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 init_lseg(struct pnfs_layout_hdr
*lo
, struct pnfs_layout_segment
*lseg
)
339 INIT_LIST_HEAD(&lseg
->pls_list
);
340 INIT_LIST_HEAD(&lseg
->pls_lc_list
);
341 atomic_set(&lseg
->pls_refcount
, 1);
343 set_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
);
344 lseg
->pls_layout
= lo
;
347 static void pnfs_free_lseg(struct pnfs_layout_segment
*lseg
)
349 struct inode
*ino
= lseg
->pls_layout
->plh_inode
;
351 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
355 pnfs_layout_remove_lseg(struct pnfs_layout_hdr
*lo
,
356 struct pnfs_layout_segment
*lseg
)
358 struct inode
*inode
= lo
->plh_inode
;
360 WARN_ON(test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
361 list_del_init(&lseg
->pls_list
);
362 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
363 atomic_dec(&lo
->plh_refcount
);
364 if (list_empty(&lo
->plh_segs
)) {
365 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
366 clear_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
368 rpc_wake_up(&NFS_SERVER(inode
)->roc_rpcwaitq
);
372 pnfs_put_lseg(struct pnfs_layout_segment
*lseg
)
374 struct pnfs_layout_hdr
*lo
;
380 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
381 atomic_read(&lseg
->pls_refcount
),
382 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
384 lo
= lseg
->pls_layout
;
385 inode
= lo
->plh_inode
;
387 if (atomic_dec_and_lock(&lseg
->pls_refcount
, &inode
->i_lock
)) {
388 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
389 spin_unlock(&inode
->i_lock
);
392 pnfs_get_layout_hdr(lo
);
393 pnfs_layout_remove_lseg(lo
, lseg
);
394 spin_unlock(&inode
->i_lock
);
395 pnfs_free_lseg(lseg
);
396 pnfs_put_layout_hdr(lo
);
399 EXPORT_SYMBOL_GPL(pnfs_put_lseg
);
401 static void pnfs_free_lseg_async_work(struct work_struct
*work
)
403 struct pnfs_layout_segment
*lseg
;
404 struct pnfs_layout_hdr
*lo
;
406 lseg
= container_of(work
, struct pnfs_layout_segment
, pls_work
);
407 lo
= lseg
->pls_layout
;
409 pnfs_free_lseg(lseg
);
410 pnfs_put_layout_hdr(lo
);
413 static void pnfs_free_lseg_async(struct pnfs_layout_segment
*lseg
)
415 INIT_WORK(&lseg
->pls_work
, pnfs_free_lseg_async_work
);
416 schedule_work(&lseg
->pls_work
);
420 pnfs_put_lseg_locked(struct pnfs_layout_segment
*lseg
)
425 assert_spin_locked(&lseg
->pls_layout
->plh_inode
->i_lock
);
427 dprintk("%s: lseg %p ref %d valid %d\n", __func__
, lseg
,
428 atomic_read(&lseg
->pls_refcount
),
429 test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
));
430 if (atomic_dec_and_test(&lseg
->pls_refcount
)) {
431 struct pnfs_layout_hdr
*lo
= lseg
->pls_layout
;
432 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
))
434 pnfs_get_layout_hdr(lo
);
435 pnfs_layout_remove_lseg(lo
, lseg
);
436 pnfs_free_lseg_async(lseg
);
439 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked
);
442 end_offset(u64 start
, u64 len
)
447 return end
>= start
? end
: NFS4_MAX_UINT64
;
451 * is l2 fully contained in l1?
453 * [----------------------------------)
458 pnfs_lseg_range_contained(const struct pnfs_layout_range
*l1
,
459 const struct pnfs_layout_range
*l2
)
461 u64 start1
= l1
->offset
;
462 u64 end1
= end_offset(start1
, l1
->length
);
463 u64 start2
= l2
->offset
;
464 u64 end2
= end_offset(start2
, l2
->length
);
466 return (start1
<= start2
) && (end1
>= end2
);
470 * is l1 and l2 intersecting?
472 * [----------------------------------)
477 pnfs_lseg_range_intersecting(const struct pnfs_layout_range
*l1
,
478 const struct pnfs_layout_range
*l2
)
480 u64 start1
= l1
->offset
;
481 u64 end1
= end_offset(start1
, l1
->length
);
482 u64 start2
= l2
->offset
;
483 u64 end2
= end_offset(start2
, l2
->length
);
485 return (end1
== NFS4_MAX_UINT64
|| end1
> start2
) &&
486 (end2
== NFS4_MAX_UINT64
|| end2
> start1
);
490 should_free_lseg(const struct pnfs_layout_range
*lseg_range
,
491 const struct pnfs_layout_range
*recall_range
)
493 return (recall_range
->iomode
== IOMODE_ANY
||
494 lseg_range
->iomode
== recall_range
->iomode
) &&
495 pnfs_lseg_range_intersecting(lseg_range
, recall_range
);
498 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment
*lseg
,
499 struct list_head
*tmp_list
)
501 if (!atomic_dec_and_test(&lseg
->pls_refcount
))
503 pnfs_layout_remove_lseg(lseg
->pls_layout
, lseg
);
504 list_add(&lseg
->pls_list
, tmp_list
);
508 /* Returns 1 if lseg is removed from list, 0 otherwise */
509 static int mark_lseg_invalid(struct pnfs_layout_segment
*lseg
,
510 struct list_head
*tmp_list
)
514 if (test_and_clear_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
)) {
515 /* Remove the reference keeping the lseg in the
516 * list. It will now be removed when all
517 * outstanding io is finished.
519 dprintk("%s: lseg %p ref %d\n", __func__
, lseg
,
520 atomic_read(&lseg
->pls_refcount
));
521 if (pnfs_lseg_dec_and_remove_zero(lseg
, tmp_list
))
528 * Compare 2 layout stateid sequence ids, to see which is newer,
529 * taking into account wraparound issues.
531 static bool pnfs_seqid_is_newer(u32 s1
, u32 s2
)
533 return (s32
)(s1
- s2
) > 0;
537 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
538 * @lo: layout header containing the lsegs
539 * @tmp_list: list head where doomed lsegs should go
540 * @recall_range: optional recall range argument to match (may be NULL)
541 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
543 * Walk the list of lsegs in the layout header, and tear down any that should
544 * be destroyed. If "recall_range" is specified then the segment must match
545 * that range. If "seq" is non-zero, then only match segments that were handed
546 * out at or before that sequence.
548 * Returns number of matching invalid lsegs remaining in list after scanning
549 * it and purging them.
