009 * first byte that lies outside the pnfs_layout_range. FIXME?
2012 if (pgio
->pg_lseg
) {
2013 seg_end
= end_offset(pgio
->pg_lseg
->pls_range
.offset
,
2014 pgio
->pg_lseg
->pls_range
.length
);
2015 req_start
= req_offset(req
);
2016 WARN_ON_ONCE(req_start
>= seg_end
);
2017 /* start of request is past the last byte of this segment */
2018 if (req_start
>= seg_end
) {
2019 /* reference the new lseg */
2020 if (pgio
->pg_ops
->pg_cleanup
)
2021 pgio
->pg_ops
->pg_cleanup(pgio
);
2022 if (pgio
->pg_ops
->pg_init
)
2023 pgio
->pg_ops
->pg_init(pgio
, req
);
2027 /* adjust 'size' iff there are fewer bytes left in the
2028 * segment than what nfs_generic_pg_test returned */
2029 seg_left
= seg_end
- req_start
;
2030 if (seg_left
< size
)
2031 size
= (unsigned int)seg_left
;
2036 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
2038 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2040 struct nfs_pageio_descriptor pgio
;
2042 /* Resend all requests through the MDS */
2043 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
2044 hdr
->completion_ops
);
2045 set_bit(NFS_CONTEXT_RESEND_WRITES
, &hdr
->args
.context
->flags
);
2046 return nfs_pageio_resend(&pgio
, hdr
);
2048 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
2050 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
2053 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
2054 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2055 PNFS_LAYOUTRET_ON_ERROR
) {
2056 pnfs_return_layout(hdr
->inode
);
2058 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2059 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
2063 * Called by non rpc-based layout drivers
2065 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
2067 if (likely(!hdr
->pnfs_error
)) {
2068 pnfs_set_layoutcommit(hdr
->inode
, hdr
->lseg
,
2069 hdr
->mds_offset
+ hdr
->res
.count
);
2070 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2072 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
2073 if (unlikely(hdr
->pnfs_error
))
2074 pnfs_ld_handle_write_error(hdr
);
2075 hdr
->mds_ops
->rpc_release(hdr
);
2077 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
2080 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
2081 struct nfs_pgio_header
*hdr
)
2083 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2085 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2086 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2087 nfs_pageio_reset_write_mds(desc
);
2088 mirror
->pg_recoalesce
= 1;
2090 nfs_pgio_data_destroy(hdr
);
2094 static enum pnfs_try_status
2095 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
2096 const struct rpc_call_ops
*call_ops
,
2097 struct pnfs_layout_segment
*lseg
,
2100 struct inode
*inode
= hdr
->inode
;
2101 enum pnfs_try_status trypnfs
;
2102 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2104 hdr
->mds_ops
= call_ops
;
2106 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
2107 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
2108 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
2109 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2110 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
2111 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2116 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
2117 struct nfs_pgio_header
*hdr
, int how
)
2119 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2120 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2121 enum pnfs_try_status trypnfs
;
2123 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
2124 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
2125 pnfs_write_through_mds(desc
, hdr
);
2128 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
2130 pnfs_put_lseg(hdr
->lseg
);
2131 nfs_pgio_header_free(hdr
);
2135 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
2137 struct nfs_pgio_header
*hdr
;
2140 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2142 desc
->pg_error
= -ENOMEM
;
2143 return desc
->pg_error
;
2145 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
2147 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2148 ret
= nfs_generic_pgio(desc
, hdr
);
2150 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
2154 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
2156 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2158 struct nfs_pageio_descriptor pgio
;
2160 /* Resend all requests through the MDS */
2161 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
2162 return nfs_pageio_resend(&pgio
, hdr
);
2164 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
2166 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
2168 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
2169 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2170 PNFS_LAYOUTRET_ON_ERROR
) {
2171 pnfs_return_layout(hdr
->inode
);
2173 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2174 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
2178 * Called by non rpc-based layout drivers
2180 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
2182 if (likely(!hdr
->pnfs_error
)) {
2183 __nfs4_read_done_cb(hdr
);
2184 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2186 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
2187 if (unlikely(hdr
->pnfs_error
))
2188 pnfs_ld_handle_read_error(hdr
);
2189 hdr
->mds_ops
->rpc_release(hdr
);
2191 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
2194 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
2195 struct nfs_pgio_header
*hdr
)
2197 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2199 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2200 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2201 nfs_pageio_reset_read_mds(desc
);
2202 mirror
->pg_recoalesce
= 1;
2204 nfs_pgio_data_destroy(hdr
);
2209 * Call the appropriate parallel I/O subsystem read function.
