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>
33 #include <linux/sort.h>
37 #include "nfs4trace.h"
38 #include "delegation.h"
42 #define NFSDBG_FACILITY NFSDBG_PNFS
43 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
48 * protects pnfs_modules_tbl.
50 static DEFINE_SPINLOCK(pnfs_spinlock
);
53 * pnfs_modules_tbl holds all pnfs modules
55 static LIST_HEAD(pnfs_modules_tbl
);
57 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
);
58 static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
59 struct list_head
*free_me
,
60 const struct pnfs_layout_range
*range
,
62 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
63 struct list_head
*tmp_list
);
65 /* Return the registered pnfs layout driver module matching given id */
66 static struct pnfs_layoutdriver_type
*
67 find_pnfs_driver_locked(u32 id
)
69 struct pnfs_layoutdriver_type
*local
;
71 list_for_each_entry(local
, &pnfs_modules_tbl
, pnfs_tblid
)
76 dprintk("%s: Searching for id %u, found %p\n", __func__
, id
, local
);
80 static struct pnfs_layoutdriver_type
*
81 find_pnfs_driver(u32 id
)
83 struct pnfs_layoutdriver_type
*local
;
85 spin_lock(&pnfs_spinlock
);
86 local
= find_pnfs_driver_locked(id
);
87 if (local
!= NULL
&& !try_module_get(local
->owner
)) {
88 dprintk("%s: Could not grab reference on module\n", __func__
);
91 spin_unlock(&pnfs_spinlock
);
96 unset_pnfs_layoutdriver(struct nfs_server
*nfss
)
98 if (nfss
->pnfs_curr_ld
) {
99 if (nfss
->pnfs_curr_ld
->clear_layoutdriver
)
100 nfss
->pnfs_curr_ld
->clear_layoutdriver(nfss
);
101 /* Decrement the MDS count. Purge the deviceid cache if zero */
102 if (atomic_dec_and_test(&nfss
->nfs_client
->cl_mds_count
))
103 nfs4_deviceid_purge_client(nfss
->nfs_client
);
104 module_put(nfss
->pnfs_curr_ld
->owner
);
106 nfss
->pnfs_curr_ld
= NULL
;
110 * When the server sends a list of layout types, we choose one in the order
111 * given in the list below.
113 * FIXME: should this list be configurable in some fashion? module param?
114 * mount option? something else?
116 static const u32 ld_prefs
[] = {
121 LAYOUT_NFSV4_1_FILES
,
126 ld_cmp(const void *e1
, const void *e2
)
128 u32 ld1
= *((u32
*)e1
);
129 u32 ld2
= *((u32
*)e2
);
132 for (i
= 0; ld_prefs
[i
] != 0; i
++) {
133 if (ld1
== ld_prefs
[i
])
136 if (ld2
== ld_prefs
[i
])
143 * Try to set the server's pnfs module to the pnfs layout type specified by id.
144 * Currently only one pNFS layout driver per filesystem is supported.
146 * @ids array of layout types supported by MDS.
149 set_pnfs_layoutdriver(struct nfs_server
*server
, const struct nfs_fh
*mntfh
,
150 struct nfs_fsinfo
*fsinfo
)
152 struct pnfs_layoutdriver_type
*ld_type
= NULL
;
156 if (fsinfo
->nlayouttypes
== 0)
158 if (!(server
->nfs_client
->cl_exchange_flags
&
159 (EXCHGID4_FLAG_USE_NON_PNFS
| EXCHGID4_FLAG_USE_PNFS_MDS
))) {
160 printk(KERN_ERR
"NFS: %s: cl_exchange_flags 0x%x\n",
161 __func__
, server
->nfs_client
->cl_exchange_flags
);
165 sort(fsinfo
->layouttype
, fsinfo
->nlayouttypes
,
166 sizeof(*fsinfo
->layouttype
), ld_cmp
, NULL
);
168 for (i
= 0; i
< fsinfo
->nlayouttypes
; i
++) {
169 id
= fsinfo
->layouttype
[i
];
170 ld_type
= find_pnfs_driver(id
);
172 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX
,
174 ld_type
= find_pnfs_driver(id
);
181 dprintk("%s: No pNFS module found!\n", __func__
);
185 server
->pnfs_curr_ld
= ld_type
;
186 if (ld_type
->set_layoutdriver
187 && ld_type
->set_layoutdriver(server
, mntfh
)) {
188 printk(KERN_ERR
"NFS: %s: Error initializing pNFS layout "
189 "driver %u.\n", __func__
, id
);
190 module_put(ld_type
->owner
);
193 /* Bump the MDS count */
194 atomic_inc(&server
->nfs_client
->cl_mds_count
);
196 dprintk("%s: pNFS module for %u set\n", __func__
, id
);
200 dprintk("%s: Using NFSv4 I/O\n", __func__
);
201 server
->pnfs_curr_ld
= NULL
;
205 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
207 int status
= -EINVAL
;
208 struct pnfs_layoutdriver_type
*tmp
;
210 if (ld_type
->id
== 0) {
211 printk(KERN_ERR
"NFS: %s id 0 is reserved\n", __func__
);
214 if (!ld_type
->alloc_lseg
|| !ld_type
->free_lseg
) {
215 printk(KERN_ERR
"NFS: %s Layout driver must provide "
216 "alloc_lseg and free_lseg.\n", __func__
);
220 spin_lock(&pnfs_spinlock
);
221 tmp
= find_pnfs_driver_locked(ld_type
->id
);
223 list_add(&ld_type
->pnfs_tblid
, &pnfs_modules_tbl
);
225 dprintk("%s Registering id:%u name:%s\n", __func__
, ld_type
->id
,
228 printk(KERN_ERR
"NFS: %s Module with id %d already loaded!\n",
229 __func__
, ld_type
->id
);
231 spin_unlock(&pnfs_spinlock
);
235 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver
);
238 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type
*ld_type
)
240 dprintk("%s Deregistering id:%u\n", __func__
, ld_type
->id
);
241 spin_lock(&pnfs_spinlock
);
242 list_del(&ld_type
->pnfs_tblid
);
243 spin_unlock(&pnfs_spinlock
);
245 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver
);
248 * pNFS client layout cache
251 /* Need to hold i_lock if caller does not already hold reference */
253 pnfs_get_layout_hdr(struct pnfs_layout_hdr
*lo
)
255 refcount_inc(&lo
->plh_refcount
);
258 static struct pnfs_layout_hdr
*
259 pnfs_alloc_layout_hdr(struct inode
*ino
, gfp_t gfp_flags
)
261 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
262 return ld
->alloc_layout_hdr(ino
, gfp_flags
);
266 pnfs_free_layout_hdr(struct pnfs_layout_hdr
*lo
)
268 struct nfs_server
*server
= NFS_SERVER(lo
->plh_inode
);
269 struct pnfs_layoutdriver_type
*ld
= server
->pnfs_curr_ld
;
271 if (!list_empty(&lo
->plh_layouts
)) {
272 struct nfs_client
*clp
= server
->nfs_client
;
274 spin_lock(&clp
->cl_lock
);
275 list_del_init(&lo
->plh_layouts
);
276 spin_unlock(&clp
->cl_lock
);
278 put_cred(lo
->plh_lc_cred
);
279 return ld
->free_layout_hdr(lo
);
283 pnfs_detach_layout_hdr(struct pnfs_layout_hdr
*lo
)
285 struct nfs_inode
*nfsi
= NFS_I(lo
->plh_inode
);
286 dprintk("%s: freeing layout cache %p\n", __func__
, lo
);
288 /* Reset MDS Threshold I/O counters */
294 pnfs_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
300 inode
= lo
->plh_inode
;
301 pnfs_layoutreturn_before_put_layout_hdr(lo
);
303 if (refcount_dec_and_lock(&lo
->plh_refcount
, &inode
->i_lock
)) {
304 if (!list_empty(&lo
->plh_segs
))
305 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
306 pnfs_detach_layout_hdr(lo
);
307 spin_unlock(&inode
->i_lock
);
308 pnfs_free_layout_hdr(lo
);
313 pnfs_set_plh_return_info(struct pnfs_layout_hdr
*lo
, enum pnfs_iomode iomode
,
316 if (lo
->plh_return_iomode
!= 0 && lo
->plh_return_iomode
!= iomode
)
318 lo
->plh_return_iomode
= iomode
;
319 set_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
321 WARN_ON_ONCE(lo
->plh_return_seq
!= 0 && lo
->plh_return_seq
!= seq
);
322 lo
->plh_return_seq
= seq
;
327 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr
*lo
)
329 struct pnfs_layout_segment
*lseg
;
330 lo
->plh_return_iomode
= 0;
331 lo
->plh_return_seq
= 0;
332 clear_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
333 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
334 if (!test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
336 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
340 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr
*lo
)
342 clear_bit_unlock(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
343 clear_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
);
344 smp_mb__after_atomic();
345 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
);
346 rpc_wake_up(&NFS_SERVER(lo
->plh_inode
)->roc_rpcwaitq
);
350 pnfs_clear_lseg_state(struct pnfs_layout_segment
*lseg
,
351 struct list_head
*free_me
)
353 clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
354 clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
355 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
356 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
357 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
358 pnfs_lseg_dec_and_remove_zero(lseg
, free_me
);
362 * Update the seqid of a layout stateid
364 bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid
*dst
,
365 struct pnfs_layout_range
*dst_range
,
368 struct pnfs_layout_hdr
*lo
;
369 struct pnfs_layout_range range
= {
370 .iomode
= IOMODE_ANY
,
372 .length
= NFS4_MAX_UINT64
,
378 spin_lock(&inode
->i_lock
);
379 lo
= NFS_I(inode
)->layout
;
380 if (lo
&& nfs4_stateid_match_other(dst
, &lo
->plh_stateid
)) {
381 err
= pnfs_mark_matching_lsegs_return(lo
, &head
, &range
, 0);
383 dst
->seqid
= lo
->plh_stateid
.seqid
;
388 spin_unlock(&inode
->i_lock
);
389 pnfs_free_lseg_list(&head
);
394 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
396 * In order to continue using the pnfs_layout_hdr, a full recovery
398 * Note that caller must hold inode->i_lock.