552 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr
*lo
,
553 struct list_head
*tmp_list
,
554 const struct pnfs_layout_range
*recall_range
,
557 struct pnfs_layout_segment
*lseg
, *next
;
560 dprintk("%s:Begin lo %p\n", __func__
, lo
);
562 if (list_empty(&lo
->plh_segs
))
564 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
566 should_free_lseg(&lseg
->pls_range
, recall_range
)) {
567 if (seq
&& pnfs_seqid_is_newer(lseg
->pls_seq
, seq
))
569 dprintk("%s: freeing lseg %p iomode %d seq %u"
570 "offset %llu length %llu\n", __func__
,
571 lseg
, lseg
->pls_range
.iomode
, lseg
->pls_seq
,
572 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
573 if (!mark_lseg_invalid(lseg
, tmp_list
))
576 dprintk("%s:Return %i\n", __func__
, remaining
);
580 /* note free_me must contain lsegs from a single layout_hdr */
582 pnfs_free_lseg_list(struct list_head
*free_me
)
584 struct pnfs_layout_segment
*lseg
, *tmp
;
586 if (list_empty(free_me
))
589 list_for_each_entry_safe(lseg
, tmp
, free_me
, pls_list
) {
590 list_del(&lseg
->pls_list
);
591 pnfs_free_lseg(lseg
);
596 pnfs_destroy_layout(struct nfs_inode
*nfsi
)
598 struct pnfs_layout_hdr
*lo
;
601 spin_lock(&nfsi
->vfs_inode
.i_lock
);
604 pnfs_get_layout_hdr(lo
);
605 pnfs_mark_layout_stateid_invalid(lo
, &tmp_list
);
606 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RO_FAILED
);
607 pnfs_layout_clear_fail_bit(lo
, NFS_LAYOUT_RW_FAILED
);
608 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
609 pnfs_free_lseg_list(&tmp_list
);
610 pnfs_put_layout_hdr(lo
);
612 spin_unlock(&nfsi
->vfs_inode
.i_lock
);
614 EXPORT_SYMBOL_GPL(pnfs_destroy_layout
);
617 pnfs_layout_add_bulk_destroy_list(struct inode
*inode
,
618 struct list_head
*layout_list
)
620 struct pnfs_layout_hdr
*lo
;
623 spin_lock(&inode
->i_lock
);
624 lo
= NFS_I(inode
)->layout
;
625 if (lo
!= NULL
&& list_empty(&lo
->plh_bulk_destroy
)) {
626 pnfs_get_layout_hdr(lo
);
627 list_add(&lo
->plh_bulk_destroy
, layout_list
);
630 spin_unlock(&inode
->i_lock
);
634 /* Caller must hold rcu_read_lock and clp->cl_lock */
636 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client
*clp
,
637 struct nfs_server
*server
,
638 struct list_head
*layout_list
)
640 struct pnfs_layout_hdr
*lo
, *next
;
643 list_for_each_entry_safe(lo
, next
, &server
->layouts
, plh_layouts
) {
644 inode
= igrab(lo
->plh_inode
);
647 list_del_init(&lo
->plh_layouts
);
648 if (pnfs_layout_add_bulk_destroy_list(inode
, layout_list
))
651 spin_unlock(&clp
->cl_lock
);
653 spin_lock(&clp
->cl_lock
);
661 pnfs_layout_free_bulk_destroy_list(struct list_head
*layout_list
,
664 struct pnfs_layout_hdr
*lo
;
666 LIST_HEAD(lseg_list
);
669 while (!list_empty(layout_list
)) {
670 lo
= list_entry(layout_list
->next
, struct pnfs_layout_hdr
,
672 dprintk("%s freeing layout for inode %lu\n", __func__
,
673 lo
->plh_inode
->i_ino
);
674 inode
= lo
->plh_inode
;
676 pnfs_layoutcommit_inode(inode
, false);
678 spin_lock(&inode
->i_lock
);
679 list_del_init(&lo
->plh_bulk_destroy
);
680 if (pnfs_mark_layout_stateid_invalid(lo
, &lseg_list
)) {
682 set_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
685 spin_unlock(&inode
->i_lock
);
686 pnfs_free_lseg_list(&lseg_list
);
687 /* Free all lsegs that are attached to commit buckets */
688 nfs_commit_inode(inode
, 0);
689 pnfs_put_layout_hdr(lo
);
696 pnfs_destroy_layouts_byfsid(struct nfs_client
*clp
,
697 struct nfs_fsid
*fsid
,
700 struct nfs_server
*server
;
701 LIST_HEAD(layout_list
);
703 spin_lock(&clp
->cl_lock
);
706 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
707 if (memcmp(&server
->fsid
, fsid
, sizeof(*fsid
)) != 0)
709 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
715 spin_unlock(&clp
->cl_lock
);
717 if (list_empty(&layout_list
))
719 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
723 pnfs_destroy_layouts_byclid(struct nfs_client
*clp
,
726 struct nfs_server
*server
;
727 LIST_HEAD(layout_list
);
729 spin_lock(&clp
->cl_lock
);
732 list_for_each_entry_rcu(server
, &clp
->cl_superblocks
, client_link
) {
733 if (pnfs_layout_bulk_destroy_byserver_locked(clp
,
739 spin_unlock(&clp
->cl_lock
);
741 if (list_empty(&layout_list
))
743 return pnfs_layout_free_bulk_destroy_list(&layout_list
, is_recall
);
747 * Called by the state manger to remove all layouts established under an
751 pnfs_destroy_all_layouts(struct nfs_client
*clp
)
753 nfs4_deviceid_mark_client_invalid(clp
);
754 nfs4_deviceid_purge_client(clp
);
756 pnfs_destroy_layouts_byclid(clp
, false);
759 /* update lo->plh_stateid with new if is more recent */
761 pnfs_set_layout_stateid(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*new,
764 u32 oldseq
, newseq
, new_barrier
;
765 int empty
= list_empty(&lo
->plh_segs
);
767 oldseq
= be32_to_cpu(lo
->plh_stateid
.seqid
);
768 newseq
= be32_to_cpu(new->seqid
);
769 if (empty
|| pnfs_seqid_is_newer(newseq
, oldseq
)) {
770 nfs4_stateid_copy(&lo
->plh_stateid
, new);
771 if (update_barrier
) {
772 new_barrier
= be32_to_cpu(new->seqid
);
774 /* Because of wraparound, we want to keep the barrier
775 * "close" to the current seqids.
777 new_barrier
= newseq
- atomic_read(&lo
->plh_outstanding
);
779 if (empty
|| pnfs_seqid_is_newer(new_barrier
, lo
->plh_barrier
))
780 lo
->plh_barrier
= new_barrier
;
785 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr
*lo
,
786 const nfs4_stateid
*stateid
)
788 u32 seqid
= be32_to_cpu(stateid
->seqid
);
790 return !pnfs_seqid_is_newer(seqid
, lo
->plh_barrier
);
793 /* lget is set to 1 if called from inside send_layoutget call chain */
795 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr
*lo
)
797 return lo
->plh_block_lgets
||
798 test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
);
802 * Get layout from server.
803 * for now, assume that whole file layouts are requested.
805 * arg->length: all ones
807 static struct pnfs_layout_segment
*
808 send_layoutget(struct pnfs_layout_hdr
*lo
,
809 struct nfs_open_context
*ctx
,
810 nfs4_stateid
*stateid
,
811 const struct pnfs_layout_range
*range
,
812 long *timeout
, gfp_t gfp_flags
)
814 struct inode
*ino
= lo
->plh_inode
;
815 struct nfs_server
*server
= NFS_SERVER(ino
);
816 struct nfs4_layoutget
*lgp
;
819 dprintk("--> %s\n", __func__
);
822 * Synchronously retrieve layout information from server and
823 * store in lseg. If we race with a concurrent seqid morphing
824 * op, then re-send the LAYOUTGET.
826 lgp
= kzalloc(sizeof(*lgp
), gfp_flags
);
828 return ERR_PTR(-ENOMEM
);
830 i_size
= i_size_read(ino
);
832 lgp
->args
.minlength
= PAGE_SIZE
;
833 if (lgp
->args
.minlength
> range
->length
)
834 lgp
->args
.minlength
= range
->length
;
835 if (range
->iomode
== IOMODE_READ
) {
836 if (range
->offset
>= i_size
)
837 lgp
->args
.minlength
= 0;
838 else if (i_size
- range
->offset
< lgp
->args
.minlength
)
839 lgp
->args
.minlength
= i_size
- range
->offset
;
841 lgp
->args
.maxcount
= PNFS_LAYOUT_MAXSIZE
;
842 pnfs_copy_range(&lgp
->args
.range
, range
);
843 lgp
->args
.type
= server
->pnfs_curr_ld
->id
;
844 lgp
->args
.inode
= ino
;
845 lgp
->args
.ctx
= get_nfs_open_context(ctx
);
846 nfs4_stateid_copy(&lgp
->args
.stateid
, stateid
);
847 lgp
->gfp_flags
= gfp_flags
;
848 lgp
->cred
= lo
->plh_lc_cred
;
850 return nfs4_proc_layoutget(lgp
, timeout
, gfp_flags
);
853 static void pnfs_clear_layoutcommit(struct inode
*inode
,
854 struct list_head
*head
)
856 struct nfs_inode
*nfsi
= NFS_I(inode
);
857 struct pnfs_layout_segment
*lseg
, *tmp
;
859 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
861 list_for_each_entry_safe(lseg
, tmp
, &nfsi
->layout
->plh_segs
, pls_list
) {
862 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
864 pnfs_lseg_dec_and_remove_zero(lseg
, head
);
868 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr
*lo
)
870 clear_bit_unlock(NFS_LAYOUT_RETURN
, &lo
->plh_flags
);
871 smp_mb__after_atomic();
872 wake_up_bit(&lo
->plh_flags
, NFS_LAYOUT_RETURN
);
873 rpc_wake_up(&NFS_SERVER(lo
->plh_inode
)->roc_rpcwaitq
);
877 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr
*lo
)
879 if (test_and_set_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
881 lo
->plh_return_iomode
= 0;
882 lo
->plh_return_seq
= 0;
883 pnfs_get_layout_hdr(lo
);
884 clear_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
889 pnfs_send_layoutreturn(struct pnfs_layout_hdr
*lo
, const nfs4_stateid
*stateid
,
890 enum pnfs_iomode iomode
, bool sync
)
892 struct inode
*ino
= lo
->plh_inode
;
893 struct nfs4_layoutreturn
*lrp
;
896 lrp
= kzalloc(sizeof(*lrp
), GFP_NOFS
);
897 if (unlikely(lrp
== NULL
)) {
899 spin_lock(&ino
->i_lock
);
900 pnfs_clear_layoutreturn_waitbit(lo
);
901 spin_unlock(&ino
->i_lock
);
902 pnfs_put_layout_hdr(lo
);
906 nfs4_stateid_copy(&lrp
->args
.stateid
, stateid
);
907 lrp
->args
.layout_type
= NFS_SERVER(ino
)->pnfs_curr_ld
->id
;
908 lrp
->args
.inode
= ino
;
909 lrp
->args
.range
.iomode
= iomode
;
910 lrp
->args
.range
.offset
= 0;
911 lrp
->args
.range
.length
= NFS4_MAX_UINT64
;
912 lrp
->args
.layout
= lo
;
913 lrp
->clp
= NFS_SERVER(ino
)->nfs_client
;
914 lrp
->cred
= lo
->plh_lc_cred
;
916 status
= nfs4_proc_layoutreturn(lrp
, sync
);
918 dprintk("<-- %s status: %d\n", __func__
, status
);
922 /* Return true if layoutreturn is needed */
924 pnfs_layout_need_return(struct pnfs_layout_hdr
*lo
)
926 struct pnfs_layout_segment
*s
;
928 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
931 /* Defer layoutreturn until all lsegs are done */
932 list_for_each_entry(s
, &lo
->plh_segs
, pls_list
) {
933 if (test_bit(NFS_LSEG_LAYOUTRETURN
, &s
->pls_flags
))
940 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr
*lo
)
942 struct inode
*inode
= lo
->plh_inode
;
944 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
))
946 spin_lock(&inode
->i_lock
);
947 if (pnfs_layout_need_return(lo
)) {
948 nfs4_stateid stateid
;
949 enum pnfs_iomode iomode
;
952 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
953 stateid
.seqid
= cpu_to_be32(lo
->plh_return_seq
);
954 iomode
= lo
->plh_return_iomode
;
955 send
= pnfs_prepare_layoutreturn(lo
);
956 spin_unlock(&inode
->i_lock
);
958 /* Send an async layoutreturn so we dont deadlock */
959 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
962 spin_unlock(&inode
->i_lock
);
966 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
967 * when the layout segment list is empty.
