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);
56 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
57 enum pnfs_iomode iomode, bool sync);
59 /* Return the registered pnfs layout driver module matching given id */
60 static struct pnfs_layoutdriver_type *
61 find_pnfs_driver_locked(u32 id)
63 struct pnfs_layoutdriver_type *local;
65 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
70 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
74 static struct pnfs_layoutdriver_type *
75 find_pnfs_driver(u32 id)
77 struct pnfs_layoutdriver_type *local;
79 spin_lock(&pnfs_spinlock);
80 local = find_pnfs_driver_locked(id);
81 if (local != NULL && !try_module_get(local->owner)) {
82 dprintk("%s: Could not grab reference on module\n", __func__);
85 spin_unlock(&pnfs_spinlock);
90 unset_pnfs_layoutdriver(struct nfs_server *nfss)
92 if (nfss->pnfs_curr_ld) {
93 if (nfss->pnfs_curr_ld->clear_layoutdriver)
94 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
95 /* Decrement the MDS count. Purge the deviceid cache if zero */
96 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
97 nfs4_deviceid_purge_client(nfss->nfs_client);
98 module_put(nfss->pnfs_curr_ld->owner);
100 nfss->pnfs_curr_ld = NULL;
104 * Try to set the server's pnfs module to the pnfs layout type specified by id.
105 * Currently only one pNFS layout driver per filesystem is supported.
107 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
110 set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
113 struct pnfs_layoutdriver_type *ld_type = NULL;
117 if (!(server->nfs_client->cl_exchange_flags &
118 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
119 printk(KERN_ERR "NFS: %s: id %u cl_exchange_flags 0x%x\n",
120 __func__, id, server->nfs_client->cl_exchange_flags);
123 ld_type = find_pnfs_driver(id);
125 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
126 ld_type = find_pnfs_driver(id);
128 dprintk("%s: No pNFS module found for %u.\n",
133 server->pnfs_curr_ld = ld_type;
134 if (ld_type->set_layoutdriver
135 && ld_type->set_layoutdriver(server, mntfh)) {
136 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
137 "driver %u.\n", __func__, id);
138 module_put(ld_type->owner);
141 /* Bump the MDS count */
142 atomic_inc(&server->nfs_client->cl_mds_count);
144 dprintk("%s: pNFS module for %u set\n", __func__, id);
148 dprintk("%s: Using NFSv4 I/O\n", __func__);
149 server->pnfs_curr_ld = NULL;
153 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
155 int status = -EINVAL;
156 struct pnfs_layoutdriver_type *tmp;
158 if (ld_type->id == 0) {
159 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
162 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
163 printk(KERN_ERR "NFS: %s Layout driver must provide "
164 "alloc_lseg and free_lseg.\n", __func__);
168 spin_lock(&pnfs_spinlock);
169 tmp = find_pnfs_driver_locked(ld_type->id);
171 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
173 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
176 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
177 __func__, ld_type->id);
179 spin_unlock(&pnfs_spinlock);
183 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
186 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
188 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
189 spin_lock(&pnfs_spinlock);
190 list_del(&ld_type->pnfs_tblid);
191 spin_unlock(&pnfs_spinlock);
193 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
196 * pNFS client layout cache
199 /* Need to hold i_lock if caller does not already hold reference */
201 pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
203 atomic_inc(&lo->plh_refcount);
206 static struct pnfs_layout_hdr *
207 pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
209 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
210 return ld->alloc_layout_hdr(ino, gfp_flags);
214 pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
216 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
217 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
219 if (!list_empty(&lo->plh_layouts)) {
220 struct nfs_client *clp = server->nfs_client;
222 spin_lock(&clp->cl_lock);
223 list_del_init(&lo->plh_layouts);
224 spin_unlock(&clp->cl_lock);
226 put_rpccred(lo->plh_lc_cred);
227 return ld->free_layout_hdr(lo);
231 pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
233 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
234 dprintk("%s: freeing layout cache %p\n", __func__, lo);
236 /* Reset MDS Threshold I/O counters */
242 pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
244 struct inode *inode = lo->plh_inode;
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 pnfs_iomode_to_fail_bit(u32 iomode)
258 return iomode == IOMODE_RW ?
259 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
263 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
265 lo->plh_retry_timestamp = jiffies;
266 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
267 atomic_inc(&lo->plh_refcount);
271 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
273 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
274 atomic_dec(&lo->plh_refcount);
278 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
280 struct inode *inode = lo->plh_inode;
281 struct pnfs_layout_range range = {
284 .length = NFS4_MAX_UINT64,
288 spin_lock(&inode->i_lock);
289 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
290 pnfs_mark_matching_lsegs_invalid(lo, &head, &range);
291 spin_unlock(&inode->i_lock);
292 pnfs_free_lseg_list(&head);
293 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
294 iomode == IOMODE_RW ? "RW" : "READ");
298 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
300 unsigned long start, end;
301 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
303 if (test_bit(fail_bit, &lo->plh_flags) == 0)
306 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
307 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
308 /* It is time to retry the failed layoutgets */
309 pnfs_layout_clear_fail_bit(lo, fail_bit);
316 init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
318 INIT_LIST_HEAD(&lseg->pls_list);
319 INIT_LIST_HEAD(&lseg->pls_lc_list);
320 atomic_set(&lseg->pls_refcount, 1);
322 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
323 lseg->pls_layout = lo;
326 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
328 struct inode *ino = lseg->pls_layout->plh_inode;
330 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
334 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
335 struct pnfs_layout_segment *lseg)
337 struct inode *inode = lo->plh_inode;
339 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
340 list_del_init(&lseg->pls_list);
341 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
342 atomic_dec(&lo->plh_refcount);
343 if (list_empty(&lo->plh_segs))
344 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
345 rpc_wake_up(&NFS_SERVER(inode)->roc_rpcwaitq);
348 /* Return true if layoutreturn is needed */
350 pnfs_layout_need_return(struct pnfs_layout_hdr *lo,
351 struct pnfs_layout_segment *lseg)
353 struct pnfs_layout_segment *s;
355 if (!test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
358 list_for_each_entry(s, &lo->plh_segs, pls_list)
359 if (s != lseg && test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
366 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
368 if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
370 lo->plh_return_iomode = 0;
371 pnfs_get_layout_hdr(lo);
372 clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
376 static void pnfs_layoutreturn_before_put_lseg(struct pnfs_layout_segment *lseg,
377 struct pnfs_layout_hdr *lo, struct inode *inode)
379 lo = lseg->pls_layout;
380 inode = lo->plh_inode;
382 spin_lock(&inode->i_lock);
383 if (pnfs_layout_need_return(lo, lseg)) {
384 nfs4_stateid stateid;
385 enum pnfs_iomode iomode;
388 stateid = lo->plh_stateid;
389 iomode = lo->plh_return_iomode;
390 send = pnfs_prepare_layoutreturn(lo);
391 spin_unlock(&inode->i_lock);
393 /* Send an async layoutreturn so we dont deadlock */
394 pnfs_send_layoutreturn(lo, stateid, iomode, false);
397 spin_unlock(&inode->i_lock);
401 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
403 struct pnfs_layout_hdr *lo;
409 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