401 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr
*lo
,
402 struct list_head
*lseg_list
)
404 struct pnfs_layout_range range
= {
405 .iomode
= IOMODE_ANY
,
407 .length
= NFS4_MAX_UINT64
,
409 struct pnfs_layout_segment
*lseg
, *next
;
411 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
412 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
413 pnfs_clear_lseg_state(lseg
, lseg_list
);
414 pnfs_clear_layoutreturn_info(lo
);
415 pnfs_free_returned_lsegs(lo
, lseg_list
, &range
, 0);
416 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
) &&
417 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
418 pnfs_clear_layoutreturn_waitbit(lo
);
419 return !list_empty(&lo
->plh_segs
);
423 pnfs_iomode_to_fail_bit(u32 iomode
)
425 return iomode
== IOMODE_RW
?
426 NFS_LAYOUT_RW_FAILED
: NFS_LAYOUT_RO_FAILED
;
430 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
432 lo
->plh_retry_timestamp
= jiffies
;
433 if (!test_and_set_bit(fail_bit
, &lo
->plh_flags
))
434 refcount_inc(&lo
->plh_refcount
);
438 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr
*lo
, int fail_bit
)
440 if (test_and_clear_bit(fail_bit
, &lo
->plh_flags
))
441 refcount_dec(&lo
->plh_refcount
);
445 pnfs_layout_io_set_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
447 struct inode
*inode
= lo
->plh_inode
;
448 struct pnfs_layout_range range
= {
451 .length
= NFS4_MAX_UINT64
,
455 spin_lock(&inode
->i_lock
);
456 pnfs_layout_set_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
457 pnfs_mark_matching_lsegs_invalid(lo
, &head
, &range
, 0);
458 spin_unlock(&inode
->i_lock
);
459 pnfs_free_lseg_list(&head
);
460 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__
,
461 iomode
== IOMODE_RW
? "RW" : "READ");
465 pnfs_layout_io_test_failed(struct pnfs_layout_hdr
*lo
, u32 iomode
)
467 unsigned long start
, end
;
468 int fail_bit
= pnfs_iomode_to_fail_bit(iomode
);
470 if (test_bit(fail_bit
, &lo
->plh_flags
) == 0)
473 start
= end
- PNFS_LAYOUTGET_RETRY_TIMEOUT
;
474 if (!time_in_range(lo
->plh_retry_timestamp
, start
, end
)) {
475 /* It is time to retry the failed layoutgets */
476 pnfs_layout_clear_fail_bit(lo
, fail_bit
);
483 pnfs_init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
,
484 const struct pnfs_layout_range
*range
,
485 const nfs4_stateid
*stateid
)
487 INIT_LIST_HEAD(&lseg
->pls_list
);
488 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
489 refcount_set(&lseg
->pls_refcount
, 1);
490 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
491 lseg
->pls_layout
= lo
;
492 lseg
->pls_range
= *range
;
493 lseg
->pls_seq
= be32_to_cpu(stateid
->seqid
);
496 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
499 struct inode
*inode
= lseg
->pls_layout
->plh_inode
;
500 NFS_SERVER(inode
)->pnfs_curr_ld
->free_lseg(lseg
);
505 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
506 struct pnfs_layout_segment
*lseg
)
508 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
509 list_del_init(&lseg
->pls_list
);
510 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
511 refcount_dec(&lo
->plh_refcount
);
512 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
))
514 if (list_empty(&lo
->plh_segs
) &&
515 !test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) &&
516 !test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
517 if (atomic_read(&lo
->plh_outstanding
) == 0)
518 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
519 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
524 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr
*lo
,
525 struct pnfs_layout_segment
*lseg
)
527 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
528 pnfs_layout_is_valid(lo
)) {
529 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
530 list_move_tail(&lseg
->pls_list
, &lo
->plh_return_segs
);
537 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
539 struct pnfs_layout_hdr
*lo
;
545 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
546 refcount_read(&lseg
->pls_refcount
),
547 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
549 lo
= lseg
->pls_layout
;
550 inode
= lo
->plh_inode
;
552 if (refcount_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
553 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
554 spin_unlock(&inode
->i_lock
);
557 pnfs_get_layout_hdr(lo
);
558 pnfs_layout_remove_lseg(lo
, lseg
);
559 if (pnfs_cache_lseg_for_layoutreturn(lo
, lseg
))
561 spin_unlock(&inode
->i_lock
);
562 pnfs_free_lseg(lseg
);
563 pnfs_put_layout_hdr(lo
);
566 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
569 * is l2 fully contained in l1?
571 * [----------------------------------)
576 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
577 const struct pnfs_layout_range
*l2
)
579 u64 start1
= l1
->offset
;
580 u64 end1
= pnfs_end_offset(start1
, l1
->length
);
581 u64 start2
= l2
->offset
;
582 u64 end2
= pnfs_end_offset(start2
, l2
->length
);
584 return (start1
<= start2
) && (end1
>= end2
);
587 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
588 struct list_head
*tmp_list
)
590 if (!refcount_dec_and_test(&lseg
->pls_refcount
))
592 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
593 list_add(&lseg
->pls_list
, tmp_list
);
597 /* Returns 1 if lseg is removed from list, 0 otherwise */
598 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
599 struct list_head
*tmp_list
)
603 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
604 /* Remove the reference keeping the lseg in the
605 * list. It will now be removed when all
606 * outstanding io is finished.
608 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
609 refcount_read(&lseg
->pls_refcount
));
610 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
617 * Compare 2 layout stateid sequence ids, to see which is newer,
618 * taking into account wraparound issues.
620 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
622 return (s32
)(s1
- s2
) > 0;
626 pnfs_should_free_range(const struct pnfs_layout_range
*lseg_range
,
627 const struct pnfs_layout_range
*recall_range
)
629 return (recall_range
->iomode
== IOMODE_ANY
||
630 lseg_range
->iomode
== recall_range
->iomode
) &&
631 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
635 pnfs_match_lseg_recall(const struct pnfs_layout_segment
*lseg
,
636 const struct pnfs_layout_range
*recall_range
,
639 if (seq
!= 0 && pnfs_seqid_is_newer(lseg
->pls_seq
, seq
))
641 if (recall_range
== NULL
)
643 return pnfs_should_free_range(&lseg
->pls_range
, recall_range
);
647 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
648 * @lo: layout header containing the lsegs
649 * @tmp_list: list head where doomed lsegs should go
650 * @recall_range: optional recall range argument to match (may be NULL)
651 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
653 * Walk the list of lsegs in the layout header, and tear down any that should
654 * be destroyed. If "recall_range" is specified then the segment must match
655 * that range. If "seq" is non-zero, then only match segments that were handed
656 * out at or before that sequence.
658 * Returns number of matching invalid lsegs remaining in list after scanning
659 * it and purging them.
662 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
663 struct list_head
*tmp_list
,
664 const struct pnfs_layout_range
*recall_range
,
667 struct pnfs_layout_segment
*lseg
, *next
;
670 dprintk("%s:Begin lo %p\n", __func__
, lo
);
672 if (list_empty(&lo
->plh_segs
))
674 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
675 if (pnfs_match_lseg_recall(lseg
, recall_range
, seq
)) {
676 dprintk("%s: freeing lseg %p iomode %d seq %u "
677 "offset %llu length %llu\n", __func__
,
678 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_seq
,
679 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
680 if (!mark_lseg_invalid(lseg
, tmp_list
))
683 dprintk("%s:Return %i\n", __func__
, remaining
);
688 pnfs_free_returned_lsegs(struct pnfs_layout_hdr
*lo
,
689 struct list_head
*free_me
,
690 const struct pnfs_layout_range
*range
,
693 struct pnfs_layout_segment
*lseg
, *next
;
695 list_for_each_entry_safe(lseg
, next
, &lo
->plh_return_segs
, pls_list
) {
696 if (pnfs_match_lseg_recall(lseg
, range
, seq
))
697 list_move_tail(&lseg
->pls_list
, free_me
);
701 /* note free_me must contain lsegs from a single layout_hdr */
703 pnfs_free_lseg_list(struct list_head
*free_me
)
705 struct pnfs_layout_segment
*lseg
, *tmp
;
707 if (list_empty(free_me
))
710 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
711 list_del(&lseg
->pls_list
);
712 pnfs_free_lseg(lseg
);
717 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
719 struct pnfs_layout_hdr
*lo
;
722 spin_lock(&nfsi
->vfs_inode
.i_lock
);
725 pnfs_get_layout_hdr(lo
);
726 pnfs_mark_layout_stateid_invalid(lo
, &tmp_list
);
727 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
728 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
729 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
730 pnfs_free_lseg_list(&tmp_list
);
731 nfs_commit_inode(&nfsi
->vfs_inode
, 0);
732 pnfs_put_layout_hdr(lo
);
734 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
736 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
739 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
740 struct list_head
*layout_list
)
742 struct pnfs_layout_hdr
*lo
;
745 spin_lock(&inode
->i_lock
);
746 lo
= NFS_I(inode
)->layout
;
747 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
748 pnfs_get_layout_hdr(lo
);
749 list_add(&lo
->plh_bulk_destroy
, layout_list
);
752 spin_unlock(&inode
->i_lock
);
756 /* Caller must hold rcu_read_lock and clp->cl_lock */
758 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
759 struct nfs_server
*server
,
760 struct list_head
*layout_list
)
761 __must_hold(&clp
->cl_lock
)
764 struct pnfs_layout_hdr
*lo
, *next
;
767 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
768 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
) ||
769 test_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
) ||
770 !list_empty(&lo
->plh_bulk_destroy
))
772 /* If the sb is being destroyed, just bail */
773 if (!nfs_sb_active(server
->super
))
775 inode
= igrab(lo
->plh_inode
);
777 list_del_init(&lo
->plh_layouts
);
778 if (pnfs_layout_add_bulk_destroy_list(inode
,
782 spin_unlock(&clp
->cl_lock
);
786 spin_unlock(&clp
->cl_lock
);
787 set_bit(NFS_LAYOUT_INODE_FREEING
, &lo
->plh_flags
);
789 nfs_sb_deactive(server
->super
);
790 spin_lock(&clp
->cl_lock
);
798 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
801 struct pnfs_layout_hdr
*lo
;
803 LIST_HEAD(lseg_list
);
806 while (!list_empty(layout_list
)) {
807 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
809 dprintk("%s freeing layout for inode %lu\n", __func__
,
810 lo
->plh_inode
->i_ino
);
811 inode
= lo
->plh_inode
;
813 pnfs_layoutcommit_inode(inode
, false);
815 spin_lock(&inode
->i_lock
);
816 list_del_init(&lo
->plh_bulk_destroy
);
817 if (pnfs_mark_layout_stateid_invalid(lo
, &lseg_list
)) {
819 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
822 spin_unlock(&inode
->i_lock
);
823 pnfs_free_lseg_list(&lseg_list
);
824 /* Free all lsegs that are attached to commit buckets */
825 nfs_commit_inode(inode
, 0);
826 pnfs_put_layout_hdr(lo
);
827 nfs_iput_and_deactive(inode
);
833 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
834 struct nfs_fsid
*fsid
,
837 struct nfs_server
*server
;
838 LIST_HEAD(layout_list
);
840 spin_lock(&clp
->cl_lock
);
843 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
844 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
846 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
852 spin_unlock(&clp
->cl_lock
);
854 if (list_empty(&layout_list
))
856 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
860 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
863 struct nfs_server
*server
;
864 LIST_HEAD(layout_list
);
866 spin_lock(&clp
->cl_lock
);
869 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
870 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
876 spin_unlock(&clp
->cl_lock
);
878 if (list_empty(&layout_list
))
880 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
884 * Called by the state manger to remove all layouts established under an
888 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
890 nfs4_deviceid_mark_client_invalid(clp
);
891 nfs4_deviceid_purge_client(clp
);
893 pnfs_destroy_layouts_byclid(clp
, false);
896 /* update lo->plh_stateid with new if is more recent */
898 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
901 u32 oldseq
, newseq
, new_barrier
= 0;
903 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
904 newseq
= be32_to_cpu(new->seqid
);
906 if (!pnfs_layout_is_valid(lo
)) {
907 nfs4_stateid_copy(&lo
->plh_stateid
, new);
908 lo
->plh_barrier
= newseq
;
909 pnfs_clear_layoutreturn_info(lo
);
910 clear_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
913 if (pnfs_seqid_is_newer(newseq
, oldseq
)) {
914 nfs4_stateid_copy(&lo
->plh_stateid
, new);
916 * Because of wraparound, we want to keep the barrier
917 * "close" to the current seqids.