969 * Note that a pnfs_layout_hdr can exist with an empty layout segment
970 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
971 * deviceid is marked invalid.
974 _pnfs_return_layout(struct inode
*ino
)
976 struct pnfs_layout_hdr
*lo
= NULL
;
977 struct nfs_inode
*nfsi
= NFS_I(ino
);
979 nfs4_stateid stateid
;
980 int status
= 0, empty
;
983 dprintk("NFS: %s for inode %lu\n", __func__
, ino
->i_ino
);
985 spin_lock(&ino
->i_lock
);
988 spin_unlock(&ino
->i_lock
);
989 dprintk("NFS: %s no layout to return\n", __func__
);
992 nfs4_stateid_copy(&stateid
, &nfsi
->layout
->plh_stateid
);
993 /* Reference matched in nfs4_layoutreturn_release */
994 pnfs_get_layout_hdr(lo
);
995 empty
= list_empty(&lo
->plh_segs
);
996 pnfs_clear_layoutcommit(ino
, &tmp_list
);
997 pnfs_mark_matching_lsegs_invalid(lo
, &tmp_list
, NULL
, 0);
999 if (NFS_SERVER(ino
)->pnfs_curr_ld
->return_range
) {
1000 struct pnfs_layout_range range
= {
1001 .iomode
= IOMODE_ANY
,
1003 .length
= NFS4_MAX_UINT64
,
1005 NFS_SERVER(ino
)->pnfs_curr_ld
->return_range(lo
, &range
);
1008 /* Don't send a LAYOUTRETURN if list was initially empty */
1010 spin_unlock(&ino
->i_lock
);
1011 dprintk("NFS: %s no layout segments to return\n", __func__
);
1012 goto out_put_layout_hdr
;
1015 set_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
1016 send
= pnfs_prepare_layoutreturn(lo
);
1017 spin_unlock(&ino
->i_lock
);
1018 pnfs_free_lseg_list(&tmp_list
);
1020 status
= pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1022 pnfs_put_layout_hdr(lo
);
1024 dprintk("<-- %s status: %d\n", __func__
, status
);
1027 EXPORT_SYMBOL_GPL(_pnfs_return_layout
);
1030 pnfs_commit_and_return_layout(struct inode
*inode
)
1032 struct pnfs_layout_hdr
*lo
;
1035 spin_lock(&inode
->i_lock
);
1036 lo
= NFS_I(inode
)->layout
;
1038 spin_unlock(&inode
->i_lock
);
1041 pnfs_get_layout_hdr(lo
);
1042 /* Block new layoutgets and read/write to ds */
1043 lo
->plh_block_lgets
++;
1044 spin_unlock(&inode
->i_lock
);
1045 filemap_fdatawait(inode
->i_mapping
);
1046 ret
= pnfs_layoutcommit_inode(inode
, true);
1048 ret
= _pnfs_return_layout(inode
);
1049 spin_lock(&inode
->i_lock
);
1050 lo
->plh_block_lgets
--;
1051 spin_unlock(&inode
->i_lock
);
1052 pnfs_put_layout_hdr(lo
);
1056 bool pnfs_roc(struct inode
*ino
)
1058 struct nfs_inode
*nfsi
= NFS_I(ino
);
1059 struct nfs_open_context
*ctx
;
1060 struct nfs4_state
*state
;
1061 struct pnfs_layout_hdr
*lo
;
1062 struct pnfs_layout_segment
*lseg
, *tmp
;
1063 nfs4_stateid stateid
;
1064 LIST_HEAD(tmp_list
);
1065 bool found
= false, layoutreturn
= false, roc
= false;
1067 spin_lock(&ino
->i_lock
);
1069 if (!lo
|| test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
))
1072 /* no roc if we hold a delegation */
1073 if (nfs4_check_delegation(ino
, FMODE_READ
))
1076 list_for_each_entry(ctx
, &nfsi
->open_files
, list
) {
1078 /* Don't return layout if there is open file state */
1079 if (state
!= NULL
&& state
->state
!= 0)
1083 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1084 /* always send layoutreturn if being marked so */
1085 if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED
,
1087 layoutreturn
= pnfs_prepare_layoutreturn(lo
);
1089 list_for_each_entry_safe(lseg
, tmp
, &lo
->plh_segs
, pls_list
)
1090 /* If we are sending layoutreturn, invalidate all valid lsegs */
1091 if (layoutreturn
|| test_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
)) {
1092 mark_lseg_invalid(lseg
, &tmp_list
);
1095 /* ROC in two conditions:
1096 * 1. there are ROC lsegs
1097 * 2. we don't send layoutreturn
1099 if (found
&& !layoutreturn
) {
1100 /* lo ref dropped in pnfs_roc_release() */
1101 pnfs_get_layout_hdr(lo
);
1106 spin_unlock(&ino
->i_lock
);
1107 pnfs_free_lseg_list(&tmp_list
);
1108 pnfs_layoutcommit_inode(ino
, true);
1110 pnfs_send_layoutreturn(lo
, &stateid
, IOMODE_ANY
, true);
1114 void pnfs_roc_release(struct inode
*ino
)
1116 struct pnfs_layout_hdr
*lo
;
1118 spin_lock(&ino
->i_lock
);
1119 lo
= NFS_I(ino
)->layout
;
1120 pnfs_clear_layoutreturn_waitbit(lo
);
1121 if (atomic_dec_and_test(&lo
->plh_refcount
)) {
1122 pnfs_detach_layout_hdr(lo
);
1123 spin_unlock(&ino
->i_lock
);
1124 pnfs_free_layout_hdr(lo
);
1126 spin_unlock(&ino
->i_lock
);
1129 void pnfs_roc_set_barrier(struct inode
*ino
, u32 barrier
)
1131 struct pnfs_layout_hdr
*lo
;
1133 spin_lock(&ino
->i_lock
);
1134 lo
= NFS_I(ino
)->layout
;
1135 pnfs_mark_layout_returned_if_empty(lo
);
1136 if (pnfs_seqid_is_newer(barrier
, lo
->plh_barrier
))
1137 lo
->plh_barrier
= barrier
;
1138 spin_unlock(&ino
->i_lock
);
1139 trace_nfs4_layoutreturn_on_close(ino
, 0);
1142 void pnfs_roc_get_barrier(struct inode
*ino
, u32
*barrier
)
1144 struct nfs_inode
*nfsi
= NFS_I(ino
);
1145 struct pnfs_layout_hdr
*lo
;
1148 spin_lock(&ino
->i_lock
);
1150 current_seqid
= be32_to_cpu(lo
->plh_stateid
.seqid
);
1152 /* Since close does not return a layout stateid for use as
1153 * a barrier, we choose the worst-case barrier.
1155 *barrier
= current_seqid
+ atomic_read(&lo
->plh_outstanding
);
1156 spin_unlock(&ino
->i_lock
);
1159 bool pnfs_wait_on_layoutreturn(struct inode
*ino
, struct rpc_task
*task
)
1161 struct nfs_inode
*nfsi
= NFS_I(ino
);
1162 struct pnfs_layout_hdr
*lo
;
1165 /* we might not have grabbed lo reference. so need to check under
1167 spin_lock(&ino
->i_lock
);
1169 if (lo
&& test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
))
1171 spin_unlock(&ino
->i_lock
);
1174 rpc_sleep_on(&NFS_SERVER(ino
)->roc_rpcwaitq
, task
, NULL
);
1180 * Compare two layout segments for sorting into layout cache.