410 atomic_read(&lseg->pls_refcount),
411 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
413 /* Handle the case where refcount != 1 */
414 if (atomic_add_unless(&lseg->pls_refcount, -1, 1))
417 lo = lseg->pls_layout;
418 inode = lo->plh_inode;
419 /* Do we need a layoutreturn? */
420 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
421 pnfs_layoutreturn_before_put_lseg(lseg, lo, inode);
423 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
424 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
425 spin_unlock(&inode->i_lock);
428 pnfs_get_layout_hdr(lo);
429 pnfs_layout_remove_lseg(lo, lseg);
430 spin_unlock(&inode->i_lock);
431 pnfs_free_lseg(lseg);
432 pnfs_put_layout_hdr(lo);
435 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
437 static void pnfs_free_lseg_async_work(struct work_struct *work)
439 struct pnfs_layout_segment *lseg;
440 struct pnfs_layout_hdr *lo;
442 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
443 lo = lseg->pls_layout;
445 pnfs_free_lseg(lseg);
446 pnfs_put_layout_hdr(lo);
449 static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
451 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
452 schedule_work(&lseg->pls_work);
456 pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
461 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
463 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
464 atomic_read(&lseg->pls_refcount),
465 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
466 if (atomic_dec_and_test(&lseg->pls_refcount)) {
467 struct pnfs_layout_hdr *lo = lseg->pls_layout;
468 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
470 pnfs_get_layout_hdr(lo);
471 pnfs_layout_remove_lseg(lo, lseg);
472 pnfs_free_lseg_async(lseg);
475 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
478 end_offset(u64 start, u64 len)
483 return end >= start ? end : NFS4_MAX_UINT64;
487 * is l2 fully contained in l1?
489 * [----------------------------------)
494 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
495 const struct pnfs_layout_range *l2)
497 u64 start1 = l1->offset;
498 u64 end1 = end_offset(start1, l1->length);
499 u64 start2 = l2->offset;
500 u64 end2 = end_offset(start2, l2->length);
502 return (start1 <= start2) && (end1 >= end2);
506 * is l1 and l2 intersecting?
508 * [----------------------------------)
513 pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
514 const struct pnfs_layout_range *l2)
516 u64 start1 = l1->offset;
517 u64 end1 = end_offset(start1, l1->length);
518 u64 start2 = l2->offset;
519 u64 end2 = end_offset(start2, l2->length);
521 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
522 (end2 == NFS4_MAX_UINT64 || end2 > start1);
526 should_free_lseg(const struct pnfs_layout_range *lseg_range,
527 const struct pnfs_layout_range *recall_range)
529 return (recall_range->iomode == IOMODE_ANY ||
530 lseg_range->iomode == recall_range->iomode) &&
531 pnfs_lseg_range_intersecting(lseg_range, recall_range);
534 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
535 struct list_head *tmp_list)
537 if (!atomic_dec_and_test(&lseg->pls_refcount))
539 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
540 list_add(&lseg->pls_list, tmp_list);
544 /* Returns 1 if lseg is removed from list, 0 otherwise */
545 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
546 struct list_head *tmp_list)
550 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
551 /* Remove the reference keeping the lseg in the
552 * list. It will now be removed when all
553 * outstanding io is finished.
555 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
556 atomic_read(&lseg->pls_refcount));
557 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
563 /* Returns count of number of matching invalid lsegs remaining in list
567 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
568 struct list_head *tmp_list,
569 struct pnfs_layout_range *recall_range)
571 struct pnfs_layout_segment *lseg, *next;
572 int invalid = 0, removed = 0;
574 dprintk("%s:Begin lo %p\n", __func__, lo);
576 if (list_empty(&lo->plh_segs))
578 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
580 should_free_lseg(&lseg->pls_range, recall_range)) {
581 dprintk("%s: freeing lseg %p iomode %d "
582 "offset %llu length %llu\n", __func__,
583 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
584 lseg->pls_range.length);
586 removed += mark_lseg_invalid(lseg, tmp_list);
588 dprintk("%s:Return %i\n", __func__, invalid - removed);
589 return invalid - removed;
592 /* note free_me must contain lsegs from a single layout_hdr */
594 pnfs_free_lseg_list(struct list_head *free_me)
596 struct pnfs_layout_segment *lseg, *tmp;
598 if (list_empty(free_me))
601 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
602 list_del(&lseg->pls_list);
603 pnfs_free_lseg(lseg);
608 pnfs_destroy_layout(struct nfs_inode *nfsi)
610 struct pnfs_layout_hdr *lo;
613 spin_lock(&nfsi->vfs_inode.i_lock);
616 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
617 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
618 pnfs_get_layout_hdr(lo);
619 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
620 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
621 pnfs_clear_retry_layoutget(lo);
622 spin_unlock(&nfsi->vfs_inode.i_lock);
623 pnfs_free_lseg_list(&tmp_list);
624 pnfs_put_layout_hdr(lo);
626 spin_unlock(&nfsi->vfs_inode.i_lock);
628 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
631 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
632 struct list_head *layout_list)
634 struct pnfs_layout_hdr *lo;
637 spin_lock(&inode->i_lock);
638 lo = NFS_I(inode)->layout;
639 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
640 pnfs_get_layout_hdr(lo);
641 list_add(&lo->plh_bulk_destroy, layout_list);
644 spin_unlock(&inode->i_lock);
648 /* Caller must hold rcu_read_lock and clp->cl_lock */
650 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
651 struct nfs_server *server,
652 struct list_head *layout_list)
654 struct pnfs_layout_hdr *lo, *next;
657 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
658 inode = igrab(lo->plh_inode);
661 list_del_init(&lo->plh_layouts);
662 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
665 spin_unlock(&clp->cl_lock);
667 spin_lock(&clp->cl_lock);
675 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
678 struct pnfs_layout_hdr *lo;
680 struct pnfs_layout_range range = {
681 .iomode = IOMODE_ANY,
683 .length = NFS4_MAX_UINT64,
685 LIST_HEAD(lseg_list);
688 while (!list_empty(layout_list)) {
689 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
691 dprintk("%s freeing layout for inode %lu\n", __func__,
692 lo->plh_inode->i_ino);
693 inode = lo->plh_inode;
695 pnfs_layoutcommit_inode(inode, false);
697 spin_lock(&inode->i_lock);
698 list_del_init(&lo->plh_bulk_destroy);
699 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
701 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
702 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
704 spin_unlock(&inode->i_lock);
705 pnfs_free_lseg_list(&lseg_list);
706 pnfs_put_layout_hdr(lo);
713 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
714 struct nfs_fsid *fsid,
717 struct nfs_server *server;
718 LIST_HEAD(layout_list);
720 spin_lock(&clp->cl_lock);
723 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
724 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
726 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
732 spin_unlock(&clp->cl_lock);
734 if (list_empty(&layout_list))
736 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
740 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
743 struct nfs_server *server;
744 LIST_HEAD(layout_list);
746 spin_lock(&clp->cl_lock);
749 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
750 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
756 spin_unlock(&clp->cl_lock);
758 if (list_empty(&layout_list))
760 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
764 * Called by the state manger to remove all layouts established under an
768 pnfs_destroy_all_layouts(struct nfs_client *clp)
770 nfs4_deviceid_mark_client_invalid(clp);
771 nfs4_deviceid_purge_client(clp);
773 pnfs_destroy_layouts_byclid(clp, false);
777 * Compare 2 layout stateid sequence ids, to see which is newer,
778 * taking into account wraparound issues.