919 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
922 new_barrier
= be32_to_cpu(new->seqid
);
923 else if (new_barrier
== 0)
925 if (pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
926 lo
->plh_barrier
= new_barrier
;
930 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
931 const nfs4_stateid
*stateid
)
933 u32 seqid
= be32_to_cpu(stateid
->seqid
);
935 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
938 /* lget is set to 1 if called from inside send_layoutget call chain */
940 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
)
942 return lo
->plh_block_lgets
||
943 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
946 static struct nfs_server
*
947 pnfs_find_server(struct inode
*inode
, struct nfs_open_context
*ctx
)
949 struct nfs_server
*server
;
952 server
= NFS_SERVER(inode
);
954 struct dentry
*parent_dir
= dget_parent(ctx
->dentry
);
955 server
= NFS_SERVER(parent_dir
->d_inode
);
961 static void nfs4_free_pages(struct page
**pages
, size_t size
)
968 for (i
= 0; i
< size
; i
++) {
971 __free_page(pages
[i
]);
976 static struct page
**nfs4_alloc_pages(size_t size
, gfp_t gfp_flags
)
981 pages
= kmalloc_array(size
, sizeof(struct page
*), gfp_flags
);
983 dprintk("%s: can't alloc array of %zu pages\n", __func__
, size
);
987 for (i
= 0; i
< size
; i
++) {
988 pages
[i
] = alloc_page(gfp_flags
);
990 dprintk("%s: failed to allocate page\n", __func__
);
991 nfs4_free_pages(pages
, i
);
999 static struct nfs4_layoutget
*
1000 pnfs_alloc_init_layoutget_args(struct inode
*ino
,
1001 struct nfs_open_context
*ctx
,
1002 const nfs4_stateid
*stateid
,
1003 const struct pnfs_layout_range
*range
,
1006 struct nfs_server
*server
= pnfs_find_server(ino
, ctx
);
1007 size_t max_reply_sz
= server
->pnfs_curr_ld
->max_layoutget_response
;
1008 size_t max_pages
= max_response_pages(server
);
1009 struct nfs4_layoutget
*lgp
;
1011 dprintk("--> %s\n", __func__
);
1013 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
1018 size_t npages
= (max_reply_sz
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1019 if (npages
< max_pages
)
1023 lgp
->args
.layout
.pages
= nfs4_alloc_pages(max_pages
, gfp_flags
);
1024 if (!lgp
->args
.layout
.pages
) {
1028 lgp
->args
.layout
.pglen
= max_pages
* PAGE_SIZE
;
1029 lgp
->res
.layoutp
= &lgp
->args
.layout
;
1031 /* Don't confuse uninitialised result and success */
1032 lgp
->res
.status
= -NFS4ERR_DELAY
;
1034 lgp
->args
.minlength
= PAGE_SIZE
;
1035 if (lgp
->args
.minlength
> range
->length
)
1036 lgp
->args
.minlength
= range
->length
;
1038 loff_t i_size
= i_size_read(ino
);
1040 if (range
->iomode
== IOMODE_READ
) {
1041 if (range
->offset
>= i_size
)
1042 lgp
->args
.minlength
= 0;
1043 else if (i_size
- range
->offset
< lgp
->args
.minlength
)
1044 lgp
->args
.minlength
= i_size
- range
->offset
;
1047 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
1048 pnfs_copy_range(&lgp
->args
.range
, range
);
1049 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
1050 lgp
->args
.inode
= ino
;
1051 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
1052 nfs4_stateid_copy(&lgp
->args
.stateid
, stateid
);
1053 lgp
->gfp_flags
= gfp_flags
;
1054 lgp
->cred
= get_cred(ctx
->cred
);
1058 void pnfs_layoutget_free(struct nfs4_layoutget
*lgp
)
1060 size_t max_pages
= lgp
->args
.layout
.pglen
/ PAGE_SIZE
;
1062 nfs4_free_pages(lgp
->args
.layout
.pages
, max_pages
);
1063 if (lgp
->args
.inode
)
1064 pnfs_put_layout_hdr(NFS_I(lgp
->args
.inode
)->layout
);
1065 put_cred(lgp
->cred
);
1066 put_nfs_open_context(lgp
->args
.ctx
);
1070 static void pnfs_clear_layoutcommit(struct inode
*inode
,
1071 struct list_head
*head
)
1073 struct nfs_inode
*nfsi
= NFS_I(inode
);
1074 struct pnfs_layout_segment
*lseg
, *tmp
;
1076 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
1078 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
1079 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
1081 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
1085 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr
*lo
,
1086 const nfs4_stateid
*arg_stateid
,
1087 const struct pnfs_layout_range
*range
,
1088 const nfs4_stateid
*stateid
)
1090 struct inode
*inode
= lo
->plh_inode
;
1093 spin_lock(&inode
->i_lock
);
1094 if (!pnfs_layout_is_valid(lo
) || !arg_stateid
||
1095 !nfs4_stateid_match_other(&lo
->plh_stateid
, arg_stateid
))
1098 u32 seq
= be32_to_cpu(arg_stateid
->seqid
);
1100 pnfs_mark_matching_lsegs_invalid(lo
, &freeme
, range
, seq
);
1101 pnfs_free_returned_lsegs(lo
, &freeme
, range
, seq
);
1102 pnfs_set_layout_stateid(lo
, stateid
, true);
1104 pnfs_mark_layout_stateid_invalid(lo
, &freeme
);
1106 pnfs_clear_layoutreturn_waitbit(lo
);
1107 spin_unlock(&inode
->i_lock
);
1108 pnfs_free_lseg_list(&freeme
);
1113 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr
*lo
,
1114 nfs4_stateid
*stateid
,
1115 enum pnfs_iomode
*iomode
)
1117 /* Serialise LAYOUTGET/LAYOUTRETURN */
1118 if (atomic_read(&lo
->plh_outstanding
) != 0)
1120 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
))
1122 set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
1123 pnfs_get_layout_hdr(lo
);
1124 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
)) {
1125 if (stateid
!= NULL
) {
1126 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
1127 if (lo
->plh_return_seq
!= 0)
1128 stateid
->seqid
= cpu_to_be32(lo
->plh_return_seq
);
1131 *iomode
= lo
->plh_return_iomode
;
1132 pnfs_clear_layoutreturn_info(lo
);
1135 if (stateid
!= NULL
)
1136 nfs4_stateid_copy(stateid
, &lo
->plh_stateid
);
1138 *iomode
= IOMODE_ANY
;
1143 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args
*args
,
1144 struct pnfs_layout_hdr
*lo
,
1145 const nfs4_stateid
*stateid
,
1146 enum pnfs_iomode iomode
)
1148 struct inode
*inode
= lo
->plh_inode
;
1150 args
->layout_type
= NFS_SERVER(inode
)->pnfs_curr_ld
->id
;
1151 args
->inode
= inode
;
1152 args
->range
.iomode
= iomode
;
1153 args
->range
.offset
= 0;
1154 args
->range
.length
= NFS4_MAX_UINT64
;
1156 nfs4_stateid_copy(&args
->stateid
, stateid
);
1160 pnfs_send_layoutreturn(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*stateid
,
1161 enum pnfs_iomode iomode
, bool sync
)
1163 struct inode
*ino
= lo
->plh_inode
;
1164 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1165 struct nfs4_layoutreturn
*lrp
;
1168 lrp
= kzalloc(sizeof(*lrp
), GFP_NOFS
);
1169 if (unlikely(lrp
== NULL
)) {
1171 spin_lock(&ino
->i_lock
);
1172 pnfs_clear_layoutreturn_waitbit(lo
);
1173 spin_unlock(&ino
->i_lock
);
1174 pnfs_put_layout_hdr(lo
);
1178 pnfs_init_layoutreturn_args(&lrp
->args
, lo
, stateid
, iomode
);
1179 lrp
->args
.ld_private
= &lrp
->ld_private
;
1180 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
1181 lrp
->cred
= lo
->plh_lc_cred
;
1182 if (ld
->prepare_layoutreturn
)
1183 ld
->prepare_layoutreturn(&lrp
->args
);
1185 status
= nfs4_proc_layoutreturn(lrp
, sync
);
1187 dprintk("<-- %s status: %d\n", __func__
, status
);
1191 /* Return true if layoutreturn is needed */
1193 pnfs_layout_need_return(struct pnfs_layout_hdr
*lo
)
1195 struct pnfs_layout_segment
*s
;
1196 enum pnfs_iomode iomode
;
1199 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1202 seq
= lo
->plh_return_seq
;
1203 iomode
= lo
->plh_return_iomode
;
1205 /* Defer layoutreturn until all recalled lsegs are done */
1206 list_for_each_entry(s
, &lo
->plh_segs
, pls_list
) {
1207 if (seq
&& pnfs_seqid_is_newer(s
->pls_seq
, seq
))
1209 if (iomode
!= IOMODE_ANY
&& s
->pls_range
.iomode
!= iomode
)
1211 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &s
->pls_flags
))
1218 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
1220 struct inode
*inode
= lo
->plh_inode
;
1222 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1224 spin_lock(&inode
->i_lock
);
1225 if (pnfs_layout_need_return(lo
)) {
1226 nfs4_stateid stateid
;
1227 enum pnfs_iomode iomode
;
1230 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
1231 spin_unlock(&inode
->i_lock
);
1233 /* Send an async layoutreturn so we dont deadlock */
1234 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
1237 spin_unlock(&inode
->i_lock
);
1241 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1242 * when the layout segment list is empty.