1181 * We want to preferentially return RW over RO layouts, so ensure those
1185 pnfs_lseg_range_cmp(const struct pnfs_layout_range
*l1
,
1186 const struct pnfs_layout_range
*l2
)
1190 /* high offset > low offset */
1191 d
= l1
->offset
- l2
->offset
;
1195 /* short length > long length */
1196 d
= l2
->length
- l1
->length
;
1200 /* read > read/write */
1201 return (int)(l1
->iomode
== IOMODE_READ
) - (int)(l2
->iomode
== IOMODE_READ
);
1205 pnfs_lseg_range_is_after(const struct pnfs_layout_range
*l1
,
1206 const struct pnfs_layout_range
*l2
)
1208 return pnfs_lseg_range_cmp(l1
, l2
) > 0;
1212 pnfs_lseg_no_merge(struct pnfs_layout_segment
*lseg
,
1213 struct pnfs_layout_segment
*old
)
1219 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1220 struct pnfs_layout_segment
*lseg
,
1221 bool (*is_after
)(const struct pnfs_layout_range
*,
1222 const struct pnfs_layout_range
*),
1223 bool (*do_merge
)(struct pnfs_layout_segment
*,
1224 struct pnfs_layout_segment
*),
1225 struct list_head
*free_me
)
1227 struct pnfs_layout_segment
*lp
, *tmp
;
1229 dprintk("%s:Begin\n", __func__
);
1231 list_for_each_entry_safe(lp
, tmp
, &lo
->plh_segs
, pls_list
) {
1232 if (test_bit(NFS_LSEG_VALID
, &lp
->pls_flags
) == 0)
1234 if (do_merge(lseg
, lp
)) {
1235 mark_lseg_invalid(lp
, free_me
);
1238 if (is_after(&lseg
->pls_range
, &lp
->pls_range
))
1240 list_add_tail(&lseg
->pls_list
, &lp
->pls_list
);
1241 dprintk("%s: inserted lseg %p "
1242 "iomode %d offset %llu length %llu before "
1243 "lp %p iomode %d offset %llu length %llu\n",
1244 __func__
, lseg
, lseg
->pls_range
.iomode
,
1245 lseg
->pls_range
.offset
, lseg
->pls_range
.length
,
1246 lp
, lp
->pls_range
.iomode
, lp
->pls_range
.offset
,
1247 lp
->pls_range
.length
);
1250 list_add_tail(&lseg
->pls_list
, &lo
->plh_segs
);
1251 dprintk("%s: inserted lseg %p "
1252 "iomode %d offset %llu length %llu at tail\n",
1253 __func__
, lseg
, lseg
->pls_range
.iomode
,
1254 lseg
->pls_range
.offset
, lseg
->pls_range
.length
);
1256 pnfs_get_layout_hdr(lo
);
1258 dprintk("%s:Return\n", __func__
);
1260 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg
);
1263 pnfs_layout_insert_lseg(struct pnfs_layout_hdr
*lo
,
1264 struct pnfs_layout_segment
*lseg
,
1265 struct list_head
*free_me
)
1267 struct inode
*inode
= lo
->plh_inode
;
1268 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
1270 if (ld
->add_lseg
!= NULL
)
1271 ld
->add_lseg(lo
, lseg
, free_me
);
1273 pnfs_generic_layout_insert_lseg(lo
, lseg
,
1274 pnfs_lseg_range_is_after
,
1279 static struct pnfs_layout_hdr
*
1280 alloc_init_layout_hdr(struct inode
*ino
,
1281 struct nfs_open_context
*ctx
,
1284 struct pnfs_layout_hdr
*lo
;
1286 lo
= pnfs_alloc_layout_hdr(ino
, gfp_flags
);
1289 atomic_set(&lo
->plh_refcount
, 1);
1290 INIT_LIST_HEAD(&lo
->plh_layouts
);
1291 INIT_LIST_HEAD(&lo
->plh_segs
);
1292 INIT_LIST_HEAD(&lo
->plh_bulk_destroy
);
1293 lo
->plh_inode
= ino
;
1294 lo
->plh_lc_cred
= get_rpccred(ctx
->cred
);
1295 lo
->plh_flags
|= 1 << NFS_LAYOUT_INVALID_STID
;
1299 static struct pnfs_layout_hdr
*
1300 pnfs_find_alloc_layout(struct inode
*ino
,
1301 struct nfs_open_context
*ctx
,
1303 __releases(&ino
->i_lock
)
1304 __acquires(&ino
->i_lock
)
1306 struct nfs_inode
*nfsi
= NFS_I(ino
);
1307 struct pnfs_layout_hdr
*new = NULL
;
1309 dprintk("%s Begin ino=%p layout=%p\n", __func__
, ino
, nfsi
->layout
);
1311 if (nfsi
->layout
!= NULL
)
1313 spin_unlock(&ino
->i_lock
);
1314 new = alloc_init_layout_hdr(ino
, ctx
, gfp_flags
);
1315 spin_lock(&ino
->i_lock
);
1317 if (likely(nfsi
->layout
== NULL
)) { /* Won the race? */
1320 } else if (new != NULL
)
1321 pnfs_free_layout_hdr(new);
1323 pnfs_get_layout_hdr(nfsi
->layout
);
1324 return nfsi
->layout
;
1328 * iomode matching rules:
1329 * iomode lseg strict match
1331 * ----- ----- ------ -----
1336 * READ READ N/A true
1337 * READ RW true false
1338 * READ RW false true
1341 pnfs_lseg_range_match(const struct pnfs_layout_range
*ls_range
,
1342 const struct pnfs_layout_range
*range
,
1345 struct pnfs_layout_range range1
;
1347 if ((range
->iomode
== IOMODE_RW
&&
1348 ls_range
->iomode
!= IOMODE_RW
) ||
1349 (range
->iomode
!= ls_range
->iomode
&&
1350 strict_iomode
== true) ||
1351 !pnfs_lseg_range_intersecting(ls_range
, range
))
1354 /* range1 covers only the first byte in the range */
1357 return pnfs_lseg_range_contained(ls_range
, &range1
);
1361 * lookup range in layout
1363 static struct pnfs_layout_segment
*
1364 pnfs_find_lseg(struct pnfs_layout_hdr
*lo
,
1365 struct pnfs_layout_range
*range
,
1368 struct pnfs_layout_segment
*lseg
, *ret
= NULL
;
1370 dprintk("%s:Begin\n", __func__
);
1372 list_for_each_entry(lseg
, &lo
->plh_segs
, pls_list
) {
1373 if (test_bit(NFS_LSEG_VALID
, &lseg
->pls_flags
) &&
1374 !test_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
) &&
1375 pnfs_lseg_range_match(&lseg
->pls_range
, range
,
1377 ret
= pnfs_get_lseg(lseg
);
1382 dprintk("%s:Return lseg %p ref %d\n",
1383 __func__
, ret
, ret
? atomic_read(&ret
->pls_refcount
) : 0);
1388 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1389 * to the MDS or over pNFS
1391 * The nfs_inode read_io and write_io fields are cumulative counters reset
1392 * when there are no layout segments. Note that in pnfs_update_layout iomode
1393 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1396 * A return of true means use MDS I/O.
1399 * If a file's size is smaller than the file size threshold, data accesses
1400 * SHOULD be sent to the metadata server. If an I/O request has a length that
1401 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1402 * server. If both file size and I/O size are provided, the client SHOULD
1403 * reach or exceed both thresholds before sending its read or write
1404 * requests to the data server.
1406 static bool pnfs_within_mdsthreshold(struct nfs_open_context
*ctx
,
1407 struct inode
*ino
, int iomode
)
1409 struct nfs4_threshold
*t
= ctx
->mdsthreshold
;
1410 struct nfs_inode
*nfsi
= NFS_I(ino
);
1411 loff_t fsize
= i_size_read(ino
);
1412 bool size
= false, size_set
= false, io
= false, io_set
= false, ret
= false;
1417 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1418 __func__
, t
->bm
, t
->rd_sz
, t
->wr_sz
, t
->rd_io_sz
, t
->wr_io_sz
);
1422 if (t
->bm
& THRESHOLD_RD
) {
1423 dprintk("%s fsize %llu\n", __func__
, fsize
);
1425 if (fsize
< t
->rd_sz
)
1428 if (t
->bm
& THRESHOLD_RD_IO
) {
1429 dprintk("%s nfsi->read_io %llu\n", __func__
,
1432 if (nfsi
->read_io
< t
->rd_io_sz
)
1437 if (t
->bm
& THRESHOLD_WR
) {
1438 dprintk("%s fsize %llu\n", __func__
, fsize
);
1440 if (fsize
< t
->wr_sz
)
1443 if (t
->bm
& THRESHOLD_WR_IO
) {
1444 dprintk("%s nfsi->write_io %llu\n", __func__
,
1447 if (nfsi
->write_io
< t
->wr_io_sz
)
1452 if (size_set
&& io_set
) {
1455 } else if (size
|| io
)
1458 dprintk("<-- %s size %d io %d ret %d\n", __func__
, size
, io
, ret
);
1462 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr
*lo
)
1465 * send layoutcommit as it can hold up layoutreturn due to lseg
1468 pnfs_layoutcommit_inode(lo
->plh_inode
, false);
1469 return !wait_on_bit_action(&lo
->plh_flags
, NFS_LAYOUT_RETURN
,
1470 nfs_wait_bit_killable
,
1471 TASK_UNINTERRUPTIBLE
);
1474 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr
*lo
)
1476 unsigned long *bitlock
= &lo
->plh_flags
;
1478 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET
, bitlock
);
1479 smp_mb__after_atomic();
1480 wake_up_bit(bitlock
, NFS_LAYOUT_FIRST_LAYOUTGET
);
1484 * Layout segment is retreived from the server if not cached.