780 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
782 return (s32)(s1 - s2) > 0;
785 /* update lo->plh_stateid with new if is more recent */
787 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
790 u32 oldseq, newseq, new_barrier;
791 int empty = list_empty(&lo->plh_segs);
793 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
794 newseq = be32_to_cpu(new->seqid);
795 if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
796 nfs4_stateid_copy(&lo->plh_stateid, new);
797 if (update_barrier) {
798 new_barrier = be32_to_cpu(new->seqid);
800 /* Because of wraparound, we want to keep the barrier
801 * "close" to the current seqids.
803 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
805 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
806 lo->plh_barrier = new_barrier;
811 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
812 const nfs4_stateid *stateid)
814 u32 seqid = be32_to_cpu(stateid->seqid);
816 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
819 /* lget is set to 1 if called from inside send_layoutget call chain */
821 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
823 return lo->plh_block_lgets ||
824 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
828 pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
829 struct pnfs_layout_range *range,
830 struct nfs4_state *open_state)
834 dprintk("--> %s\n", __func__);
835 spin_lock(&lo->plh_inode->i_lock);
836 if (pnfs_layoutgets_blocked(lo)) {
838 } else if (!nfs4_valid_open_stateid(open_state)) {
840 } else if (list_empty(&lo->plh_segs) ||
841 test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
845 seq = read_seqbegin(&open_state->seqlock);
846 nfs4_stateid_copy(dst, &open_state->stateid);
847 } while (read_seqretry(&open_state->seqlock, seq));
849 nfs4_stateid_copy(dst, &lo->plh_stateid);
850 spin_unlock(&lo->plh_inode->i_lock);
851 dprintk("<-- %s\n", __func__);
856 * Get layout from server.
857 * for now, assume that whole file layouts are requested.
859 * arg->length: all ones
861 static struct pnfs_layout_segment *
862 send_layoutget(struct pnfs_layout_hdr *lo,
863 struct nfs_open_context *ctx,
864 struct pnfs_layout_range *range,
867 struct inode *ino = lo->plh_inode;
868 struct nfs_server *server = NFS_SERVER(ino);
869 struct nfs4_layoutget *lgp;
870 struct pnfs_layout_segment *lseg;
873 dprintk("--> %s\n", __func__);
876 * Synchronously retrieve layout information from server and
877 * store in lseg. If we race with a concurrent seqid morphing
878 * op, then re-send the LAYOUTGET.
881 lgp = kzalloc(sizeof(*lgp), gfp_flags);
885 i_size = i_size_read(ino);
887 lgp->args.minlength = PAGE_CACHE_SIZE;
888 if (lgp->args.minlength > range->length)
889 lgp->args.minlength = range->length;
890 if (range->iomode == IOMODE_READ) {
891 if (range->offset >= i_size)
892 lgp->args.minlength = 0;
893 else if (i_size - range->offset < lgp->args.minlength)
894 lgp->args.minlength = i_size - range->offset;
896 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
897 lgp->args.range = *range;
898 lgp->args.type = server->pnfs_curr_ld->id;
899 lgp->args.inode = ino;
900 lgp->args.ctx = get_nfs_open_context(ctx);
901 lgp->gfp_flags = gfp_flags;
902 lgp->cred = lo->plh_lc_cred;
904 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
905 } while (lseg == ERR_PTR(-EAGAIN));
908 switch (PTR_ERR(lseg)) {
913 /* remember that LAYOUTGET failed and suspend trying */
914 pnfs_layout_io_set_failed(lo, range->iomode);
918 pnfs_layout_clear_fail_bit(lo,
919 pnfs_iomode_to_fail_bit(range->iomode));
924 static void pnfs_clear_layoutcommit(struct inode *inode,
925 struct list_head *head)
927 struct nfs_inode *nfsi = NFS_I(inode);
928 struct pnfs_layout_segment *lseg, *tmp;
930 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
932 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
933 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
935 pnfs_lseg_dec_and_remove_zero(lseg, head);
939 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
941 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
942 smp_mb__after_atomic();
943 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
944 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
948 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
949 enum pnfs_iomode iomode, bool sync)
951 struct inode *ino = lo->plh_inode;
952 struct nfs4_layoutreturn *lrp;
955 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
956 if (unlikely(lrp == NULL)) {
958 spin_lock(&ino->i_lock);
959 pnfs_clear_layoutreturn_waitbit(lo);
960 spin_unlock(&ino->i_lock);
961 pnfs_put_layout_hdr(lo);
965 lrp->args.stateid = stateid;
966 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
967 lrp->args.inode = ino;
968 lrp->args.range.iomode = iomode;
969 lrp->args.range.offset = 0;
970 lrp->args.range.length = NFS4_MAX_UINT64;
971 lrp->args.layout = lo;
972 lrp->clp = NFS_SERVER(ino)->nfs_client;
973 lrp->cred = lo->plh_lc_cred;
975 status = nfs4_proc_layoutreturn(lrp, sync);
977 dprintk("<-- %s status: %d\n", __func__, status);
982 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
983 * when the layout segment list is empty.
985 * Note that a pnfs_layout_hdr can exist with an empty layout segment
986 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
987 * deviceid is marked invalid.