1244 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1245 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1246 * deviceid is marked invalid.
1249 _pnfs_return_layout(struct inode
*ino
)
1251 struct pnfs_layout_hdr
*lo
= NULL
;
1252 struct nfs_inode
*nfsi
= NFS_I(ino
);
1253 LIST_HEAD(tmp_list
);
1254 nfs4_stateid stateid
;
1256 bool send
, valid_layout
;
1258 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
1260 spin_lock(&ino
->i_lock
);
1263 spin_unlock(&ino
->i_lock
);
1264 dprintk("NFS: %s no layout to return\n", __func__
);
1267 /* Reference matched in nfs4_layoutreturn_release */
1268 pnfs_get_layout_hdr(lo
);
1269 /* Is there an outstanding layoutreturn ? */
1270 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1271 spin_unlock(&ino
->i_lock
);
1272 if (wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1273 TASK_UNINTERRUPTIBLE
))
1274 goto out_put_layout_hdr
;
1275 spin_lock(&ino
->i_lock
);
1277 valid_layout
= pnfs_layout_is_valid(lo
);
1278 pnfs_clear_layoutcommit(ino
, &tmp_list
);
1279 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
, 0);
1281 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
1282 struct pnfs_layout_range range
= {
1283 .iomode
= IOMODE_ANY
,
1285 .length
= NFS4_MAX_UINT64
,
1287 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
, &range
);
1290 /* Don't send a LAYOUTRETURN if list was initially empty */
1291 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
) ||
1293 spin_unlock(&ino
->i_lock
);
1294 dprintk("NFS: %s no layout segments to return\n", __func__
);
1295 goto out_put_layout_hdr
;
1298 send
= pnfs_prepare_layoutreturn(lo
, &stateid
, NULL
);
1299 spin_unlock(&ino
->i_lock
);
1301 status
= pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1303 pnfs_free_lseg_list(&tmp_list
);
1304 pnfs_put_layout_hdr(lo
);
1306 dprintk("<-- %s status: %d\n", __func__
, status
);
1311 pnfs_commit_and_return_layout(struct inode
*inode
)
1313 struct pnfs_layout_hdr
*lo
;
1316 spin_lock(&inode
->i_lock
);
1317 lo
= NFS_I(inode
)->layout
;
1319 spin_unlock(&inode
->i_lock
);
1322 pnfs_get_layout_hdr(lo
);
1323 /* Block new layoutgets and read/write to ds */
1324 lo
->plh_block_lgets
++;
1325 spin_unlock(&inode
->i_lock
);
1326 filemap_fdatawait(inode
->i_mapping
);
1327 ret
= pnfs_layoutcommit_inode(inode
, true);
1329 ret
= _pnfs_return_layout(inode
);
1330 spin_lock(&inode
->i_lock
);
1331 lo
->plh_block_lgets
--;
1332 spin_unlock(&inode
->i_lock
);
1333 pnfs_put_layout_hdr(lo
);
1337 bool pnfs_roc(struct inode
*ino
,
1338 struct nfs4_layoutreturn_args
*args
,
1339 struct nfs4_layoutreturn_res
*res
,
1340 const struct cred
*cred
)
1342 struct nfs_inode
*nfsi
= NFS_I(ino
);
1343 struct nfs_open_context
*ctx
;
1344 struct nfs4_state
*state
;
1345 struct pnfs_layout_hdr
*lo
;
1346 struct pnfs_layout_segment
*lseg
, *next
;
1347 nfs4_stateid stateid
;
1348 enum pnfs_iomode iomode
= 0;
1349 bool layoutreturn
= false, roc
= false;
1350 bool skip_read
= false;
1352 if (!nfs_have_layout(ino
))
1356 spin_lock(&ino
->i_lock
);
1358 if (!lo
|| !pnfs_layout_is_valid(lo
) ||
1359 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1363 pnfs_get_layout_hdr(lo
);
1364 if (test_bit(NFS_LAYOUT_RETURN_LOCK
, &lo
->plh_flags
)) {
1365 spin_unlock(&ino
->i_lock
);
1367 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1368 TASK_UNINTERRUPTIBLE
);
1369 pnfs_put_layout_hdr(lo
);
1373 /* no roc if we hold a delegation */
1374 if (nfs4_check_delegation(ino
, FMODE_READ
)) {
1375 if (nfs4_check_delegation(ino
, FMODE_WRITE
))
1380 list_for_each_entry_rcu(ctx
, &nfsi
->open_files
, list
) {
1384 /* Don't return layout if there is open file state */
1385 if (state
->state
& FMODE_WRITE
)
1387 if (state
->state
& FMODE_READ
)
1392 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
) {
1393 if (skip_read
&& lseg
->pls_range
.iomode
== IOMODE_READ
)
1395 /* If we are sending layoutreturn, invalidate all valid lsegs */
1396 if (!test_and_clear_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
))
1399 * Note: mark lseg for return so pnfs_layout_remove_lseg
1400 * doesn't invalidate the layout for us.
1402 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
1403 if (!mark_lseg_invalid(lseg
, &lo
->plh_return_segs
))
1405 pnfs_set_plh_return_info(lo
, lseg
->pls_range
.iomode
, 0);
1408 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
1411 /* ROC in two conditions:
1412 * 1. there are ROC lsegs
1413 * 2. we don't send layoutreturn
1415 /* lo ref dropped in pnfs_roc_release() */
1416 layoutreturn
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
1417 /* If the creds don't match, we can't compound the layoutreturn */
1418 if (!layoutreturn
|| cred
!= lo
->plh_lc_cred
)
1422 pnfs_init_layoutreturn_args(args
, lo
, &stateid
, iomode
);
1423 res
->lrs_present
= 0;
1424 layoutreturn
= false;
1427 spin_unlock(&ino
->i_lock
);
1429 pnfs_layoutcommit_inode(ino
, true);
1431 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(ino
)->pnfs_curr_ld
;
1432 if (ld
->prepare_layoutreturn
)
1433 ld
->prepare_layoutreturn(args
);
1434 pnfs_put_layout_hdr(lo
);
1438 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, true);
1439 pnfs_put_layout_hdr(lo
);
1443 void pnfs_roc_release(struct nfs4_layoutreturn_args
*args
,
1444 struct nfs4_layoutreturn_res
*res
,
1447 struct pnfs_layout_hdr
*lo
= args
->layout
;
1448 const nfs4_stateid
*arg_stateid
= NULL
;
1449 const nfs4_stateid
*res_stateid
= NULL
;
1450 struct nfs4_xdr_opaque_data
*ld_private
= args
->ld_private
;
1453 arg_stateid
= &args
->stateid
;
1454 if (res
->lrs_present
)
1455 res_stateid
= &res
->stateid
;
1457 pnfs_layoutreturn_free_lsegs(lo
, arg_stateid
, &args
->range
,
1459 if (ld_private
&& ld_private
->ops
&& ld_private
->ops
->free
)
1460 ld_private
->ops
->free(ld_private
);
1461 pnfs_put_layout_hdr(lo
);
1462 trace_nfs4_layoutreturn_on_close(args
->inode
, 0);
1465 bool pnfs_wait_on_layoutreturn(struct inode
*ino
, struct rpc_task
*task
)
1467 struct nfs_inode
*nfsi
= NFS_I(ino
);
1468 struct pnfs_layout_hdr
*lo
;
1471 /* we might not have grabbed lo reference. so need to check under
1473 spin_lock(&ino
->i_lock
);
1475 if (lo
&& test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1476 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1479 spin_unlock(&ino
->i_lock
);
1484 * Compare two layout segments for sorting into layout cache.
