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 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);
349 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
351 struct pnfs_layout_hdr *lo;
357 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
358 atomic_read(&lseg->pls_refcount),
359 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
361 lo = lseg->pls_layout;
362 inode = lo->plh_inode;
364 if (atomic_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
365 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
366 spin_unlock(&inode->i_lock);
369 pnfs_get_layout_hdr(lo);
370 pnfs_layout_remove_lseg(lo, lseg);
371 spin_unlock(&inode->i_lock);
372 pnfs_free_lseg(lseg);
373 pnfs_put_layout_hdr(lo);
376 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
378 static void pnfs_free_lseg_async_work(struct work_struct *work)
380 struct pnfs_layout_segment *lseg;
381 struct pnfs_layout_hdr *lo;
383 lseg = container_of(work, struct pnfs_layout_segment, pls_work);
384 lo = lseg->pls_layout;
386 pnfs_free_lseg(lseg);
387 pnfs_put_layout_hdr(lo);
390 static void pnfs_free_lseg_async(struct pnfs_layout_segment *lseg)
392 INIT_WORK(&lseg->pls_work, pnfs_free_lseg_async_work);
393 schedule_work(&lseg->pls_work);
397 pnfs_put_lseg_locked(struct pnfs_layout_segment *lseg)
402 assert_spin_locked(&lseg->pls_layout->plh_inode->i_lock);
404 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
405 atomic_read(&lseg->pls_refcount),
406 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
407 if (atomic_dec_and_test(&lseg->pls_refcount)) {
408 struct pnfs_layout_hdr *lo = lseg->pls_layout;
409 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
411 pnfs_get_layout_hdr(lo);
412 pnfs_layout_remove_lseg(lo, lseg);
413 pnfs_free_lseg_async(lseg);
416 EXPORT_SYMBOL_GPL(pnfs_put_lseg_locked);
419 end_offset(u64 start, u64 len)
424 return end >= start ? end : NFS4_MAX_UINT64;
428 * is l2 fully contained in l1?
430 * [----------------------------------)
435 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
436 const struct pnfs_layout_range *l2)
438 u64 start1 = l1->offset;
439 u64 end1 = end_offset(start1, l1->length);
440 u64 start2 = l2->offset;
441 u64 end2 = end_offset(start2, l2->length);
443 return (start1 <= start2) && (end1 >= end2);
447 * is l1 and l2 intersecting?
449 * [----------------------------------)
454 pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
455 const struct pnfs_layout_range *l2)
457 u64 start1 = l1->offset;
458 u64 end1 = end_offset(start1, l1->length);
459 u64 start2 = l2->offset;
460 u64 end2 = end_offset(start2, l2->length);
462 return (end1 == NFS4_MAX_UINT64 || end1 > start2) &&
463 (end2 == NFS4_MAX_UINT64 || end2 > start1);
467 should_free_lseg(const struct pnfs_layout_range *lseg_range,
468 const struct pnfs_layout_range *recall_range)
470 return (recall_range->iomode == IOMODE_ANY ||
471 lseg_range->iomode == recall_range->iomode) &&
472 pnfs_lseg_range_intersecting(lseg_range, recall_range);
475 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
476 struct list_head *tmp_list)
478 if (!atomic_dec_and_test(&lseg->pls_refcount))
480 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
481 list_add(&lseg->pls_list, tmp_list);
485 /* Returns 1 if lseg is removed from list, 0 otherwise */
486 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
487 struct list_head *tmp_list)
491 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
492 /* Remove the reference keeping the lseg in the
493 * list. It will now be removed when all
494 * outstanding io is finished.
496 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
497 atomic_read(&lseg->pls_refcount));
498 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
504 /* Returns count of number of matching invalid lsegs remaining in list
508 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
509 struct list_head *tmp_list,
510 const struct pnfs_layout_range *recall_range)
512 struct pnfs_layout_segment *lseg, *next;
515 dprintk("%s:Begin lo %p\n", __func__, lo);
517 if (list_empty(&lo->plh_segs))
519 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
521 should_free_lseg(&lseg->pls_range, recall_range)) {
522 dprintk("%s: freeing lseg %p iomode %d "
523 "offset %llu length %llu\n", __func__,
524 lseg, lseg->pls_range.iomode, lseg->pls_range.offset,
525 lseg->pls_range.length);
526 if (!mark_lseg_invalid(lseg, tmp_list))
529 dprintk("%s:Return %i\n", __func__, remaining);
533 /* note free_me must contain lsegs from a single layout_hdr */
535 pnfs_free_lseg_list(struct list_head *free_me)
537 struct pnfs_layout_segment *lseg, *tmp;
539 if (list_empty(free_me))
542 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
543 list_del(&lseg->pls_list);
544 pnfs_free_lseg(lseg);
549 pnfs_destroy_layout(struct nfs_inode *nfsi)
551 struct pnfs_layout_hdr *lo;
554 spin_lock(&nfsi->vfs_inode.i_lock);
557 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
558 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
559 pnfs_get_layout_hdr(lo);
560 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
561 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
562 spin_unlock(&nfsi->vfs_inode.i_lock);
563 pnfs_free_lseg_list(&tmp_list);
564 pnfs_put_layout_hdr(lo);
566 spin_unlock(&nfsi->vfs_inode.i_lock);
568 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
571 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
572 struct list_head *layout_list)
574 struct pnfs_layout_hdr *lo;