990 _pnfs_return_layout(struct inode *ino)
992 struct pnfs_layout_hdr *lo = NULL;
993 struct nfs_inode *nfsi = NFS_I(ino);
995 nfs4_stateid stateid;
996 int status = 0, empty;
999 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1001 spin_lock(&ino->i_lock);
1004 spin_unlock(&ino->i_lock);
1005 dprintk("NFS: %s no layout to return\n", __func__);
1008 stateid = nfsi->layout->plh_stateid;
1009 /* Reference matched in nfs4_layoutreturn_release */
1010 pnfs_get_layout_hdr(lo);
1011 empty = list_empty(&lo->plh_segs);
1012 pnfs_clear_layoutcommit(ino, &tmp_list);
1013 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
1015 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1016 struct pnfs_layout_range range = {
1017 .iomode = IOMODE_ANY,
1019 .length = NFS4_MAX_UINT64,
1021 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1024 /* Don't send a LAYOUTRETURN if list was initially empty */
1026 spin_unlock(&ino->i_lock);
1027 dprintk("NFS: %s no layout segments to return\n", __func__);
1028 goto out_put_layout_hdr;
1031 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1032 send = pnfs_prepare_layoutreturn(lo);
1033 spin_unlock(&ino->i_lock);
1034 pnfs_free_lseg_list(&tmp_list);
1036 status = pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
1038 pnfs_put_layout_hdr(lo);
1040 dprintk("<-- %s status: %d\n", __func__, status);
1043 EXPORT_SYMBOL_GPL(_pnfs_return_layout);
1046 pnfs_commit_and_return_layout(struct inode *inode)
1048 struct pnfs_layout_hdr *lo;
1051 spin_lock(&inode->i_lock);
1052 lo = NFS_I(inode)->layout;
1054 spin_unlock(&inode->i_lock);
1057 pnfs_get_layout_hdr(lo);
1058 /* Block new layoutgets and read/write to ds */
1059 lo->plh_block_lgets++;
1060 spin_unlock(&inode->i_lock);
1061 filemap_fdatawait(inode->i_mapping);
1062 ret = pnfs_layoutcommit_inode(inode, true);
1064 ret = _pnfs_return_layout(inode);
1065 spin_lock(&inode->i_lock);
1066 lo->plh_block_lgets--;
1067 spin_unlock(&inode->i_lock);
1068 pnfs_put_layout_hdr(lo);
1072 bool pnfs_roc(struct inode *ino)
1074 struct nfs_inode *nfsi = NFS_I(ino);
1075 struct nfs_open_context *ctx;
1076 struct nfs4_state *state;
1077 struct pnfs_layout_hdr *lo;
1078 struct pnfs_layout_segment *lseg, *tmp;
1079 nfs4_stateid stateid;
1080 LIST_HEAD(tmp_list);
1081 bool found = false, layoutreturn = false, roc = false;
1083 spin_lock(&ino->i_lock);
1085 if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1088 /* no roc if we hold a delegation */
1089 if (nfs4_check_delegation(ino, FMODE_READ))
1092 list_for_each_entry(ctx, &nfsi->open_files, list) {
1094 /* Don't return layout if there is open file state */
1095 if (state != NULL && state->state != 0)
1099 stateid = lo->plh_stateid;
1100 /* always send layoutreturn if being marked so */
1101 if (test_and_clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1103 layoutreturn = pnfs_prepare_layoutreturn(lo);
1105 pnfs_clear_retry_layoutget(lo);
1106 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1107 /* If we are sending layoutreturn, invalidate all valid lsegs */
1108 if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
1109 mark_lseg_invalid(lseg, &tmp_list);
1112 /* ROC in two conditions:
1113 * 1. there are ROC lsegs
1114 * 2. we don't send layoutreturn
1116 if (found && !layoutreturn) {
1117 /* lo ref dropped in pnfs_roc_release() */
1118 pnfs_get_layout_hdr(lo);
1123 spin_unlock(&ino->i_lock);
1124 pnfs_free_lseg_list(&tmp_list);
1125 pnfs_layoutcommit_inode(ino, true);
1127 pnfs_send_layoutreturn(lo, stateid, IOMODE_ANY, true);
1131 void pnfs_roc_release(struct inode *ino)
1133 struct pnfs_layout_hdr *lo;
1135 spin_lock(&ino->i_lock);
1136 lo = NFS_I(ino)->layout;
1137 pnfs_clear_layoutreturn_waitbit(lo);
1138 if (atomic_dec_and_test(&lo->plh_refcount)) {
1139 pnfs_detach_layout_hdr(lo);
1140 spin_unlock(&ino->i_lock);
1141 pnfs_free_layout_hdr(lo);
1143 spin_unlock(&ino->i_lock);
1146 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
1148 struct pnfs_layout_hdr *lo;
1150 spin_lock(&ino->i_lock);
1151 lo = NFS_I(ino)->layout;
1152 pnfs_mark_layout_returned_if_empty(lo);
1153 if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
1154 lo->plh_barrier = barrier;
1155 spin_unlock(&ino->i_lock);
1156 trace_nfs4_layoutreturn_on_close(ino, 0);
1159 void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
1161 struct nfs_inode *nfsi = NFS_I(ino);
1162 struct pnfs_layout_hdr *lo;
1165 spin_lock(&ino->i_lock);
1167 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
1169 /* Since close does not return a layout stateid for use as
1170 * a barrier, we choose the worst-case barrier.
1172 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1173 spin_unlock(&ino->i_lock);
1176 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1178 struct nfs_inode *nfsi = NFS_I(ino);
1179 struct pnfs_layout_hdr *lo;
1182 /* we might not have grabbed lo reference. so need to check under
1184 spin_lock(&ino->i_lock);
1186 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
1188 spin_unlock(&ino->i_lock);
1191 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1197 * Compare two layout segments for sorting into layout cache.