1485 * The appropriate layout segment is referenced and returned to the caller.
1487 struct pnfs_layout_segment
*
1488 pnfs_update_layout(struct inode
*ino
,
1489 struct nfs_open_context
*ctx
,
1492 enum pnfs_iomode iomode
,
1496 struct pnfs_layout_range arg
= {
1501 unsigned pg_offset
, seq
;
1502 struct nfs_server
*server
= NFS_SERVER(ino
);
1503 struct nfs_client
*clp
= server
->nfs_client
;
1504 struct pnfs_layout_hdr
*lo
= NULL
;
1505 struct pnfs_layout_segment
*lseg
= NULL
;
1506 nfs4_stateid stateid
;
1508 unsigned long giveup
= jiffies
+ rpc_get_timeout(server
->client
);
1511 if (!pnfs_enabled_sb(NFS_SERVER(ino
))) {
1512 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1513 PNFS_UPDATE_LAYOUT_NO_PNFS
);
1517 if (iomode
== IOMODE_READ
&& i_size_read(ino
) == 0) {
1518 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1519 PNFS_UPDATE_LAYOUT_RD_ZEROLEN
);
1523 if (pnfs_within_mdsthreshold(ctx
, ino
, iomode
)) {
1524 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1525 PNFS_UPDATE_LAYOUT_MDSTHRESH
);
1531 spin_lock(&ino
->i_lock
);
1532 lo
= pnfs_find_alloc_layout(ino
, ctx
, gfp_flags
);
1534 spin_unlock(&ino
->i_lock
);
1535 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1536 PNFS_UPDATE_LAYOUT_NOMEM
);
1540 /* Do we even need to bother with this? */
1541 if (test_bit(NFS_LAYOUT_BULK_RECALL
, &lo
->plh_flags
)) {
1542 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1543 PNFS_UPDATE_LAYOUT_BULK_RECALL
);
1544 dprintk("%s matches recall, use MDS\n", __func__
);
1548 /* if LAYOUTGET already failed once we don't try again */
1549 if (pnfs_layout_io_test_failed(lo
, iomode
)) {
1550 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1551 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL
);
1555 lseg
= pnfs_find_lseg(lo
, &arg
, strict_iomode
);
1557 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1558 PNFS_UPDATE_LAYOUT_FOUND_CACHED
);
1562 if (!nfs4_valid_open_stateid(ctx
->state
)) {
1563 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1564 PNFS_UPDATE_LAYOUT_INVALID_OPEN
);
1569 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1570 * stateid, or it has been invalidated, then we must use the open
1573 if (test_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
)) {
1576 * The first layoutget for the file. Need to serialize per
1577 * RFC 5661 Errata 3208.
1579 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET
,
1581 spin_unlock(&ino
->i_lock
);
1582 wait_on_bit(&lo
->plh_flags
, NFS_LAYOUT_FIRST_LAYOUTGET
,
1583 TASK_UNINTERRUPTIBLE
);
1584 pnfs_put_layout_hdr(lo
);
1585 dprintk("%s retrying\n", __func__
);
1591 seq
= read_seqbegin(&ctx
->state
->seqlock
);
1592 nfs4_stateid_copy(&stateid
, &ctx
->state
->stateid
);
1593 } while (read_seqretry(&ctx
->state
->seqlock
, seq
));
1595 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1599 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1600 * for LAYOUTRETURN even if first is true.
1602 if (test_bit(NFS_LAYOUT_RETURN
, &lo
->plh_flags
)) {
1603 spin_unlock(&ino
->i_lock
);
1604 dprintk("%s wait for layoutreturn\n", __func__
);
1605 if (pnfs_prepare_to_retry_layoutget(lo
)) {
1607 pnfs_clear_first_layoutget(lo
);
1608 pnfs_put_layout_hdr(lo
);
1609 dprintk("%s retrying\n", __func__
);
1610 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
,
1611 lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1614 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1615 PNFS_UPDATE_LAYOUT_RETURN
);
1616 goto out_put_layout_hdr
;
1619 if (pnfs_layoutgets_blocked(lo
)) {
1620 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1621 PNFS_UPDATE_LAYOUT_BLOCKED
);
1624 atomic_inc(&lo
->plh_outstanding
);
1625 spin_unlock(&ino
->i_lock
);
1627 if (list_empty(&lo
->plh_layouts
)) {
1628 /* The lo must be on the clp list if there is any
1629 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1631 spin_lock(&clp
->cl_lock
);
1632 if (list_empty(&lo
->plh_layouts
))
1633 list_add_tail(&lo
->plh_layouts
, &server
->layouts
);
1634 spin_unlock(&clp
->cl_lock
);
1637 pg_offset
= arg
.offset
& ~PAGE_MASK
;
1639 arg
.offset
-= pg_offset
;
1640 arg
.length
+= pg_offset
;
1642 if (arg
.length
!= NFS4_MAX_UINT64
)
1643 arg
.length
= PAGE_ALIGN(arg
.length
);
1645 lseg
= send_layoutget(lo
, ctx
, &stateid
, &arg
, &timeout
, gfp_flags
);
1646 trace_pnfs_update_layout(ino
, pos
, count
, iomode
, lo
, lseg
,
1647 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET
);
1649 switch(PTR_ERR(lseg
)) {
1650 case -ERECALLCONFLICT
:
1651 if (time_after(jiffies
, giveup
))
1655 pnfs_put_layout_hdr(lo
);
1657 pnfs_clear_first_layoutget(lo
);
1659 trace_pnfs_update_layout(ino
, pos
, count
,
1660 iomode
, lo
, lseg
, PNFS_UPDATE_LAYOUT_RETRY
);
1665 if (!nfs_error_is_fatal(PTR_ERR(lseg
))) {
1666 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1671 pnfs_layout_clear_fail_bit(lo
, pnfs_iomode_to_fail_bit(iomode
));
1674 atomic_dec(&lo
->plh_outstanding
);
1677 pnfs_clear_first_layoutget(lo
);
1678 pnfs_put_layout_hdr(lo
);
1680 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1681 "(%s, offset: %llu, length: %llu)\n",
1682 __func__
, ino
->i_sb
->s_id
,
1683 (unsigned long long)NFS_FILEID(ino
),
1684 IS_ERR_OR_NULL(lseg
) ? "not found" : "found",
1685 iomode
==IOMODE_RW
? "read/write" : "read-only",
1686 (unsigned long long)pos
,
1687 (unsigned long long)count
);
1690 spin_unlock(&ino
->i_lock
);
1691 goto out_put_layout_hdr
;
1693 EXPORT_SYMBOL_GPL(pnfs_update_layout
);
1696 pnfs_sanity_check_layout_range(struct pnfs_layout_range
*range
)
1698 switch (range
->iomode
) {
1705 if (range
->offset
== NFS4_MAX_UINT64
)
1707 if (range
->length
== 0)
1709 if (range
->length
!= NFS4_MAX_UINT64
&&
1710 range
->length
> NFS4_MAX_UINT64
- range
->offset
)
1715 struct pnfs_layout_segment
*
1716 pnfs_layout_process(struct nfs4_layoutget
*lgp
)
1718 struct pnfs_layout_hdr
*lo
= NFS_I(lgp
->args
.inode
)->layout
;
1719 struct nfs4_layoutget_res
*res
= &lgp
->res
;
1720 struct pnfs_layout_segment
*lseg
;
1721 struct inode
*ino
= lo
->plh_inode
;
1724 if (!pnfs_sanity_check_layout_range(&res
->range
))
1725 return ERR_PTR(-EINVAL
);
1727 /* Inject layout blob into I/O device driver */
1728 lseg
= NFS_SERVER(ino
)->pnfs_curr_ld
->alloc_lseg(lo
, res
, lgp
->gfp_flags
);
1729 if (IS_ERR_OR_NULL(lseg
)) {
1731 lseg
= ERR_PTR(-ENOMEM
);