1485 * We want to preferentially return RW over RO layouts, so ensure those
1489 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1490 const struct pnfs_layout_range
*l2
)
1494 /* high offset > low offset */
1495 d
= l1
->offset
- l2
->offset
;
1499 /* short length > long length */
1500 d
= l2
->length
- l1
->length
;
1504 /* read > read/write */
1505 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1509 pnfs_lseg_range_is_after(const struct pnfs_layout_range
*l1
,
1510 const struct pnfs_layout_range
*l2
)
1512 return pnfs_lseg_range_cmp(l1
, l2
) > 0;
1516 pnfs_lseg_no_merge(struct pnfs_layout_segment
*lseg
,
1517 struct pnfs_layout_segment
*old
)
1523 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1524 struct pnfs_layout_segment
*lseg
,
1525 bool (*is_after
)(const struct pnfs_layout_range
*,
1526 const struct pnfs_layout_range
*),
1527 bool (*do_merge
)(struct pnfs_layout_segment
*,
1528 struct pnfs_layout_segment
*),
1529 struct list_head
*free_me
)
1531 struct pnfs_layout_segment
*lp
, *tmp
;
1533 dprintk("%s:Begin\n", __func__
);
1535 list_for_each_entry_safe(lp
, tmp
, &lo
->plh_segs
, pls_list
) {
1536 if (test_bit(NFS_LSEG_VALID
, &lp
->pls_flags
) == 0)
1538 if (do_merge(lseg
, lp
)) {
1539 mark_lseg_invalid(lp
, free_me
);
1542 if (is_after(&lseg
->pls_range
, &lp
->pls_range
))
1544 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1545 dprintk("%s: inserted lseg %p "
1546 "iomode %d offset %llu length %llu before "
1547 "lp %p iomode %d offset %llu length %llu\n",
1548 __func__
, lseg
, lseg
->pls_range
.iomode
,
1549 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1550 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1551 lp
->pls_range
.length
);
1554 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1555 dprintk("%s: inserted lseg %p "
1556 "iomode %d offset %llu length %llu at tail\n",
1557 __func__
, lseg
, lseg
->pls_range
.iomode
,
1558 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1560 pnfs_get_layout_hdr(lo
);
1562 dprintk("%s:Return\n", __func__
);
1564 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg
);
1567 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1568 struct pnfs_layout_segment
*lseg
,
1569 struct list_head
*free_me
)
1571 struct inode
*inode
= lo
->plh_inode
;
1572 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
1574 if (ld
->add_lseg
!= NULL
)
1575 ld
->add_lseg(lo
, lseg
, free_me
);
1577 pnfs_generic_layout_insert_lseg(lo
, lseg
,
1578 pnfs_lseg_range_is_after
,
1583 static struct pnfs_layout_hdr
*
1584 alloc_init_layout_hdr(struct inode
*ino
,
1585 struct nfs_open_context
*ctx
,
1588 struct pnfs_layout_hdr
*lo
;
1590 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1593 refcount_set(&lo
->plh_refcount
, 1);
1594 INIT_LIST_HEAD(&lo
->plh_layouts
);
1595 INIT_LIST_HEAD(&lo
->plh_segs
);
1596 INIT_LIST_HEAD(&lo
->plh_return_segs
);
1597 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1598 lo
->plh_inode
= ino
;
1599 lo
->plh_lc_cred
= get_cred(ctx
->cred
);
1600 lo
->plh_flags
|= 1 << NFS_LAYOUT_INVALID_STID
;
1604 static struct pnfs_layout_hdr
*
1605 pnfs_find_alloc_layout(struct inode
*ino
,
1606 struct nfs_open_context
*ctx
,
1608 __releases(&ino
->i_lock
)
1609 __acquires(&ino
->i_lock
)
1611 struct nfs_inode
*nfsi
= NFS_I(ino
);
1612 struct pnfs_layout_hdr
*new = NULL
;
1614 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1616 if (nfsi
->layout
!= NULL
)
1618 spin_unlock(&ino
->i_lock
);
1619 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1620 spin_lock(&ino
->i_lock
);
1622 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1625 } else if (new != NULL
)
1626 pnfs_free_layout_hdr(new);
1628 pnfs_get_layout_hdr(nfsi
->layout
);
1629 return nfsi
->layout
;
1633 * iomode matching rules:
1634 * iomode lseg strict match
1636 * ----- ----- ------ -----
1641 * READ READ N/A true
1642 * READ RW true false
1643 * READ RW false true
1646 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1647 const struct pnfs_layout_range
*range
,
1650 struct pnfs_layout_range range1
;
1652 if ((range
->iomode
== IOMODE_RW
&&
1653 ls_range
->iomode
!= IOMODE_RW
) ||
1654 (range
->iomode
!= ls_range
->iomode
&&
1656 !pnfs_lseg_range_intersecting(ls_range
, range
))
1659 /* range1 covers only the first byte in the range */
1662 return pnfs_lseg_range_contained(ls_range
, &range1
);
1666 * lookup range in layout
1668 static struct pnfs_layout_segment
*
1669 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1670 struct pnfs_layout_range
*range
,
1673 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1675 dprintk("%s:Begin\n", __func__
);
1677 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1678 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1679 pnfs_lseg_range_match(&lseg
->pls_range
, range
,
1681 ret
= pnfs_get_lseg(lseg
);
1686 dprintk("%s:Return lseg %p ref %d\n",
1687 __func__
, ret
, ret
? refcount_read(&ret
->pls_refcount
) : 0);
1692 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1693 * to the MDS or over pNFS
1695 * The nfs_inode read_io and write_io fields are cumulative counters reset
1696 * when there are no layout segments. Note that in pnfs_update_layout iomode
1697 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1700 * A return of true means use MDS I/O.
1703 * If a file's size is smaller than the file size threshold, data accesses
1704 * SHOULD be sent to the metadata server. If an I/O request has a length that
1705 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1706 * server. If both file size and I/O size are provided, the client SHOULD
1707 * reach or exceed both thresholds before sending its read or write
1708 * requests to the data server.
1710 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1711 struct inode
*ino
, int iomode
)
1713 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1714 struct nfs_inode
*nfsi
= NFS_I(ino
);
1715 loff_t fsize
= i_size_read(ino
);
1716 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1721 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1722 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1726 if (t
->bm
& THRESHOLD_RD
) {
1727 dprintk("%s fsize %llu\n", __func__
, fsize
);
1729 if (fsize
< t
->rd_sz
)
1732 if (t
->bm
& THRESHOLD_RD_IO
) {
1733 dprintk("%s nfsi->read_io %llu\n", __func__
,
1736 if (nfsi
->read_io
< t
->rd_io_sz
)
1741 if (t
->bm
& THRESHOLD_WR
) {
1742 dprintk("%s fsize %llu\n", __func__
, fsize
);
1744 if (fsize
< t
->wr_sz
)
1747 if (t
->bm
& THRESHOLD_WR_IO
) {
1748 dprintk("%s nfsi->write_io %llu\n", __func__
,
1751 if (nfsi
->write_io
< t
->wr_io_sz
)
1756 if (size_set
&& io_set
) {
1759 } else if (size
|| io
)
1762 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1766 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr
*lo
)
1769 * send layoutcommit as it can hold up layoutreturn due to lseg
1772 pnfs_layoutcommit_inode(lo
->plh_inode
, false);
1773 return wait_on_bit_action(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1774 nfs_wait_bit_killable
,
1778 static void nfs_layoutget_begin(struct pnfs_layout_hdr
*lo
)
1780 atomic_inc(&lo
->plh_outstanding
);
1783 static void nfs_layoutget_end(struct pnfs_layout_hdr
*lo
)
1785 if (atomic_dec_and_test(&lo
->plh_outstanding
))
1786 wake_up_var(&lo
->plh_outstanding
);
1789 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr
*lo
)
1791 unsigned long *bitlock
= &lo
->plh_flags
;
1793 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET
, bitlock
);
1794 smp_mb__after_atomic();
1795 wake_up_bit(bitlock
, NFS_LAYOUT_FIRST_LAYOUTGET
);
1798 static void _add_to_server_list(struct pnfs_layout_hdr
*lo
,
1799 struct nfs_server
*server
)
1801 if (list_empty(&lo
->plh_layouts
)) {
1802 struct nfs_client
*clp
= server
->nfs_client
;
1804 /* The lo must be on the clp list if there is any
1805 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1807 spin_lock(&clp
->cl_lock
);
1808 if (list_empty(&lo
->plh_layouts
))
1809 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1810 spin_unlock(&clp
->cl_lock
);
1815 * Layout segment is retreived from the server if not cached.
1816 * The appropriate layout segment is referenced and returned to the caller.
1818 struct pnfs_layout_segment
*
1819 pnfs_update_layout(struct inode
*ino
,
1820 struct nfs_open_context
*ctx
,
1823 enum pnfs_iomode iomode
,
1827 struct pnfs_layout_range arg
= {
1833 struct nfs_server
*server
= NFS_SERVER(ino
);
1834 struct nfs_client
*clp
= server
->nfs_client
;
1835 struct pnfs_layout_hdr
*lo
= NULL
;
1836 struct pnfs_layout_segment
*lseg
= NULL
;
1837 struct nfs4_layoutget
*lgp
;
1838 nfs4_stateid stateid
;
1840 unsigned long giveup
= jiffies
+ (clp
->cl_lease_time
<< 1);
1843 if (!pnfs_enabled_sb(NFS_SERVER(ino
))) {
1844 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1845 PNFS_UPDATE_LAYOUT_NO_PNFS
);
1849 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
)) {
1850 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1851 PNFS_UPDATE_LAYOUT_MDSTHRESH
);
1856 lseg
= ERR_PTR(nfs4_client_recover_expired_lease(clp
));
1860 spin_lock(&ino
->i_lock
);
1861 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1863 spin_unlock(&ino
->i_lock
);
1864 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1865 PNFS_UPDATE_LAYOUT_NOMEM
);
1869 /* Do we even need to bother with this? */
1870 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1871 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1872 PNFS_UPDATE_LAYOUT_BULK_RECALL
);
1873 dprintk("%s matches recall, use MDS\n", __func__
);
1877 /* if LAYOUTGET already failed once we don't try again */
1878 if (pnfs_layout_io_test_failed(lo
, iomode
)) {
1879 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1880 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL
);
1885 * If the layout segment list is empty, but there are outstanding
1886 * layoutget calls, then they might be subject to a layoutrecall.
1888 if (list_empty(&lo
->plh_segs
) &&
1889 atomic_read(&lo
->plh_outstanding
) != 0) {
1890 spin_unlock(&ino
->i_lock
);
1891 lseg
= ERR_PTR(wait_var_event_killable(&lo
->plh_outstanding
,
1892 !atomic_read(&lo
->plh_outstanding
)));
1893 if (IS_ERR(lseg
) || !list_empty(&lo
->plh_segs
))
1894 goto out_put_layout_hdr
;
1895 pnfs_put_layout_hdr(lo
);
1899 lseg
= pnfs_find_lseg(lo
, &arg
, strict_iomode
);
1901 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1902 PNFS_UPDATE_LAYOUT_FOUND_CACHED
);
1906 if (!nfs4_valid_open_stateid(ctx
->state
)) {
1907 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1908 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1913 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1914 * stateid, or it has been invalidated, then we must use the open
1917 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
)) {
1921 * The first layoutget for the file. Need to serialize per
1922 * RFC 5661 Errata 3208.