1198 * We want to preferentially return RW over RO layouts, so ensure those
1202 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1203 const struct pnfs_layout_range *l2)
1207 /* high offset > low offset */
1208 d = l1->offset - l2->offset;
1212 /* short length > long length */
1213 d = l2->length - l1->length;
1217 /* read > read/write */
1218 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1222 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1223 const struct pnfs_layout_range *l2)
1225 return pnfs_lseg_range_cmp(l1, l2) > 0;
1229 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1230 struct pnfs_layout_segment *old)
1236 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1237 struct pnfs_layout_segment *lseg,
1238 bool (*is_after)(const struct pnfs_layout_range *,
1239 const struct pnfs_layout_range *),
1240 bool (*do_merge)(struct pnfs_layout_segment *,
1241 struct pnfs_layout_segment *),
1242 struct list_head *free_me)
1244 struct pnfs_layout_segment *lp, *tmp;
1246 dprintk("%s:Begin\n", __func__);
1248 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1249 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1251 if (do_merge(lseg, lp)) {
1252 mark_lseg_invalid(lp, free_me);
1255 if (is_after(&lseg->pls_range, &lp->pls_range))
1257 list_add_tail(&lseg->pls_list, &lp->pls_list);
1258 dprintk("%s: inserted lseg %p "
1259 "iomode %d offset %llu length %llu before "
1260 "lp %p iomode %d offset %llu length %llu\n",
1261 __func__, lseg, lseg->pls_range.iomode,
1262 lseg->pls_range.offset, lseg->pls_range.length,
1263 lp, lp->pls_range.iomode, lp->pls_range.offset,
1264 lp->pls_range.length);
1267 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1268 dprintk("%s: inserted lseg %p "
1269 "iomode %d offset %llu length %llu at tail\n",
1270 __func__, lseg, lseg->pls_range.iomode,
1271 lseg->pls_range.offset, lseg->pls_range.length);
1273 pnfs_get_layout_hdr(lo);
1275 dprintk("%s:Return\n", __func__);
1277 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1280 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1281 struct pnfs_layout_segment *lseg,
1282 struct list_head *free_me)
1284 struct inode *inode = lo->plh_inode;
1285 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1287 if (ld->add_lseg != NULL)
1288 ld->add_lseg(lo, lseg, free_me);
1290 pnfs_generic_layout_insert_lseg(lo, lseg,
1291 pnfs_lseg_range_is_after,
1296 static struct pnfs_layout_hdr *
1297 alloc_init_layout_hdr(struct inode *ino,
1298 struct nfs_open_context *ctx,
1301 struct pnfs_layout_hdr *lo;
1303 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1306 atomic_set(&lo->plh_refcount, 1);
1307 INIT_LIST_HEAD(&lo->plh_layouts);
1308 INIT_LIST_HEAD(&lo->plh_segs);
1309 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1310 lo->plh_inode = ino;
1311 lo->plh_lc_cred = get_rpccred(ctx->cred);
1315 static struct pnfs_layout_hdr *
1316 pnfs_find_alloc_layout(struct inode *ino,
1317 struct nfs_open_context *ctx,
1320 struct nfs_inode *nfsi = NFS_I(ino);
1321 struct pnfs_layout_hdr *new = NULL;
1323 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1325 if (nfsi->layout != NULL)
1327 spin_unlock(&ino->i_lock);
1328 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1329 spin_lock(&ino->i_lock);
1331 if (likely(nfsi->layout == NULL)) { /* Won the race? */
1334 } else if (new != NULL)
1335 pnfs_free_layout_hdr(new);
1337 pnfs_get_layout_hdr(nfsi->layout);
1338 return nfsi->layout;
1342 * iomode matching rules:
1353 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1354 const struct pnfs_layout_range *range)
1356 struct pnfs_layout_range range1;
1358 if ((range->iomode == IOMODE_RW &&
1359 ls_range->iomode != IOMODE_RW) ||
1360 !pnfs_lseg_range_intersecting(ls_range, range))
1363 /* range1 covers only the first byte in the range */
1366 return pnfs_lseg_range_contained(ls_range, &range1);
1370 * lookup range in layout
1372 static struct pnfs_layout_segment *
1373 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1374 struct pnfs_layout_range *range)
1376 struct pnfs_layout_segment *lseg, *ret = NULL;
1378 dprintk("%s:Begin\n", __func__);
1380 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1381 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1382 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1383 pnfs_lseg_range_match(&lseg->pls_range, range)) {
1384 ret = pnfs_get_lseg(lseg);
1389 dprintk("%s:Return lseg %p ref %d\n",
1390 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1395 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1396 * to the MDS or over pNFS
1398 * The nfs_inode read_io and write_io fields are cumulative counters reset
1399 * when there are no layout segments. Note that in pnfs_update_layout iomode
1400 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1403 * A return of true means use MDS I/O.
1406 * If a file's size is smaller than the file size threshold, data accesses
1407 * SHOULD be sent to the metadata server. If an I/O request has a length that
1408 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1409 * server. If both file size and I/O size are provided, the client SHOULD
1410 * reach or exceed both thresholds before sending its read or write
1411 * requests to the data server.
1413 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1414 struct inode *ino, int iomode)
1416 struct nfs4_threshold *t = ctx->mdsthreshold;
1417 struct nfs_inode *nfsi = NFS_I(ino);
1418 loff_t fsize = i_size_read(ino);
1419 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1424 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1425 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1429 if (t->bm & THRESHOLD_RD) {
1430 dprintk("%s fsize %llu\n", __func__, fsize);
1432 if (fsize < t->rd_sz)
1435 if (t->bm & THRESHOLD_RD_IO) {
1436 dprintk("%s nfsi->read_io %llu\n", __func__,
1439 if (nfsi->read_io < t->rd_io_sz)
1444 if (t->bm & THRESHOLD_WR) {
1445 dprintk("%s fsize %llu\n", __func__, fsize);
1447 if (fsize < t->wr_sz)
1450 if (t->bm & THRESHOLD_WR_IO) {
1451 dprintk("%s nfsi->write_io %llu\n", __func__,
1454 if (nfsi->write_io < t->wr_io_sz)
1459 if (size_set && io_set) {
1462 } else if (size || io)
1465 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1469 /* stop waiting if someone clears NFS_LAYOUT_RETRY_LAYOUTGET bit. */
1470 static int pnfs_layoutget_retry_bit_wait(struct wait_bit_key *key, int mode)
1472 if (!test_bit(NFS_LAYOUT_RETRY_LAYOUTGET, key->flags))
1474 return nfs_wait_bit_killable(key, mode);
1477 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1479 if (!pnfs_should_retry_layoutget(lo))
1482 * send layoutcommit as it can hold up layoutreturn due to lseg
1485 pnfs_layoutcommit_inode(lo->plh_inode, false);
1486 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1487 pnfs_layoutget_retry_bit_wait,
1488 TASK_UNINTERRUPTIBLE);
1491 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1493 unsigned long *bitlock = &lo->plh_flags;
1495 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1496 smp_mb__after_atomic();
1497 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1501 * Layout segment is retreived from the server if not cached.
1502 * The appropriate layout segment is referenced and returned to the caller.