1733 dprintk("%s: Could not allocate layout: error %ld\n",
1734 __func__
, PTR_ERR(lseg
));
1738 init_lseg(lo
, lseg
);
1739 lseg
->pls_range
= res
->range
;
1740 lseg
->pls_seq
= be32_to_cpu(res
->stateid
.seqid
);
1742 spin_lock(&ino
->i_lock
);
1743 if (pnfs_layoutgets_blocked(lo
)) {
1744 dprintk("%s forget reply due to state\n", __func__
);
1748 if (nfs4_stateid_match_other(&lo
->plh_stateid
, &res
->stateid
)) {
1749 /* existing state ID, make sure the sequence number matches. */
1750 if (pnfs_layout_stateid_blocked(lo
, &res
->stateid
)) {
1751 dprintk("%s forget reply due to sequence\n", __func__
);
1754 pnfs_set_layout_stateid(lo
, &res
->stateid
, false);
1757 * We got an entirely new state ID. Mark all segments for the
1758 * inode invalid, and don't bother validating the stateid
1761 pnfs_mark_matching_lsegs_invalid(lo
, &free_me
, NULL
, 0);
1763 nfs4_stateid_copy(&lo
->plh_stateid
, &res
->stateid
);
1764 lo
->plh_barrier
= be32_to_cpu(res
->stateid
.seqid
);
1767 clear_bit(NFS_LAYOUT_INVALID_STID
, &lo
->plh_flags
);
1769 pnfs_get_lseg(lseg
);
1770 pnfs_layout_insert_lseg(lo
, lseg
, &free_me
);
1772 if (res
->return_on_close
)
1773 set_bit(NFS_LSEG_ROC
, &lseg
->pls_flags
);
1775 spin_unlock(&ino
->i_lock
);
1776 pnfs_free_lseg_list(&free_me
);
1780 spin_unlock(&ino
->i_lock
);
1781 lseg
->pls_layout
= lo
;
1782 NFS_SERVER(ino
)->pnfs_curr_ld
->free_lseg(lseg
);
1783 return ERR_PTR(-EAGAIN
);
1787 pnfs_set_plh_return_info(struct pnfs_layout_hdr
*lo
, enum pnfs_iomode iomode
,
1790 if (lo
->plh_return_iomode
== iomode
)
1792 if (lo
->plh_return_iomode
!= 0)
1793 iomode
= IOMODE_ANY
;
1794 lo
->plh_return_iomode
= iomode
;
1795 set_bit(NFS_LAYOUT_RETURN_REQUESTED
, &lo
->plh_flags
);
1796 if (!lo
->plh_return_seq
|| pnfs_seqid_is_newer(seq
, lo
->plh_return_seq
))
1797 lo
->plh_return_seq
= seq
;
1801 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
1802 * @lo: pointer to layout header
1803 * @tmp_list: list header to be used with pnfs_free_lseg_list()
1804 * @return_range: describe layout segment ranges to be returned
1806 * This function is mainly intended for use by layoutrecall. It attempts
1807 * to free the layout segment immediately, or else to mark it for return
1808 * as soon as its reference count drops to zero.
1811 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr
*lo
,
1812 struct list_head
*tmp_list
,
1813 const struct pnfs_layout_range
*return_range
,
1816 struct pnfs_layout_segment
*lseg
, *next
;
1819 dprintk("%s:Begin lo %p\n", __func__
, lo
);
1821 if (list_empty(&lo
->plh_segs
))
1824 assert_spin_locked(&lo
->plh_inode
->i_lock
);
1826 list_for_each_entry_safe(lseg
, next
, &lo
->plh_segs
, pls_list
)
1827 if (should_free_lseg(&lseg
->pls_range
, return_range
)) {
1828 dprintk("%s: marking lseg %p iomode %d "
1829 "offset %llu length %llu\n", __func__
,
1830 lseg
, lseg
->pls_range
.iomode
,
1831 lseg
->pls_range
.offset
,
1832 lseg
->pls_range
.length
);
1833 if (mark_lseg_invalid(lseg
, tmp_list
))
1836 set_bit(NFS_LSEG_LAYOUTRETURN
, &lseg
->pls_flags
);
1840 pnfs_set_plh_return_info(lo
, return_range
->iomode
, seq
);
1845 void pnfs_error_mark_layout_for_return(struct inode
*inode
,
1846 struct pnfs_layout_segment
*lseg
)
1848 struct pnfs_layout_hdr
*lo
= NFS_I(inode
)->layout
;
1849 struct pnfs_layout_range range
= {
1850 .iomode
= lseg
->pls_range
.iomode
,
1852 .length
= NFS4_MAX_UINT64
,
1855 bool return_now
= false;
1857 spin_lock(&inode
->i_lock
);
1858 pnfs_set_plh_return_info(lo
, range
.iomode
, lseg
->pls_seq
);
1860 * mark all matching lsegs so that we are sure to have no live
1861 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1864 if (!pnfs_mark_matching_lsegs_return(lo
, &free_me
,
1865 &range
, lseg
->pls_seq
)) {
1866 nfs4_stateid stateid
;
1867 enum pnfs_iomode iomode
= lo
->plh_return_iomode
;
1869 nfs4_stateid_copy(&stateid
, &lo
->plh_stateid
);
1870 return_now
= pnfs_prepare_layoutreturn(lo
);
1871 spin_unlock(&inode
->i_lock
);
1873 pnfs_send_layoutreturn(lo
, &stateid
, iomode
, false);
1875 spin_unlock(&inode
->i_lock
);
1876 nfs_commit_inode(inode
, 0);
1878 pnfs_free_lseg_list(&free_me
);
1880 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return
);
1883 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor
*pgio
, struct nfs_page
*req
)
1885 u64 rd_size
= req
->wb_bytes
;
1887 if (pgio
->pg_lseg
== NULL
) {
1888 if (pgio
->pg_dreq
== NULL
)
1889 rd_size
= i_size_read(pgio
->pg_inode
) - req_offset(req
);
1891 rd_size
= nfs_dreq_bytes_left(pgio
->pg_dreq
);
1893 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1900 if (IS_ERR(pgio
->pg_lseg
)) {
1901 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
1902 pgio
->pg_lseg
= NULL
;
1906 /* If no lseg, fall back to read through mds */
1907 if (pgio
->pg_lseg
== NULL
)
1908 nfs_pageio_reset_read_mds(pgio
);
1911 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read
);
1914 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor
*pgio
,
1915 struct nfs_page
*req
, u64 wb_size
)
1917 if (pgio
->pg_lseg
== NULL
) {
1918 pgio
->pg_lseg
= pnfs_update_layout(pgio
->pg_inode
,
1925 if (IS_ERR(pgio
->pg_lseg
)) {
1926 pgio
->pg_error
= PTR_ERR(pgio
->pg_lseg
);
1927 pgio
->pg_lseg
= NULL
;
1931 /* If no lseg, fall back to write through mds */
1932 if (pgio
->pg_lseg
== NULL
)
1933 nfs_pageio_reset_write_mds(pgio
);
1935 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write
);
1938 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor
*desc
)
1940 if (desc
->pg_lseg
) {
1941 pnfs_put_lseg(desc
->pg_lseg
);
1942 desc
->pg_lseg
= NULL
;
1945 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup
);
1948 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1949 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1952 pnfs_generic_pg_test(struct nfs_pageio_descriptor
*pgio
,
1953 struct nfs_page
*prev
, struct nfs_page
*req
)
1956 u64 seg_end
, req_start
, seg_left
;
1958 size
= nfs_generic_pg_test(pgio
, prev
, req
);
1963 * 'size' contains the number of bytes left in the current page (up
1964 * to the original size asked for in @req->wb_bytes).
1966 * Calculate how many bytes are left in the layout segment
1967 * and if there are less bytes than 'size', return that instead.
1969 * Please also note that 'end_offset' is actually the offset of the
1970 * first byte that lies outside the pnfs_layout_range. FIXME?