1924 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
,
1926 spin_unlock(&ino
->i_lock
);
1927 lseg
= ERR_PTR(wait_on_bit(&lo
->plh_flags
,
1928 NFS_LAYOUT_FIRST_LAYOUTGET
,
1931 goto out_put_layout_hdr
;
1932 pnfs_put_layout_hdr(lo
);
1933 dprintk("%s retrying\n", __func__
);
1938 status
= nfs4_select_rw_stateid(ctx
->state
,
1939 iomode
== IOMODE_RW
? FMODE_WRITE
: FMODE_READ
,
1940 NULL
, &stateid
, NULL
);
1942 trace_pnfs_update_layout(ino
, pos
, count
,
1944 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1945 if (status
!= -EAGAIN
)
1947 spin_unlock(&ino
->i_lock
);
1948 nfs4_schedule_stateid_recovery(server
, ctx
->state
);
1949 pnfs_clear_first_layoutget(lo
);
1950 pnfs_put_layout_hdr(lo
);
1954 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1958 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1959 * for LAYOUTRETURN even if first is true.
1961 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1962 spin_unlock(&ino
->i_lock
);
1963 dprintk("%s wait for layoutreturn\n", __func__
);
1964 lseg
= ERR_PTR(pnfs_prepare_to_retry_layoutget(lo
));
1965 if (!IS_ERR(lseg
)) {
1967 pnfs_clear_first_layoutget(lo
);
1968 pnfs_put_layout_hdr(lo
);
1969 dprintk("%s retrying\n", __func__
);
1970 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
,
1971 lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1974 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1975 PNFS_UPDATE_LAYOUT_RETURN
);
1976 goto out_put_layout_hdr
;
1979 if (pnfs_layoutgets_blocked(lo
)) {
1980 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1981 PNFS_UPDATE_LAYOUT_BLOCKED
);
1984 nfs_layoutget_begin(lo
);
1985 spin_unlock(&ino
->i_lock
);
1987 _add_to_server_list(lo
, server
);
1989 pg_offset
= arg
.offset
& ~PAGE_MASK
;
1991 arg
.offset
-= pg_offset
;
1992 arg
.length
+= pg_offset
;
1994 if (arg
.length
!= NFS4_MAX_UINT64
)
1995 arg
.length
= PAGE_ALIGN(arg
.length
);
1997 lgp
= pnfs_alloc_init_layoutget_args(ino
, ctx
, &stateid
, &arg
, gfp_flags
);
1999 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, NULL
,
2000 PNFS_UPDATE_LAYOUT_NOMEM
);
2001 nfs_layoutget_end(lo
);
2002 goto out_put_layout_hdr
;
2005 lseg
= nfs4_proc_layoutget(lgp
, &timeout
);
2006 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
2007 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET
);
2008 nfs_layoutget_end(lo
);
2010 switch(PTR_ERR(lseg
)) {
2012 if (time_after(jiffies
, giveup
))
2015 case -ERECALLCONFLICT
:
2019 if (!nfs_error_is_fatal(PTR_ERR(lseg
))) {
2020 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2023 goto out_put_layout_hdr
;
2027 pnfs_clear_first_layoutget(lo
);
2028 trace_pnfs_update_layout(ino
, pos
, count
,
2029 iomode
, lo
, lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
2030 pnfs_put_layout_hdr(lo
);
2034 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2039 pnfs_clear_first_layoutget(lo
);
2040 pnfs_put_layout_hdr(lo
);
2042 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2043 "(%s, offset: %llu, length: %llu)\n",
2044 __func__
, ino
->i_sb
->s_id
,
2045 (unsigned long long)NFS_FILEID(ino
),
2046 IS_ERR_OR_NULL(lseg
) ? "not found" : "found",
2047 iomode
==IOMODE_RW
? "read/write" : "read-only",
2048 (unsigned long long)pos
,
2049 (unsigned long long)count
);
2052 spin_unlock(&ino
->i_lock
);
2053 goto out_put_layout_hdr
;
2055 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
2058 pnfs_sanity_check_layout_range(struct pnfs_layout_range
*range
)
2060 switch (range
->iomode
) {
2067 if (range
->offset
== NFS4_MAX_UINT64
)
2069 if (range
->length
== 0)
2071 if (range
->length
!= NFS4_MAX_UINT64
&&
2072 range
->length
> NFS4_MAX_UINT64
- range
->offset
)
2077 static struct pnfs_layout_hdr
*
2078 _pnfs_grab_empty_layout(struct inode
*ino
, struct nfs_open_context
*ctx
)
2080 struct pnfs_layout_hdr
*lo
;
2082 spin_lock(&ino
->i_lock
);
2083 lo
= pnfs_find_alloc_layout(ino
, ctx
, GFP_KERNEL
);
2086 if (!test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
))
2088 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
2090 if (pnfs_layoutgets_blocked(lo
))
2092 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
, &lo
->plh_flags
))
2094 nfs_layoutget_begin(lo
);
2095 spin_unlock(&ino
->i_lock
);
2096 _add_to_server_list(lo
, NFS_SERVER(ino
));
2100 spin_unlock(&ino
->i_lock
);
2101 pnfs_put_layout_hdr(lo
);
2105 extern const nfs4_stateid current_stateid
;
2107 static void _lgopen_prepare_attached(struct nfs4_opendata
*data
,
2108 struct nfs_open_context
*ctx
)
2110 struct inode
*ino
= data
->dentry
->d_inode
;
2111 struct pnfs_layout_range rng
= {
2112 .iomode
= (data
->o_arg
.fmode
& FMODE_WRITE
) ?
2113 IOMODE_RW
: IOMODE_READ
,
2115 .length
= NFS4_MAX_UINT64
,
2117 struct nfs4_layoutget
*lgp
;
2118 struct pnfs_layout_hdr
*lo
;
2120 /* Heuristic: don't send layoutget if we have cached data */
2121 if (rng
.iomode
== IOMODE_READ
&&
2122 (i_size_read(ino
) == 0 || ino
->i_mapping
->nrpages
!= 0))
2125 lo
= _pnfs_grab_empty_layout(ino
, ctx
);
2128 lgp
= pnfs_alloc_init_layoutget_args(ino
, ctx
, ¤t_stateid
,
2131 pnfs_clear_first_layoutget(lo
);
2132 pnfs_put_layout_hdr(lo
);
2136 data
->o_arg
.lg_args
= &lgp
->args
;
2137 data
->o_res
.lg_res
= &lgp
->res
;
2140 static void _lgopen_prepare_floating(struct nfs4_opendata
*data
,
2141 struct nfs_open_context
*ctx
)
2143 struct pnfs_layout_range rng
= {
2144 .iomode
= (data
->o_arg
.fmode
& FMODE_WRITE
) ?
2145 IOMODE_RW
: IOMODE_READ
,
2147 .length
= NFS4_MAX_UINT64
,
2149 struct nfs4_layoutget
*lgp
;
2151 lgp
= pnfs_alloc_init_layoutget_args(NULL
, ctx
, ¤t_stateid
,
2156 data
->o_arg
.lg_args
= &lgp
->args
;
2157 data
->o_res
.lg_res
= &lgp
->res
;
2160 void pnfs_lgopen_prepare(struct nfs4_opendata
*data
,
2161 struct nfs_open_context
*ctx
)
2163 struct nfs_server
*server
= NFS_SERVER(data
->dir
->d_inode
);
2165 if (!(pnfs_enabled_sb(server
) &&
2166 server
->pnfs_curr_ld
->flags
& PNFS_LAYOUTGET_ON_OPEN
))
2168 /* Could check on max_ops, but currently hardcoded high enough */
2169 if (!nfs_server_capable(data
->dir
->d_inode
, NFS_CAP_LGOPEN
))
2172 _lgopen_prepare_attached(data
, ctx
);
2174 _lgopen_prepare_floating(data
, ctx
);
2177 void pnfs_parse_lgopen(struct inode
*ino
, struct nfs4_layoutget
*lgp
,
2178 struct nfs_open_context
*ctx
)
2180 struct pnfs_layout_hdr
*lo
;
2181 struct pnfs_layout_segment
*lseg
;
2182 struct nfs_server
*srv
= NFS_SERVER(ino
);
2187 dprintk("%s: entered with status %i\n", __func__
, lgp
->res
.status
);
2188 if (lgp
->res
.status
) {
2189 switch (lgp
->res
.status
) {
2193 * Halt lgopen attempts if the server doesn't recognise
2194 * the "current stateid" value, the layout type, or the
2195 * layoutget operation as being valid.
2196 * Also if it complains about too many ops in the compound
2197 * or of the request/reply being too big.