2211 static enum pnfs_try_status
2212 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
2213 const struct rpc_call_ops
*call_ops
,
2214 struct pnfs_layout_segment
*lseg
)
2216 struct inode
*inode
= hdr
->inode
;
2217 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2218 enum pnfs_try_status trypnfs
;
2220 hdr
->mds_ops
= call_ops
;
2222 dprintk("%s: Reading ino:%lu %u@%llu\n",
2223 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
2225 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
2226 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2227 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
2228 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2232 /* Resend all requests through pnfs. */
2233 void pnfs_read_resend_pnfs(struct nfs_pgio_header
*hdr
)
2235 struct nfs_pageio_descriptor pgio
;
2237 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2238 nfs_pageio_init_read(&pgio
, hdr
->inode
, false,
2239 hdr
->completion_ops
);
2240 hdr
->task
.tk_status
= nfs_pageio_resend(&pgio
, hdr
);
2243 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs
);
2246 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
2248 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2249 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2250 enum pnfs_try_status trypnfs
;
2252 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
2253 if (trypnfs
== PNFS_TRY_AGAIN
)
2254 pnfs_read_resend_pnfs(hdr
);
2255 if (trypnfs
== PNFS_NOT_ATTEMPTED
|| hdr
->task
.tk_status
)
2256 pnfs_read_through_mds(desc
, hdr
);
2259 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
2261 pnfs_put_lseg(hdr
->lseg
);
2262 nfs_pgio_header_free(hdr
);
2266 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
2268 struct nfs_pgio_header
*hdr
;
2271 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2273 desc
->pg_error
= -ENOMEM
;
2274 return desc
->pg_error
;
2276 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
2277 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2278 ret
= nfs_generic_pgio(desc
, hdr
);
2280 pnfs_do_read(desc
, hdr
);
2283 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
2285 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
2287 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
2289 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
2290 smp_mb__after_atomic();
2291 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
2295 * There can be multiple RW segments.
2297 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
2299 struct pnfs_layout_segment
*lseg
;
2301 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
2302 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
2303 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
2304 list_add(&lseg
->pls_lc_list
, listp
);
2308 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
2310 struct pnfs_layout_segment
*lseg
, *tmp
;
2312 /* Matched by references in pnfs_set_layoutcommit */
2313 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
2314 list_del_init(&lseg
->pls_lc_list
);
2315 pnfs_put_lseg(lseg
);
2318 pnfs_clear_layoutcommitting(inode
);
2321 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
2323 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
2325 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
2328 pnfs_set_layoutcommit(struct inode
*inode
, struct pnfs_layout_segment
*lseg
,
2331 struct nfs_inode
*nfsi
= NFS_I(inode
);
2332 bool mark_as_dirty
= false;
2334 spin_lock(&inode
->i_lock
);
2335 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
2336 nfsi
->layout
->plh_lwb
= end_pos
;
2337 mark_as_dirty
= true;
2338 dprintk("%s: Set layoutcommit for inode %lu ",
2339 __func__
, inode
->i_ino
);
2340 } else if (end_pos
> nfsi
->layout
->plh_lwb
)
2341 nfsi
->layout
->plh_lwb
= end_pos
;
2342 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
)) {
2343 /* references matched in nfs4_layoutcommit_release */
2344 pnfs_get_lseg(lseg
);
2346 spin_unlock(&inode
->i_lock
);
2347 dprintk("%s: lseg %p end_pos %llu\n",
2348 __func__
, lseg
, nfsi
->layout
->plh_lwb
);
2350 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2351 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2353 mark_inode_dirty_sync(inode
);
2355 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
2357 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
2359 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
2361 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
2362 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
2363 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
2367 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2368 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2369 * data to disk to allow the server to recover the data if it crashes.
2370 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2371 * is off, and a COMMIT is sent to a data server, or
2372 * if WRITEs to a data server return NFS_DATA_SYNC.