1973 if (pgio
->pg_lseg
) {
1974 seg_end
= end_offset(pgio
->pg_lseg
->pls_range
.offset
,
1975 pgio
->pg_lseg
->pls_range
.length
);
1976 req_start
= req_offset(req
);
1977 WARN_ON_ONCE(req_start
>= seg_end
);
1978 /* start of request is past the last byte of this segment */
1979 if (req_start
>= seg_end
) {
1980 /* reference the new lseg */
1981 if (pgio
->pg_ops
->pg_cleanup
)
1982 pgio
->pg_ops
->pg_cleanup(pgio
);
1983 if (pgio
->pg_ops
->pg_init
)
1984 pgio
->pg_ops
->pg_init(pgio
, req
);
1988 /* adjust 'size' iff there are fewer bytes left in the
1989 * segment than what nfs_generic_pg_test returned */
1990 seg_left
= seg_end
- req_start
;
1991 if (seg_left
< size
)
1992 size
= (unsigned int)seg_left
;
1997 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test
);
1999 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2001 struct nfs_pageio_descriptor pgio
;
2003 /* Resend all requests through the MDS */
2004 nfs_pageio_init_write(&pgio
, hdr
->inode
, FLUSH_STABLE
, true,
2005 hdr
->completion_ops
);
2006 set_bit(NFS_CONTEXT_RESEND_WRITES
, &hdr
->args
.context
->flags
);
2007 return nfs_pageio_resend(&pgio
, hdr
);
2009 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds
);
2011 static void pnfs_ld_handle_write_error(struct nfs_pgio_header
*hdr
)
2014 dprintk("pnfs write error = %d\n", hdr
->pnfs_error
);
2015 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2016 PNFS_LAYOUTRET_ON_ERROR
) {
2017 pnfs_return_layout(hdr
->inode
);
2019 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2020 hdr
->task
.tk_status
= pnfs_write_done_resend_to_mds(hdr
);
2024 * Called by non rpc-based layout drivers
2026 void pnfs_ld_write_done(struct nfs_pgio_header
*hdr
)
2028 if (likely(!hdr
->pnfs_error
)) {
2029 pnfs_set_layoutcommit(hdr
->inode
, hdr
->lseg
,
2030 hdr
->mds_offset
+ hdr
->res
.count
);
2031 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2033 trace_nfs4_pnfs_write(hdr
, hdr
->pnfs_error
);
2034 if (unlikely(hdr
->pnfs_error
))
2035 pnfs_ld_handle_write_error(hdr
);
2036 hdr
->mds_ops
->rpc_release(hdr
);
2038 EXPORT_SYMBOL_GPL(pnfs_ld_write_done
);
2041 pnfs_write_through_mds(struct nfs_pageio_descriptor
*desc
,
2042 struct nfs_pgio_header
*hdr
)
2044 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2046 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2047 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2048 nfs_pageio_reset_write_mds(desc
);
2049 mirror
->pg_recoalesce
= 1;
2051 nfs_pgio_data_destroy(hdr
);
2055 static enum pnfs_try_status
2056 pnfs_try_to_write_data(struct nfs_pgio_header
*hdr
,
2057 const struct rpc_call_ops
*call_ops
,
2058 struct pnfs_layout_segment
*lseg
,
2061 struct inode
*inode
= hdr
->inode
;
2062 enum pnfs_try_status trypnfs
;
2063 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2065 hdr
->mds_ops
= call_ops
;
2067 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__
,
2068 inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
, how
);
2069 trypnfs
= nfss
->pnfs_curr_ld
->write_pagelist(hdr
, how
);
2070 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2071 nfs_inc_stats(inode
, NFSIOS_PNFS_WRITE
);
2072 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2077 pnfs_do_write(struct nfs_pageio_descriptor
*desc
,
2078 struct nfs_pgio_header
*hdr
, int how
)
2080 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2081 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2082 enum pnfs_try_status trypnfs
;
2084 trypnfs
= pnfs_try_to_write_data(hdr
, call_ops
, lseg
, how
);
2085 if (trypnfs
== PNFS_NOT_ATTEMPTED
)
2086 pnfs_write_through_mds(desc
, hdr
);
2089 static void pnfs_writehdr_free(struct nfs_pgio_header
*hdr
)
2091 pnfs_put_lseg(hdr
->lseg
);
2092 nfs_pgio_header_free(hdr
);
2096 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor
*desc
)
2098 struct nfs_pgio_header
*hdr
;
2101 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2103 desc
->pg_error
= -ENOMEM
;
2104 return desc
->pg_error
;
2106 nfs_pgheader_init(desc
, hdr
, pnfs_writehdr_free
);
2108 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2109 ret
= nfs_generic_pgio(desc
, hdr
);
2111 pnfs_do_write(desc
, hdr
, desc
->pg_ioflags
);
2115 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages
);
2117 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header
*hdr
)
2119 struct nfs_pageio_descriptor pgio
;
2121 /* Resend all requests through the MDS */
2122 nfs_pageio_init_read(&pgio
, hdr
->inode
, true, hdr
->completion_ops
);
2123 return nfs_pageio_resend(&pgio
, hdr
);
2125 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds
);
2127 static void pnfs_ld_handle_read_error(struct nfs_pgio_header
*hdr
)
2129 dprintk("pnfs read error = %d\n", hdr
->pnfs_error
);
2130 if (NFS_SERVER(hdr
->inode
)->pnfs_curr_ld
->flags
&
2131 PNFS_LAYOUTRET_ON_ERROR
) {
2132 pnfs_return_layout(hdr
->inode
);
2134 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
))
2135 hdr
->task
.tk_status
= pnfs_read_done_resend_to_mds(hdr
);
2139 * Called by non rpc-based layout drivers
2141 void pnfs_ld_read_done(struct nfs_pgio_header
*hdr
)
2143 if (likely(!hdr
->pnfs_error
)) {
2144 __nfs4_read_done_cb(hdr
);
2145 hdr
->mds_ops
->rpc_call_done(&hdr
->task
, hdr
);
2147 trace_nfs4_pnfs_read(hdr
, hdr
->pnfs_error
);
2148 if (unlikely(hdr
->pnfs_error
))
2149 pnfs_ld_handle_read_error(hdr
);
2150 hdr
->mds_ops
->rpc_release(hdr
);
2152 EXPORT_SYMBOL_GPL(pnfs_ld_read_done
);
2155 pnfs_read_through_mds(struct nfs_pageio_descriptor
*desc
,
2156 struct nfs_pgio_header
*hdr
)
2158 struct nfs_pgio_mirror
*mirror
= nfs_pgio_current_mirror(desc
);
2160 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2161 list_splice_tail_init(&hdr
->pages
, &mirror
->pg_list
);
2162 nfs_pageio_reset_read_mds(desc
);
2163 mirror
->pg_recoalesce
= 1;
2165 nfs_pgio_data_destroy(hdr
);
2170 * Call the appropriate parallel I/O subsystem read function.
2172 static enum pnfs_try_status
2173 pnfs_try_to_read_data(struct nfs_pgio_header
*hdr
,
2174 const struct rpc_call_ops
*call_ops
,
2175 struct pnfs_layout_segment
*lseg
)
2177 struct inode
*inode
= hdr
->inode
;
2178 struct nfs_server
*nfss
= NFS_SERVER(inode
);
2179 enum pnfs_try_status trypnfs
;
2181 hdr
->mds_ops
= call_ops
;
2183 dprintk("%s: Reading ino:%lu %u@%llu\n",
2184 __func__
, inode
->i_ino
, hdr
->args
.count
, hdr
->args
.offset
);
2186 trypnfs
= nfss
->pnfs_curr_ld
->read_pagelist(hdr
);
2187 if (trypnfs
!= PNFS_NOT_ATTEMPTED
)
2188 nfs_inc_stats(inode
, NFSIOS_PNFS_READ
);
2189 dprintk("%s End (trypnfs:%d)\n", __func__
, trypnfs
);
2193 /* Resend all requests through pnfs. */
2194 void pnfs_read_resend_pnfs(struct nfs_pgio_header
*hdr
)
2196 struct nfs_pageio_descriptor pgio
;
2198 if (!test_and_set_bit(NFS_IOHDR_REDO
, &hdr
->flags
)) {
2199 nfs_pageio_init_read(&pgio
, hdr
->inode
, false,
2200 hdr
->completion_ops
);
2201 hdr
->task
.tk_status
= nfs_pageio_resend(&pgio
, hdr
);
2204 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs
);
2207 pnfs_do_read(struct nfs_pageio_descriptor
*desc
, struct nfs_pgio_header
*hdr
)
2209 const struct rpc_call_ops
*call_ops
= desc
->pg_rpc_callops
;
2210 struct pnfs_layout_segment
*lseg
= desc
->pg_lseg
;
2211 enum pnfs_try_status trypnfs
;
2213 trypnfs
= pnfs_try_to_read_data(hdr
, call_ops
, lseg
);
2214 if (trypnfs
== PNFS_TRY_AGAIN
)
2215 pnfs_read_resend_pnfs(hdr
);
2216 if (trypnfs
== PNFS_NOT_ATTEMPTED
|| hdr
->task
.tk_status
)
2217 pnfs_read_through_mds(desc
, hdr
);
2220 static void pnfs_readhdr_free(struct nfs_pgio_header
*hdr
)
2222 pnfs_put_lseg(hdr
->lseg
);
2223 nfs_pgio_header_free(hdr
);
2227 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor
*desc
)
2229 struct nfs_pgio_header
*hdr
;
2232 hdr
= nfs_pgio_header_alloc(desc
->pg_rw_ops
);
2234 desc
->pg_error
= -ENOMEM
;
2235 return desc
->pg_error
;
2237 nfs_pgheader_init(desc
, hdr
, pnfs_readhdr_free
);
2238 hdr
->lseg
= pnfs_get_lseg(desc
->pg_lseg
);
2239 ret
= nfs_generic_pgio(desc
, hdr
);
2241 pnfs_do_read(desc
, hdr
);
2244 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages
);
2246 static void pnfs_clear_layoutcommitting(struct inode
*inode
)
2248 unsigned long *bitlock
= &NFS_I(inode
)->flags
;
2250 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING
, bitlock
);
2251 smp_mb__after_atomic();
2252 wake_up_bit(bitlock
, NFS_INO_LAYOUTCOMMITTING
);
2256 * There can be multiple RW segments.