2199 case -NFS4ERR_BAD_STATEID
:
2200 case -NFS4ERR_NOTSUPP
:
2201 case -NFS4ERR_REP_TOO_BIG
:
2202 case -NFS4ERR_REP_TOO_BIG_TO_CACHE
:
2203 case -NFS4ERR_REQ_TOO_BIG
:
2204 case -NFS4ERR_TOO_MANY_OPS
:
2205 case -NFS4ERR_UNKNOWN_LAYOUTTYPE
:
2206 srv
->caps
&= ~NFS_CAP_LGOPEN
;
2210 if (!lgp
->args
.inode
) {
2211 lo
= _pnfs_grab_empty_layout(ino
, ctx
);
2214 lgp
->args
.inode
= ino
;
2216 lo
= NFS_I(lgp
->args
.inode
)->layout
;
2218 lseg
= pnfs_layout_process(lgp
);
2219 if (!IS_ERR(lseg
)) {
2220 iomode
= lgp
->args
.range
.iomode
;
2221 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
2222 pnfs_put_lseg(lseg
);
2226 void nfs4_lgopen_release(struct nfs4_layoutget
*lgp
)
2229 struct inode
*inode
= lgp
->args
.inode
;
2231 struct pnfs_layout_hdr
*lo
= NFS_I(inode
)->layout
;
2232 pnfs_clear_first_layoutget(lo
);
2233 nfs_layoutget_end(lo
);
2235 pnfs_layoutget_free(lgp
);
2239 struct pnfs_layout_segment
*
2240 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
2242 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
2243 struct nfs4_layoutget_res
*res
= &lgp
->res
;
2244 struct pnfs_layout_segment
*lseg
;
2245 struct inode
*ino
= lo
->plh_inode
;
2248 if (!pnfs_sanity_check_layout_range(&res
->range
))
2249 return ERR_PTR(-EINVAL
);
2251 /* Inject layout blob into I/O device driver */
2252 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
2253 if (IS_ERR_OR_NULL(lseg
)) {
2255 lseg
= ERR_PTR(-ENOMEM
);
2257 dprintk("%s: Could not allocate layout: error %ld\n",
2258 __func__
, PTR_ERR(lseg
));
2262 pnfs_init_lseg(lo
, lseg
, &res
->range
, &res
->stateid
);
2264 spin_lock(&ino
->i_lock
);
2265 if (pnfs_layoutgets_blocked(lo
)) {
2266 dprintk("%s forget reply due to state\n", __func__
);
2270 if (!pnfs_layout_is_valid(lo
)) {
2271 /* We have a completely new layout */
2272 pnfs_set_layout_stateid(lo
, &res
->stateid
, true);
2273 } else if (nfs4_stateid_match_other(&lo
->plh_stateid
, &res
->stateid
)) {
2274 /* existing state ID, make sure the sequence number matches. */
2275 if (pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
2276 dprintk("%s forget reply due to sequence\n", __func__
);
2279 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
2282 * We got an entirely new state ID. Mark all segments for the
2283 * inode invalid, and retry the layoutget
2285 pnfs_mark_layout_stateid_invalid(lo
, &free_me
);
2289 pnfs_get_lseg(lseg
);
2290 pnfs_layout_insert_lseg(lo
, lseg
, &free_me
);
2293 if (res
->return_on_close
)
2294 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
2296 spin_unlock(&ino
->i_lock
);
2297 pnfs_free_lseg_list(&free_me
);
2301 spin_unlock(&ino
->i_lock
);
2302 lseg
->pls_layout
= lo
;
2303 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
2304 return ERR_PTR(-EAGAIN
);
2308 mark_lseg_invalid_or_return(struct pnfs_layout_segment
*lseg
,
2309 struct list_head
*tmp_list
)
2311 if (!mark_lseg_invalid(lseg
, tmp_list
))
2313 pnfs_cache_lseg_for_layoutreturn(lseg
->pls_layout
, lseg
);
2318 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2319 * @lo: pointer to layout header
2320 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2321 * @return_range: describe layout segment ranges to be returned
2322 * @seq: stateid seqid to match
2324 * This function is mainly intended for use by layoutrecall. It attempts
2325 * to free the layout segment immediately, or else to mark it for return
2326 * as soon as its reference count drops to zero.
2329 * - 0: a layoutreturn needs to be scheduled.
2330 * - EBUSY: there are layout segment that are still in use.
2331 * - ENOENT: there are no layout segments that need to be returned.
2334 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr
*lo
,
2335 struct list_head
*tmp_list
,
2336 const struct pnfs_layout_range
*return_range
,
2339 struct pnfs_layout_segment
*lseg
, *next
;
2342 dprintk("%s:Begin lo %p\n", __func__
, lo
);
2344 assert_spin_locked(&lo
->plh_inode
->i_lock
);
2346 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
2347 if (pnfs_match_lseg_recall(lseg
, return_range
, seq
)) {
2348 dprintk("%s: marking lseg %p iomode %d "
2349 "offset %llu length %llu\n", __func__
,
2350 lseg
, lseg
->pls_range
.iomode
,
2351 lseg
->pls_range
.offset
,
2352 lseg
->pls_range
.length
);
2353 if (mark_lseg_invalid_or_return(lseg
, tmp_list
))
2356 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
2360 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
2364 if (!list_empty(&lo
->plh_return_segs
)) {
2365 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
2372 void pnfs_error_mark_layout_for_return(struct inode
*inode
,
2373 struct pnfs_layout_segment
*lseg
)
2375 struct pnfs_layout_hdr
*lo
= NFS_I(inode
)->layout
;
2376 struct pnfs_layout_range range
= {
2377 .iomode
= lseg
->pls_range
.iomode
,
2379 .length
= NFS4_MAX_UINT64
,
2381 bool return_now
= false;
2383 spin_lock(&inode
->i_lock
);
2384 if (!pnfs_layout_is_valid(lo
)) {
2385 spin_unlock(&inode
->i_lock
);
2388 pnfs_set_plh_return_info(lo
, range
.iomode
, 0);
2390 * mark all matching lsegs so that we are sure to have no live
2391 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2394 if (pnfs_mark_matching_lsegs_return(lo
, &lo
->plh_return_segs
, &range
, 0) != -EBUSY
) {
2395 nfs4_stateid stateid
;
2396 enum pnfs_iomode iomode
;
2398 return_now
= pnfs_prepare_layoutreturn(lo
, &stateid
, &iomode
);
2399 spin_unlock(&inode
->i_lock
);
2401 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
2403 spin_unlock(&inode
->i_lock
);
2404 nfs_commit_inode(inode
, 0);
2407 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return
);
2410 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor
*pgio
)
2412 if (pgio
->pg_lseg
== NULL
||
2413 test_bit(NFS_LSEG_VALID
, &pgio
->pg_lseg
->pls_flags
))
2415 pnfs_put_lseg(pgio
->pg_lseg
);
2416 pgio
->pg_lseg
= NULL
;
2418 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout
);
2421 * Check for any intersection between the request and the pgio->pg_lseg,
2422 * and if none, put this pgio->pg_lseg away.
2425 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2427 if (pgio
->pg_lseg
&& !pnfs_lseg_request_intersecting(pgio
->pg_lseg
, req
)) {
2428 pnfs_put_lseg(pgio
->pg_lseg
);
2429 pgio
->pg_lseg
= NULL
;
2434 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
2436 u64 rd_size
= req
->wb_bytes
;
2438 pnfs_generic_pg_check_layout(pgio
);
2439 pnfs_generic_pg_check_range(pgio
, req
);
2440 if (pgio
->pg_lseg
== NULL
) {
2441 if (pgio
->pg_dreq
== NULL
)
2442 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
2444 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
2446 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
2447 nfs_req_openctx(req
),
2453 if (IS_ERR(pgio
->pg_lseg
)) {
2454 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2455 pgio
->pg_lseg
= NULL
;
2459 /* If no lseg, fall back to read through mds */
2460 if (pgio
->pg_lseg
== NULL
)
2461 nfs_pageio_reset_read_mds(pgio
);
2464 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
2467 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
2468 struct nfs_page
*req
, u64 wb_size
)
2470 pnfs_generic_pg_check_layout(pgio
);
2471 pnfs_generic_pg_check_range(pgio
, req
);
2472 if (pgio
->pg_lseg
== NULL
) {
2473 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
2474 nfs_req_openctx(req
),
2480 if (IS_ERR(pgio
->pg_lseg
)) {
2481 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
2482 pgio
->pg_lseg
= NULL
;
2486 /* If no lseg, fall back to write through mds */
2487 if (pgio
->pg_lseg
== NULL
)
2488 nfs_pageio_reset_write_mds(pgio
);
2490 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
2493 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor
*desc
)
2495 if (desc
->pg_lseg
) {
2496 pnfs_put_lseg(desc
->pg_lseg
);
2497 desc
->pg_lseg
= NULL
;
2500 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup
);
2503 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2504 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2507 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
,
2508 struct nfs_page
*prev
, struct nfs_page
*req
)
2511 u64 seg_end
, req_start
, seg_left
;
2513 size
= nfs_generic_pg_test(pgio
, prev
, req
);
2518 * 'size' contains the number of bytes left in the current page (up
2519 * to the original size asked for in @req->wb_bytes).
2521 * Calculate how many bytes are left in the layout segment
2522 * and if there are less bytes than 'size', return that instead.
2524 * Please also note that 'end_offset' is actually the offset of the
2525 * first byte that lies outside the pnfs_layout_range. FIXME?
2528 if (pgio
->pg_lseg
) {
2529 seg_end
= pnfs_end_offset(pgio
->pg_lseg
->pls_range
.offset
,
2530 pgio
->pg_lseg
->pls_range
.length
);
2531 req_start
= req_offset(req
);
2533 /* start of request is past the last byte of this segment */
2534 if (req_start
>= seg_end
)
2537 /* adjust 'size' iff there are fewer bytes left in the
2538 * segment than what nfs_generic_pg_test returned */
2539 seg_left
= seg_end
- req_start
;
2540 if (seg_left
< size
)
2541 size
= (unsigned int)seg_left
;
2546 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
2548 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2550 struct nfs_pageio_descriptor pgio
;
2552 /* Resend all requests through the MDS */
2553 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
2554 hdr
->completion_ops
);
2555 set_bit(NFS_CONTEXT_RESEND_WRITES
, &hdr
->args
.context
->flags
);
2556 return nfs_pageio_resend(&pgio
, hdr
);
2558 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
2560 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
2563 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
2564 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2565 PNFS_LAYOUTRET_ON_ERROR
) {
2566 pnfs_return_layout(hdr
->inode
);
2568 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2569 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
2573 * Called by non rpc-based layout drivers
2575 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
2577 if (likely(!hdr
->pnfs_error
)) {
2578 pnfs_set_layoutcommit(hdr
->inode
, hdr
->lseg
,
2579 hdr
->mds_offset
+ hdr
->res
.count
);
2580 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2582 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
2583 if (unlikely(hdr
->pnfs_error
))
2584 pnfs_ld_handle_write_error(hdr
);
2585 hdr
->mds_ops
->rpc_release(hdr
);
2587 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
2590 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
2591 struct nfs_pgio_header
*hdr
)
2593 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2595 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2596 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2597 nfs_pageio_reset_write_mds(desc
);
2598 mirror
->pg_recoalesce
= 1;
2600 hdr
->completion_ops
->completion(hdr
);
2603 static enum pnfs_try_status
2604 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
2605 const struct rpc_call_ops
*call_ops
,
2606 struct pnfs_layout_segment
*lseg
,
2609 struct inode
*inode
= hdr
->inode
;
2610 enum pnfs_try_status trypnfs
;
2611 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2613 hdr
->mds_ops
= call_ops
;
2615 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
2616 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
2617 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
2618 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2619 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
2620 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2625 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
2626 struct nfs_pgio_header
*hdr
, int how
)
2628 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2629 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2630 enum pnfs_try_status trypnfs
;
2632 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
2634 case PNFS_NOT_ATTEMPTED
:
2635 pnfs_write_through_mds(desc
, hdr
);
2636 case PNFS_ATTEMPTED
:
2638 case PNFS_TRY_AGAIN
:
2639 /* cleanup hdr and prepare to redo pnfs */
2640 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2641 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2642 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
2643 mirror
->pg_recoalesce
= 1;
2645 hdr
->mds_ops
->rpc_release(hdr
);
2649 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
2651 pnfs_put_lseg(hdr
->lseg
);
2652 nfs_pgio_header_free(hdr
);
2656 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
2658 struct nfs_pgio_header
*hdr
;
2661 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2663 desc
->pg_error
= -ENOMEM
;
2664 return desc
->pg_error
;
2666 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
2668 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2669 ret
= nfs_generic_pgio(desc
, hdr
);
2671 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
2675 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
2677 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2679 struct nfs_pageio_descriptor pgio
;
2681 /* Resend all requests through the MDS */
2682 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
2683 return nfs_pageio_resend(&pgio
, hdr
);
2685 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
2687 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
2689 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
2690 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2691 PNFS_LAYOUTRET_ON_ERROR
) {
2692 pnfs_return_layout(hdr
->inode
);
2694 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2695 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
2699 * Called by non rpc-based layout drivers
2701 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
2703 if (likely(!hdr
->pnfs_error
))
2704 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2705 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
2706 if (unlikely(hdr
->pnfs_error
))
2707 pnfs_ld_handle_read_error(hdr
);
2708 hdr
->mds_ops
->rpc_release(hdr
);
2710 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
2713 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
2714 struct nfs_pgio_header
*hdr
)
2716 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2718 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2719 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2720 nfs_pageio_reset_read_mds(desc
);
2721 mirror
->pg_recoalesce
= 1;
2723 hdr
->completion_ops
->completion(hdr
);
2727 * Call the appropriate parallel I/O subsystem read function.