1504 struct pnfs_layout_segment *
1505 pnfs_update_layout(struct inode *ino,
1506 struct nfs_open_context *ctx,
1509 enum pnfs_iomode iomode,
1512 struct pnfs_layout_range arg = {
1518 struct nfs_server *server = NFS_SERVER(ino);
1519 struct nfs_client *clp = server->nfs_client;
1520 struct pnfs_layout_hdr *lo;
1521 struct pnfs_layout_segment *lseg = NULL;
1524 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1525 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1526 PNFS_UPDATE_LAYOUT_NO_PNFS);
1530 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
1531 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1532 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
1536 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1537 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1538 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1544 spin_lock(&ino->i_lock);
1545 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1547 spin_unlock(&ino->i_lock);
1548 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1549 PNFS_UPDATE_LAYOUT_NOMEM);
1553 /* Do we even need to bother with this? */
1554 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1555 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1556 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1557 dprintk("%s matches recall, use MDS\n", __func__);
1561 /* if LAYOUTGET already failed once we don't try again */
1562 if (pnfs_layout_io_test_failed(lo, iomode) &&
1563 !pnfs_should_retry_layoutget(lo)) {
1564 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1565 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1569 first = list_empty(&lo->plh_segs);
1571 /* The first layoutget for the file. Need to serialize per
1572 * RFC 5661 Errata 3208.
1574 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1576 spin_unlock(&ino->i_lock);
1577 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1578 TASK_UNINTERRUPTIBLE);
1579 pnfs_put_layout_hdr(lo);
1583 /* Check to see if the layout for the given range
1586 lseg = pnfs_find_lseg(lo, &arg);
1588 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1589 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1595 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1596 * for LAYOUTRETURN even if first is true.
1598 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1599 spin_unlock(&ino->i_lock);
1600 dprintk("%s wait for layoutreturn\n", __func__);
1601 if (pnfs_prepare_to_retry_layoutget(lo)) {
1603 pnfs_clear_first_layoutget(lo);
1604 pnfs_put_layout_hdr(lo);
1605 dprintk("%s retrying\n", __func__);
1608 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1609 PNFS_UPDATE_LAYOUT_RETURN);
1610 goto out_put_layout_hdr;
1613 if (pnfs_layoutgets_blocked(lo)) {
1614 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1615 PNFS_UPDATE_LAYOUT_BLOCKED);
1618 atomic_inc(&lo->plh_outstanding);
1619 spin_unlock(&ino->i_lock);
1621 if (list_empty(&lo->plh_layouts)) {
1622 /* The lo must be on the clp list if there is any
1623 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1625 spin_lock(&clp->cl_lock);
1626 if (list_empty(&lo->plh_layouts))
1627 list_add_tail(&lo->plh_layouts, &server->layouts);
1628 spin_unlock(&clp->cl_lock);
1631 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1633 arg.offset -= pg_offset;
1634 arg.length += pg_offset;
1636 if (arg.length != NFS4_MAX_UINT64)
1637 arg.length = PAGE_CACHE_ALIGN(arg.length);
1639 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1640 pnfs_clear_retry_layoutget(lo);
1641 atomic_dec(&lo->plh_outstanding);
1642 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1643 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1646 pnfs_clear_first_layoutget(lo);
1647 pnfs_put_layout_hdr(lo);
1649 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1650 "(%s, offset: %llu, length: %llu)\n",
1651 __func__, ino->i_sb->s_id,
1652 (unsigned long long)NFS_FILEID(ino),
1653 lseg == NULL ? "not found" : "found",
1654 iomode==IOMODE_RW ? "read/write" : "read-only",
1655 (unsigned long long)pos,
1656 (unsigned long long)count);
1659 spin_unlock(&ino->i_lock);
1660 goto out_put_layout_hdr;
1662 EXPORT_SYMBOL_GPL(pnfs_update_layout);
1665 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1667 switch (range->iomode) {
1674 if (range->offset == NFS4_MAX_UINT64)
1676 if (range->length == 0)
1678 if (range->length != NFS4_MAX_UINT64 &&
1679 range->length > NFS4_MAX_UINT64 - range->offset)
1684 struct pnfs_layout_segment *
1685 pnfs_layout_process(struct nfs4_layoutget *lgp)
1687 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1688 struct nfs4_layoutget_res *res = &lgp->res;
1689 struct pnfs_layout_segment *lseg;
1690 struct inode *ino = lo->plh_inode;
1692 int status = -EINVAL;
1694 if (!pnfs_sanity_check_layout_range(&res->range))
1697 /* Inject layout blob into I/O device driver */
1698 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1699 if (!lseg || IS_ERR(lseg)) {
1703 status = PTR_ERR(lseg);
1704 dprintk("%s: Could not allocate layout: error %d\n",
1709 init_lseg(lo, lseg);
1710 lseg->pls_range = res->range;
1712 spin_lock(&ino->i_lock);
1713 if (pnfs_layoutgets_blocked(lo)) {
1714 dprintk("%s forget reply due to state\n", __func__);
1715 goto out_forget_reply;
1718 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
1719 /* existing state ID, make sure the sequence number matches. */
1720 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1721 dprintk("%s forget reply due to sequence\n", __func__);
1723 goto out_forget_reply;
1725 pnfs_set_layout_stateid(lo, &res->stateid, false);
1728 * We got an entirely new state ID. Mark all segments for the
1729 * inode invalid, and don't bother validating the stateid
1732 pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
1734 nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
1735 lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
1738 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1740 pnfs_get_lseg(lseg);
1741 pnfs_layout_insert_lseg(lo, lseg, &free_me);
1743 if (res->return_on_close)
1744 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1746 spin_unlock(&ino->i_lock);
1747 pnfs_free_lseg_list(&free_me);
1750 return ERR_PTR(status);
1753 spin_unlock(&ino->i_lock);
1754 lseg->pls_layout = lo;
1755 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1760 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1761 struct list_head *tmp_list,
1762 struct pnfs_layout_range *return_range)
1764 struct pnfs_layout_segment *lseg, *next;
1766 dprintk("%s:Begin lo %p\n", __func__, lo);
1768 if (list_empty(&lo->plh_segs))
1771 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
1772 if (should_free_lseg(&lseg->pls_range, return_range)) {
1773 dprintk("%s: marking lseg %p iomode %d "
1774 "offset %llu length %llu\n", __func__,
1775 lseg, lseg->pls_range.iomode,
1776 lseg->pls_range.offset,
1777 lseg->pls_range.length);
1778 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1779 mark_lseg_invalid(lseg, tmp_list);
1780 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1785 void pnfs_error_mark_layout_for_return(struct inode *inode,
1786 struct pnfs_layout_segment *lseg)
1788 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
1789 int iomode = pnfs_iomode_to_fail_bit(lseg->pls_range.iomode);
1790 struct pnfs_layout_range range = {
1791 .iomode = lseg->pls_range.iomode,
1793 .length = NFS4_MAX_UINT64,
1797 spin_lock(&inode->i_lock);
1798 /* set failure bit so that pnfs path will be retried later */
1799 pnfs_layout_set_fail_bit(lo, iomode);
1800 if (lo->plh_return_iomode == 0)
1801 lo->plh_return_iomode = range.iomode;
1802 else if (lo->plh_return_iomode != range.iomode)
1803 lo->plh_return_iomode = IOMODE_ANY;
1805 * mark all matching lsegs so that we are sure to have no live
1806 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1809 pnfs_mark_matching_lsegs_return(lo, &free_me, &range);
1810 spin_unlock(&inode->i_lock);
1811 pnfs_free_lseg_list(&free_me);
1813 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
1816 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1818 u64 rd_size = req->wb_bytes;
1820 if (pgio->pg_lseg == NULL) {
1821 if (pgio->pg_dreq == NULL)
1822 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1824 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1826 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1833 /* If no lseg, fall back to read through mds */
1834 if (pgio->pg_lseg == NULL)
1835 nfs_pageio_reset_read_mds(pgio);
1838 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1841 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1842 struct nfs_page *req, u64 wb_size)
1844 if (pgio->pg_lseg == NULL)
1845 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1851 /* If no lseg, fall back to write through mds */
1852 if (pgio->pg_lseg == NULL)
1853 nfs_pageio_reset_write_mds(pgio);
1855 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1858 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
1860 if (desc->pg_lseg) {
1861 pnfs_put_lseg(desc->pg_lseg);
1862 desc->pg_lseg = NULL;
1865 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
1868 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1869 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1872 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
1873 struct nfs_page *prev, struct nfs_page *req)
1876 u64 seg_end, req_start, seg_left;
1878 size = nfs_generic_pg_test(pgio, prev, req);
1883 * 'size' contains the number of bytes left in the current page (up
1884 * to the original size asked for in @req->wb_bytes).