2258 static void pnfs_list_write_lseg(struct inode
*inode
, struct list_head
*listp
)
2260 struct pnfs_layout_segment
*lseg
;
2262 list_for_each_entry(lseg
, &NFS_I(inode
)->layout
->plh_segs
, pls_list
) {
2263 if (lseg
->pls_range
.iomode
== IOMODE_RW
&&
2264 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
))
2265 list_add(&lseg
->pls_lc_list
, listp
);
2269 static void pnfs_list_write_lseg_done(struct inode
*inode
, struct list_head
*listp
)
2271 struct pnfs_layout_segment
*lseg
, *tmp
;
2273 /* Matched by references in pnfs_set_layoutcommit */
2274 list_for_each_entry_safe(lseg
, tmp
, listp
, pls_lc_list
) {
2275 list_del_init(&lseg
->pls_lc_list
);
2276 pnfs_put_lseg(lseg
);
2279 pnfs_clear_layoutcommitting(inode
);
2282 void pnfs_set_lo_fail(struct pnfs_layout_segment
*lseg
)
2284 pnfs_layout_io_set_failed(lseg
->pls_layout
, lseg
->pls_range
.iomode
);
2286 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail
);
2289 pnfs_set_layoutcommit(struct inode
*inode
, struct pnfs_layout_segment
*lseg
,
2292 struct nfs_inode
*nfsi
= NFS_I(inode
);
2293 bool mark_as_dirty
= false;
2295 spin_lock(&inode
->i_lock
);
2296 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
)) {
2297 nfsi
->layout
->plh_lwb
= end_pos
;
2298 mark_as_dirty
= true;
2299 dprintk("%s: Set layoutcommit for inode %lu ",
2300 __func__
, inode
->i_ino
);
2301 } else if (end_pos
> nfsi
->layout
->plh_lwb
)
2302 nfsi
->layout
->plh_lwb
= end_pos
;
2303 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT
, &lseg
->pls_flags
)) {
2304 /* references matched in nfs4_layoutcommit_release */
2305 pnfs_get_lseg(lseg
);
2307 spin_unlock(&inode
->i_lock
);
2308 dprintk("%s: lseg %p end_pos %llu\n",
2309 __func__
, lseg
, nfsi
->layout
->plh_lwb
);
2311 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2312 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2314 mark_inode_dirty_sync(inode
);
2316 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit
);
2318 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data
*data
)
2320 struct nfs_server
*nfss
= NFS_SERVER(data
->args
.inode
);
2322 if (nfss
->pnfs_curr_ld
->cleanup_layoutcommit
)
2323 nfss
->pnfs_curr_ld
->cleanup_layoutcommit(data
);
2324 pnfs_list_write_lseg_done(data
->args
.inode
, &data
->lseg_list
);
2328 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2329 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2330 * data to disk to allow the server to recover the data if it crashes.
2331 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2332 * is off, and a COMMIT is sent to a data server, or
2333 * if WRITEs to a data server return NFS_DATA_SYNC.
2336 pnfs_layoutcommit_inode(struct inode
*inode
, bool sync
)
2338 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2339 struct nfs4_layoutcommit_data
*data
;
2340 struct nfs_inode
*nfsi
= NFS_I(inode
);
2344 if (!pnfs_layoutcommit_outstanding(inode
))
2347 dprintk("--> %s inode %lu\n", __func__
, inode
->i_ino
);
2350 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING
, &nfsi
->flags
)) {
2353 status
= wait_on_bit_lock_action(&nfsi
->flags
,
2354 NFS_INO_LAYOUTCOMMITTING
,
2355 nfs_wait_bit_killable
,
2362 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2363 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
2365 goto clear_layoutcommitting
;
2368 spin_lock(&inode
->i_lock
);
2369 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
))
2372 INIT_LIST_HEAD(&data
->lseg_list
);
2373 pnfs_list_write_lseg(inode
, &data
->lseg_list
);
2375 end_pos
= nfsi
->layout
->plh_lwb
;
2377 nfs4_stateid_copy(&data
->args
.stateid
, &nfsi
->layout
->plh_stateid
);
2378 spin_unlock(&inode
->i_lock
);
2380 data
->args
.inode
= inode
;
2381 data
->cred
= get_rpccred(nfsi
->layout
->plh_lc_cred
);
2382 nfs_fattr_init(&data
->fattr
);
2383 data
->args
.bitmask
= NFS_SERVER(inode
)->cache_consistency_bitmask
;
2384 data
->res
.fattr
= &data
->fattr
;
2385 data
->args
.lastbytewritten
= end_pos
- 1;
2386 data
->res
.server
= NFS_SERVER(inode
);
2388 if (ld
->prepare_layoutcommit
) {
2389 status
= ld
->prepare_layoutcommit(&data
->args
);
2391 put_rpccred(data
->cred
);
2392 spin_lock(&inode
->i_lock
);
2393 set_bit(NFS_INO_LAYOUTCOMMIT
, &nfsi
->flags
);
2394 if (end_pos
> nfsi
->layout
->plh_lwb
)
2395 nfsi
->layout
->plh_lwb
= end_pos
;
2401 status
= nfs4_proc_layoutcommit(data
, sync
);
2404 mark_inode_dirty_sync(inode
);
2405 dprintk("<-- %s status %d\n", __func__
, status
);
2408 spin_unlock(&inode
->i_lock
);
2410 clear_layoutcommitting
:
2411 pnfs_clear_layoutcommitting(inode
);
2414 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode
);
2417 pnfs_generic_sync(struct inode
*inode
, bool datasync
)
2419 return pnfs_layoutcommit_inode(inode
, true);
2421 EXPORT_SYMBOL_GPL(pnfs_generic_sync
);
2423 struct nfs4_threshold
*pnfs_mdsthreshold_alloc(void)
2425 struct nfs4_threshold
*thp
;
2427 thp
= kzalloc(sizeof(*thp
), GFP_NOFS
);
2429 dprintk("%s mdsthreshold allocation failed\n", __func__
);
2435 #if IS_ENABLED(CONFIG_NFS_V4_2)
2437 pnfs_report_layoutstat(struct inode
*inode
, gfp_t gfp_flags
)
2439 struct pnfs_layoutdriver_type
*ld
= NFS_SERVER(inode
)->pnfs_curr_ld
;
2440 struct nfs_server
*server
= NFS_SERVER(inode
);
2441 struct nfs_inode
*nfsi
= NFS_I(inode
);
2442 struct nfs42_layoutstat_data
*data
;
2443 struct pnfs_layout_hdr
*hdr
;
2446 if (!pnfs_enabled_sb(server
) || !ld
->prepare_layoutstats
)
2449 if (!nfs_server_capable(inode
, NFS_CAP_LAYOUTSTATS
))
2452 if (test_and_set_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
))
2455 spin_lock(&inode
->i_lock
);
2456 if (!NFS_I(inode
)->layout
) {
2457 spin_unlock(&inode
->i_lock
);
2458 goto out_clear_layoutstats
;
2460 hdr
= NFS_I(inode
)->layout
;
2461 pnfs_get_layout_hdr(hdr
);
2462 spin_unlock(&inode
->i_lock
);
2464 data
= kzalloc(sizeof(*data
), gfp_flags
);
2470 data
->args
.fh
= NFS_FH(inode
);
2471 data
->args
.inode
= inode
;
2472 nfs4_stateid_copy(&data
->args
.stateid
, &hdr
->plh_stateid
);
2473 status
= ld
->prepare_layoutstats(&data
->args
);
2477 status
= nfs42_proc_layoutstats_generic(NFS_SERVER(inode
), data
);
2480 dprintk("%s returns %d\n", __func__
, status
);
2486 pnfs_put_layout_hdr(hdr
);
2487 out_clear_layoutstats
:
2488 smp_mb__before_atomic();
2489 clear_bit(NFS_INO_LAYOUTSTATS
, &nfsi
->flags
);
2490 smp_mb__after_atomic();
2493 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat
);
2496 unsigned int layoutstats_timer
;
2497 module_param(layoutstats_timer
, uint
, 0644);
2498 EXPORT_SYMBOL_GPL(layoutstats_timer
);