2729 static enum pnfs_try_status
2730 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
2731 const struct rpc_call_ops
*call_ops
,
2732 struct pnfs_layout_segment
*lseg
)
2734 struct inode
*inode
= hdr
->inode
;
2735 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2736 enum pnfs_try_status trypnfs
;
2738 hdr
->mds_ops
= call_ops
;
2740 dprintk("%s: Reading ino:%lu %u@%llu\n",
2741 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
2743 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
2744 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2745 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
2746 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2750 /* Resend all requests through pnfs. */
2751 void pnfs_read_resend_pnfs(struct nfs_pgio_header
*hdr
)
2753 struct nfs_pageio_descriptor pgio
;
2755 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2756 /* Prevent deadlocks with layoutreturn! */
2757 pnfs_put_lseg(hdr
->lseg
);
2760 nfs_pageio_init_read(&pgio
, hdr
->inode
, false,
2761 hdr
->completion_ops
);
2762 hdr
->task
.tk_status
= nfs_pageio_resend(&pgio
, hdr
);
2765 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs
);
2768 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
2770 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2771 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2772 enum pnfs_try_status trypnfs
;
2774 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
2776 case PNFS_NOT_ATTEMPTED
:
2777 pnfs_read_through_mds(desc
, hdr
);
2778 case PNFS_ATTEMPTED
:
2780 case PNFS_TRY_AGAIN
:
2781 /* cleanup hdr and prepare to redo pnfs */
2782 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2783 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2784 list_splice_init(&hdr
->pages
, &mirror
->pg_list
);
2785 mirror
->pg_recoalesce
= 1;
2787 hdr
->mds_ops
->rpc_release(hdr
);
2791 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
2793 pnfs_put_lseg(hdr
->lseg
);
2794 nfs_pgio_header_free(hdr
);
2798 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
2800 struct nfs_pgio_header
*hdr
;
2803 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2805 desc
->pg_error
= -ENOMEM
;
2806 return desc
->pg_error
;
2808 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
2809 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2810 ret
= nfs_generic_pgio(desc
, hdr
);
2812 pnfs_do_read(desc
, hdr
);
2815 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
2817 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
2819 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
2821 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
2822 smp_mb__after_atomic();
2823 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
2827 * There can be multiple RW segments.
2829 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
2831 struct pnfs_layout_segment
*lseg
;
2833 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
2834 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
2835 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
2836 list_add(&lseg
->pls_lc_list
, listp
);
2840 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
2842 struct pnfs_layout_segment
*lseg
, *tmp
;
2844 /* Matched by references in pnfs_set_layoutcommit */
2845 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
2846 list_del_init(&lseg
->pls_lc_list
);
2847 pnfs_put_lseg(lseg
);
2850 pnfs_clear_layoutcommitting(inode
);
2853 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
2855 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
2857 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
2860 pnfs_set_layoutcommit(struct inode
*inode
, struct pnfs_layout_segment
*lseg
,
2863 struct nfs_inode
*nfsi
= NFS_I(inode
);
2864 bool mark_as_dirty
= false;
2866 spin_lock(&inode
->i_lock
);
2867 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
2868 nfsi
->layout
->plh_lwb
= end_pos
;
2869 mark_as_dirty
= true;
2870 dprintk("%s: Set layoutcommit for inode %lu ",
2871 __func__
, inode
->i_ino
);
2872 } else if (end_pos
> nfsi
->layout
->plh_lwb
)
2873 nfsi
->layout
->plh_lwb
= end_pos
;
2874 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
)) {
2875 /* references matched in nfs4_layoutcommit_release */
2876 pnfs_get_lseg(lseg
);
2878 spin_unlock(&inode
->i_lock
);
2879 dprintk("%s: lseg %p end_pos %llu\n",
2880 __func__
, lseg
, nfsi
->layout
->plh_lwb
);
2882 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2883 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2885 mark_inode_dirty_sync(inode
);
2887 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
2889 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
2891 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
2893 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
2894 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
2895 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
2899 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2900 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2901 * data to disk to allow the server to recover the data if it crashes.
2902 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2903 * is off, and a COMMIT is sent to a data server, or
2904 * if WRITEs to a data server return NFS_DATA_SYNC.
2907 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
2909 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2910 struct nfs4_layoutcommit_data
*data
;
2911 struct nfs_inode
*nfsi
= NFS_I(inode
);
2915 if (!pnfs_layoutcommit_outstanding(inode
))
2918 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
2921 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
2924 status
= wait_on_bit_lock_action(&nfsi
->flags
,
2925 NFS_INO_LAYOUTCOMMITTING
,
2926 nfs_wait_bit_killable
,
2933 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2934 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
2936 goto clear_layoutcommitting
;
2939 spin_lock(&inode
->i_lock
);
2940 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
2943 INIT_LIST_HEAD(&data
->lseg_list
);
2944 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
2946 end_pos
= nfsi
->layout
->plh_lwb
;
2948 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
2949 spin_unlock(&inode
->i_lock
);
2951 data
->args
.inode
= inode
;
2952 data
->cred
= get_cred(nfsi
->layout
->plh_lc_cred
);
2953 nfs_fattr_init(&data
->fattr
);
2954 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
2955 data
->res
.fattr
= &data
->fattr
;
2957 data
->args
.lastbytewritten
= end_pos
- 1;
2959 data
->args
.lastbytewritten
= U64_MAX
;
2960 data
->res
.server
= NFS_SERVER(inode
);
2962 if (ld
->prepare_layoutcommit
) {
2963 status
= ld
->prepare_layoutcommit(&data
->args
);
2965 put_cred(data
->cred
);
2966 spin_lock(&inode
->i_lock
);
2967 set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
);
2968 if (end_pos
> nfsi
->layout
->plh_lwb
)
2969 nfsi
->layout
->plh_lwb
= end_pos
;
2975 status
= nfs4_proc_layoutcommit(data
, sync
);
2978 mark_inode_dirty_sync(inode
);
2979 dprintk("<-- %s status %d\n", __func__
, status
);
2982 spin_unlock(&inode
->i_lock
);
2984 clear_layoutcommitting
:
2985 pnfs_clear_layoutcommitting(inode
);
2988 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode
);
2991 pnfs_generic_sync(struct inode
*inode
, bool datasync
)
2993 return pnfs_layoutcommit_inode(inode
, true);
2995 EXPORT_SYMBOL_GPL(pnfs_generic_sync
);
2997 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
2999 struct nfs4_threshold
*thp
;
3001 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
3003 dprintk("%s mdsthreshold allocation failed\n", __func__
);
3009 #if IS_ENABLED(CONFIG_NFS_V4_2)
3011 pnfs_report_layoutstat(struct inode
*inode
, gfp_t gfp_flags
)
3013 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
3014 struct nfs_server
*server
= NFS_SERVER(inode
);
3015 struct nfs_inode
*nfsi
= NFS_I(inode
);
3016 struct nfs42_layoutstat_data
*data
;
3017 struct pnfs_layout_hdr
*hdr
;
3020 if (!pnfs_enabled_sb(server
) || !ld
->prepare_layoutstats
)
3023 if (!nfs_server_capable(inode
, NFS_CAP_LAYOUTSTATS
))
3026 if (test_and_set_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
))
3029 spin_lock(&inode
->i_lock
);
3030 if (!NFS_I(inode
)->layout
) {
3031 spin_unlock(&inode
->i_lock
);
3032 goto out_clear_layoutstats
;
3034 hdr
= NFS_I(inode
)->layout
;
3035 pnfs_get_layout_hdr(hdr
);
3036 spin_unlock(&inode
->i_lock
);
3038 data
= kzalloc(sizeof(*data
), gfp_flags
);
3044 data
->args
.fh
= NFS_FH(inode
);
3045 data
->args
.inode
= inode
;
3046 status
= ld
->prepare_layoutstats(&data
->args
);
3050 status
= nfs42_proc_layoutstats_generic(NFS_SERVER(inode
), data
);
3053 dprintk("%s returns %d\n", __func__
, status
);
3059 pnfs_put_layout_hdr(hdr
);
3060 out_clear_layoutstats
:
3061 smp_mb__before_atomic();
3062 clear_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
);
3063 smp_mb__after_atomic();
3066 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat
);
3069 unsigned int layoutstats_timer
;
3070 module_param(layoutstats_timer
, uint
, 0644);
3071 EXPORT_SYMBOL_GPL(layoutstats_timer
);