1886 * Calculate how many bytes are left in the layout segment
1887 * and if there are less bytes than 'size', return that instead.
1889 * Please also note that 'end_offset' is actually the offset of the
1890 * first byte that lies outside the pnfs_layout_range. FIXME?
1893 if (pgio->pg_lseg) {
1894 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1895 pgio->pg_lseg->pls_range.length);
1896 req_start = req_offset(req);
1897 WARN_ON_ONCE(req_start >= seg_end);
1898 /* start of request is past the last byte of this segment */
1899 if (req_start >= seg_end) {
1900 /* reference the new lseg */
1901 if (pgio->pg_ops->pg_cleanup)
1902 pgio->pg_ops->pg_cleanup(pgio);
1903 if (pgio->pg_ops->pg_init)
1904 pgio->pg_ops->pg_init(pgio, req);
1908 /* adjust 'size' iff there are fewer bytes left in the
1909 * segment than what nfs_generic_pg_test returned */
1910 seg_left = seg_end - req_start;
1911 if (seg_left < size)
1912 size = (unsigned int)seg_left;
1917 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1919 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1921 struct nfs_pageio_descriptor pgio;
1923 /* Resend all requests through the MDS */
1924 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
1925 hdr->completion_ops);
1926 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
1927 return nfs_pageio_resend(&pgio, hdr);
1929 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1931 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1934 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1935 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1936 PNFS_LAYOUTRET_ON_ERROR) {
1937 pnfs_return_layout(hdr->inode);
1939 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1940 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1944 * Called by non rpc-based layout drivers
1946 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1948 if (likely(!hdr->pnfs_error)) {
1949 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1950 hdr->mds_offset + hdr->res.count);
1951 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1953 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1954 if (unlikely(hdr->pnfs_error))
1955 pnfs_ld_handle_write_error(hdr);
1956 hdr->mds_ops->rpc_release(hdr);
1958 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1961 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1962 struct nfs_pgio_header *hdr)
1964 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1966 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1967 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1968 nfs_pageio_reset_write_mds(desc);
1969 mirror->pg_recoalesce = 1;
1971 nfs_pgio_data_destroy(hdr);
1975 static enum pnfs_try_status
1976 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1977 const struct rpc_call_ops *call_ops,
1978 struct pnfs_layout_segment *lseg,
1981 struct inode *inode = hdr->inode;
1982 enum pnfs_try_status trypnfs;
1983 struct nfs_server *nfss = NFS_SERVER(inode);
1985 hdr->mds_ops = call_ops;
1987 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1988 inode->i_ino, hdr->args.count, hdr->args.offset, how);
1989 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1990 if (trypnfs != PNFS_NOT_ATTEMPTED)
1991 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1992 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
1997 pnfs_do_write(struct nfs_pageio_descriptor *desc,
1998 struct nfs_pgio_header *hdr, int how)
2000 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2001 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2002 enum pnfs_try_status trypnfs;
2004 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2005 if (trypnfs == PNFS_NOT_ATTEMPTED)
2006 pnfs_write_through_mds(desc, hdr);
2009 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2011 pnfs_put_lseg(hdr->lseg);
2012 nfs_pgio_header_free(hdr);
2016 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2018 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2020 struct nfs_pgio_header *hdr;
2023 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2025 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
2028 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2030 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2031 ret = nfs_generic_pgio(desc, hdr);
2033 pnfs_do_write(desc, hdr, desc->pg_ioflags);
2037 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2039 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2041 struct nfs_pageio_descriptor pgio;
2043 /* Resend all requests through the MDS */
2044 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2045 return nfs_pageio_resend(&pgio, hdr);
2047 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2049 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2051 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2052 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2053 PNFS_LAYOUTRET_ON_ERROR) {
2054 pnfs_return_layout(hdr->inode);
2056 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2057 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2061 * Called by non rpc-based layout drivers
2063 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2065 if (likely(!hdr->pnfs_error)) {
2066 __nfs4_read_done_cb(hdr);
2067 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2069 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2070 if (unlikely(hdr->pnfs_error))
2071 pnfs_ld_handle_read_error(hdr);
2072 hdr->mds_ops->rpc_release(hdr);
2074 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2077 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2078 struct nfs_pgio_header *hdr)
2080 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2082 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2083 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2084 nfs_pageio_reset_read_mds(desc);
2085 mirror->pg_recoalesce = 1;
2087 nfs_pgio_data_destroy(hdr);
2092 * Call the appropriate parallel I/O subsystem read function.