2375 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
2377 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2378 struct nfs4_layoutcommit_data
*data
;
2379 struct nfs_inode
*nfsi
= NFS_I(inode
);
2383 if (!pnfs_layoutcommit_outstanding(inode
))
2386 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
2389 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
2392 status
= wait_on_bit_lock_action(&nfsi
->flags
,
2393 NFS_INO_LAYOUTCOMMITTING
,
2394 nfs_wait_bit_killable
,
2401 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2402 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
2404 goto clear_layoutcommitting
;
2407 spin_lock(&inode
->i_lock
);
2408 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
2411 INIT_LIST_HEAD(&data
->lseg_list
);
2412 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
2414 end_pos
= nfsi
->layout
->plh_lwb
;
2416 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
2417 spin_unlock(&inode
->i_lock
);
2419 data
->args
.inode
= inode
;
2420 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
2421 nfs_fattr_init(&data
->fattr
);
2422 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
2423 data
->res
.fattr
= &data
->fattr
;
2425 data
->args
.lastbytewritten
= end_pos
- 1;
2427 data
->args
.lastbytewritten
= U64_MAX
;
2428 data
->res
.server
= NFS_SERVER(inode
);
2430 if (ld
->prepare_layoutcommit
) {
2431 status
= ld
->prepare_layoutcommit(&data
->args
);
2433 put_rpccred(data
->cred
);
2434 spin_lock(&inode
->i_lock
);
2435 set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
);
2436 if (end_pos
> nfsi
->layout
->plh_lwb
)
2437 nfsi
->layout
->plh_lwb
= end_pos
;
2443 status
= nfs4_proc_layoutcommit(data
, sync
);
2446 mark_inode_dirty_sync(inode
);
2447 dprintk("<-- %s status %d\n", __func__
, status
);
2450 spin_unlock(&inode
->i_lock
);
2452 clear_layoutcommitting
:
2453 pnfs_clear_layoutcommitting(inode
);
2456 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode
);
2459 pnfs_generic_sync(struct inode
*inode
, bool datasync
)
2461 return pnfs_layoutcommit_inode(inode
, true);
2463 EXPORT_SYMBOL_GPL(pnfs_generic_sync
);
2465 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
2467 struct nfs4_threshold
*thp
;
2469 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
2471 dprintk("%s mdsthreshold allocation failed\n", __func__
);
2477 #if IS_ENABLED(CONFIG_NFS_V4_2)
2479 pnfs_report_layoutstat(struct inode
*inode
, gfp_t gfp_flags
)
2481 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2482 struct nfs_server
*server
= NFS_SERVER(inode
);
2483 struct nfs_inode
*nfsi
= NFS_I(inode
);
2484 struct nfs42_layoutstat_data
*data
;
2485 struct pnfs_layout_hdr
*hdr
;
2488 if (!pnfs_enabled_sb(server
) || !ld
->prepare_layoutstats
)
2491 if (!nfs_server_capable(inode
, NFS_CAP_LAYOUTSTATS
))
2494 if (test_and_set_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
))
2497 spin_lock(&inode
->i_lock
);
2498 if (!NFS_I(inode
)->layout
) {
2499 spin_unlock(&inode
->i_lock
);
2500 goto out_clear_layoutstats
;
2502 hdr
= NFS_I(inode
)->layout
;
2503 pnfs_get_layout_hdr(hdr
);
2504 spin_unlock(&inode
->i_lock
);
2506 data
= kzalloc(sizeof(*data
), gfp_flags
);
2512 data
->args
.fh
= NFS_FH(inode
);
2513 data
->args
.inode
= inode
;
2514 status
= ld
->prepare_layoutstats(&data
->args
);
2518 status
= nfs42_proc_layoutstats_generic(NFS_SERVER(inode
), data
);
2521 dprintk("%s returns %d\n", __func__
, status
);
2527 pnfs_put_layout_hdr(hdr
);
2528 out_clear_layoutstats
:
2529 smp_mb__before_atomic();
2530 clear_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
);
2531 smp_mb__after_atomic();
2534 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat
);
2537 unsigned int layoutstats_timer
;
2538 module_param(layoutstats_timer
, uint
, 0644);
2539 EXPORT_SYMBOL_GPL(layoutstats_timer
);