2094 static enum pnfs_try_status
2095 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2096 const struct rpc_call_ops *call_ops,
2097 struct pnfs_layout_segment *lseg)
2099 struct inode *inode = hdr->inode;
2100 struct nfs_server *nfss = NFS_SERVER(inode);
2101 enum pnfs_try_status trypnfs;
2103 hdr->mds_ops = call_ops;
2105 dprintk("%s: Reading ino:%lu %u@%llu\n",
2106 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2108 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2109 if (trypnfs != PNFS_NOT_ATTEMPTED)
2110 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2111 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2115 /* Resend all requests through pnfs. */
2116 int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2118 struct nfs_pageio_descriptor pgio;
2120 nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops);
2121 return nfs_pageio_resend(&pgio, hdr);
2123 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2126 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2128 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2129 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2130 enum pnfs_try_status trypnfs;
2133 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2134 if (trypnfs == PNFS_TRY_AGAIN)
2135 err = pnfs_read_resend_pnfs(hdr);
2136 if (trypnfs == PNFS_NOT_ATTEMPTED || err)
2137 pnfs_read_through_mds(desc, hdr);
2140 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2142 pnfs_put_lseg(hdr->lseg);
2143 nfs_pgio_header_free(hdr);
2147 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2149 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2151 struct nfs_pgio_header *hdr;
2154 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2156 desc->pg_completion_ops->error_cleanup(&mirror->pg_list);
2159 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2160 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2161 ret = nfs_generic_pgio(desc, hdr);
2163 pnfs_do_read(desc, hdr);
2166 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2168 static void pnfs_clear_layoutcommitting(struct inode *inode)
2170 unsigned long *bitlock = &NFS_I(inode)->flags;
2172 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2173 smp_mb__after_atomic();
2174 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2178 * There can be multiple RW segments.
2180 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2182 struct pnfs_layout_segment *lseg;
2184 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2185 if (lseg->pls_range.iomode == IOMODE_RW &&
2186 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2187 list_add(&lseg->pls_lc_list, listp);
2191 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2193 struct pnfs_layout_segment *lseg, *tmp;
2195 /* Matched by references in pnfs_set_layoutcommit */
2196 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2197 list_del_init(&lseg->pls_lc_list);
2198 pnfs_put_lseg(lseg);
2201 pnfs_clear_layoutcommitting(inode);
2204 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2206 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2208 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2211 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2214 struct nfs_inode *nfsi = NFS_I(inode);
2215 bool mark_as_dirty = false;
2217 spin_lock(&inode->i_lock);
2218 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2219 nfsi->layout->plh_lwb = end_pos;
2220 mark_as_dirty = true;
2221 dprintk("%s: Set layoutcommit for inode %lu ",
2222 __func__, inode->i_ino);
2223 } else if (end_pos > nfsi->layout->plh_lwb)
2224 nfsi->layout->plh_lwb = end_pos;
2225 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2226 /* references matched in nfs4_layoutcommit_release */
2227 pnfs_get_lseg(lseg);
2229 spin_unlock(&inode->i_lock);
2230 dprintk("%s: lseg %p end_pos %llu\n",
2231 __func__, lseg, nfsi->layout->plh_lwb);
2233 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2234 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2236 mark_inode_dirty_sync(inode);
2238 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2240 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2242 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2244 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2245 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2246 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2250 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2251 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2252 * data to disk to allow the server to recover the data if it crashes.
2253 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2254 * is off, and a COMMIT is sent to a data server, or
2255 * if WRITEs to a data server return NFS_DATA_SYNC.
2258 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2260 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2261 struct nfs4_layoutcommit_data *data;
2262 struct nfs_inode *nfsi = NFS_I(inode);
2266 if (!pnfs_layoutcommit_outstanding(inode))
2269 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2272 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2275 status = wait_on_bit_lock_action(&nfsi->flags,
2276 NFS_INO_LAYOUTCOMMITTING,
2277 nfs_wait_bit_killable,
2284 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2285 data = kzalloc(sizeof(*data), GFP_NOFS);
2287 goto clear_layoutcommitting;
2290 spin_lock(&inode->i_lock);
2291 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2294 INIT_LIST_HEAD(&data->lseg_list);
2295 pnfs_list_write_lseg(inode, &data->lseg_list);
2297 end_pos = nfsi->layout->plh_lwb;
2299 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2300 spin_unlock(&inode->i_lock);
2302 data->args.inode = inode;
2303 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
2304 nfs_fattr_init(&data->fattr);
2305 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2306 data->res.fattr = &data->fattr;
2307 data->args.lastbytewritten = end_pos - 1;
2308 data->res.server = NFS_SERVER(inode);
2310 if (ld->prepare_layoutcommit) {
2311 status = ld->prepare_layoutcommit(&data->args);
2313 put_rpccred(data->cred);
2314 spin_lock(&inode->i_lock);
2315 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2316 if (end_pos > nfsi->layout->plh_lwb)
2317 nfsi->layout->plh_lwb = end_pos;
2323 status = nfs4_proc_layoutcommit(data, sync);
2326 mark_inode_dirty_sync(inode);
2327 dprintk("<-- %s status %d\n", __func__, status);
2330 spin_unlock(&inode->i_lock);
2332 clear_layoutcommitting:
2333 pnfs_clear_layoutcommitting(inode);
2336 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2339 pnfs_generic_sync(struct inode *inode, bool datasync)
2341 return pnfs_layoutcommit_inode(inode, true);
2343 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2345 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2347 struct nfs4_threshold *thp;
2349 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2351 dprintk("%s mdsthreshold allocation failed\n", __func__);
2357 #if IS_ENABLED(CONFIG_NFS_V4_2)
2359 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2361 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2362 struct nfs_server *server = NFS_SERVER(inode);
2363 struct nfs_inode *nfsi = NFS_I(inode);
2364 struct nfs42_layoutstat_data *data;
2365 struct pnfs_layout_hdr *hdr;
2368 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2371 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2374 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2377 spin_lock(&inode->i_lock);
2378 if (!NFS_I(inode)->layout) {
2379 spin_unlock(&inode->i_lock);
2382 hdr = NFS_I(inode)->layout;
2383 pnfs_get_layout_hdr(hdr);
2384 spin_unlock(&inode->i_lock);
2386 data = kzalloc(sizeof(*data), gfp_flags);
2392 data->args.fh = NFS_FH(inode);
2393 data->args.inode = inode;
2394 nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid);
2395 status = ld->prepare_layoutstats(&data->args);
2399 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2402 dprintk("%s returns %d\n", __func__, status);
2408 pnfs_put_layout_hdr(hdr);
2409 smp_mb__before_atomic();
2410 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2411 smp_mb__after_atomic();
2414 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2417 unsigned int layoutstats_timer;
2418 module_param(layoutstats_timer, uint, 0644);
2419 EXPORT_SYMBOL_GPL(layoutstats_timer);