577 spin_lock(&inode->i_lock);
578 lo = NFS_I(inode)->layout;
579 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
580 pnfs_get_layout_hdr(lo);
581 list_add(&lo->plh_bulk_destroy, layout_list);
584 spin_unlock(&inode->i_lock);
588 /* Caller must hold rcu_read_lock and clp->cl_lock */
590 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
591 struct nfs_server *server,
592 struct list_head *layout_list)
594 struct pnfs_layout_hdr *lo, *next;
597 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
598 inode = igrab(lo->plh_inode);
601 list_del_init(&lo->plh_layouts);
602 if (pnfs_layout_add_bulk_destroy_list(inode, layout_list))
605 spin_unlock(&clp->cl_lock);
607 spin_lock(&clp->cl_lock);
615 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
618 struct pnfs_layout_hdr *lo;
620 struct pnfs_layout_range range = {
621 .iomode = IOMODE_ANY,
623 .length = NFS4_MAX_UINT64,
625 LIST_HEAD(lseg_list);
628 while (!list_empty(layout_list)) {
629 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
631 dprintk("%s freeing layout for inode %lu\n", __func__,
632 lo->plh_inode->i_ino);
633 inode = lo->plh_inode;
635 pnfs_layoutcommit_inode(inode, false);
637 spin_lock(&inode->i_lock);
638 list_del_init(&lo->plh_bulk_destroy);
639 lo->plh_block_lgets++; /* permanently block new LAYOUTGETs */
641 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
642 if (pnfs_mark_matching_lsegs_invalid(lo, &lseg_list, &range))
644 spin_unlock(&inode->i_lock);
645 pnfs_free_lseg_list(&lseg_list);
646 /* Free all lsegs that are attached to commit buckets */
647 nfs_commit_inode(inode, 0);
648 pnfs_put_layout_hdr(lo);
655 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
656 struct nfs_fsid *fsid,
659 struct nfs_server *server;
660 LIST_HEAD(layout_list);
662 spin_lock(&clp->cl_lock);
665 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
666 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
668 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
674 spin_unlock(&clp->cl_lock);
676 if (list_empty(&layout_list))
678 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
682 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
685 struct nfs_server *server;
686 LIST_HEAD(layout_list);
688 spin_lock(&clp->cl_lock);
691 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
692 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
698 spin_unlock(&clp->cl_lock);
700 if (list_empty(&layout_list))
702 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
706 * Called by the state manger to remove all layouts established under an
710 pnfs_destroy_all_layouts(struct nfs_client *clp)
712 nfs4_deviceid_mark_client_invalid(clp);
713 nfs4_deviceid_purge_client(clp);
715 pnfs_destroy_layouts_byclid(clp, false);
719 * Compare 2 layout stateid sequence ids, to see which is newer,
720 * taking into account wraparound issues.
722 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
724 return (s32)(s1 - s2) > 0;
727 /* update lo->plh_stateid with new if is more recent */
729 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
732 u32 oldseq, newseq, new_barrier;
733 int empty = list_empty(&lo->plh_segs);
735 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
736 newseq = be32_to_cpu(new->seqid);
737 if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
738 nfs4_stateid_copy(&lo->plh_stateid, new);
739 if (update_barrier) {
740 new_barrier = be32_to_cpu(new->seqid);
742 /* Because of wraparound, we want to keep the barrier
743 * "close" to the current seqids.
745 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
747 if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
748 lo->plh_barrier = new_barrier;
753 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
754 const nfs4_stateid *stateid)
756 u32 seqid = be32_to_cpu(stateid->seqid);
758 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
761 /* lget is set to 1 if called from inside send_layoutget call chain */
763 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
765 return lo->plh_block_lgets ||
766 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
770 pnfs_choose_layoutget_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
771 const struct pnfs_layout_range *range,
772 struct nfs4_state *open_state)
776 dprintk("--> %s\n", __func__);
777 spin_lock(&lo->plh_inode->i_lock);
778 if (pnfs_layoutgets_blocked(lo)) {
780 } else if (!nfs4_valid_open_stateid(open_state)) {
782 } else if (list_empty(&lo->plh_segs) ||
783 test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
787 seq = read_seqbegin(&open_state->seqlock);
788 nfs4_stateid_copy(dst, &open_state->stateid);
789 } while (read_seqretry(&open_state->seqlock, seq));
791 nfs4_stateid_copy(dst, &lo->plh_stateid);
792 spin_unlock(&lo->plh_inode->i_lock);
793 dprintk("<-- %s\n", __func__);
798 * Get layout from server.
799 * for now, assume that whole file layouts are requested.
801 * arg->length: all ones
803 static struct pnfs_layout_segment *
804 send_layoutget(struct pnfs_layout_hdr *lo,
805 struct nfs_open_context *ctx,
806 const struct pnfs_layout_range *range,
809 struct inode *ino = lo->plh_inode;
810 struct nfs_server *server = NFS_SERVER(ino);
811 struct nfs4_layoutget *lgp;
812 struct pnfs_layout_segment *lseg;
815 dprintk("--> %s\n", __func__);
818 * Synchronously retrieve layout information from server and
819 * store in lseg. If we race with a concurrent seqid morphing
820 * op, then re-send the LAYOUTGET.
823 lgp = kzalloc(sizeof(*lgp), gfp_flags);
827 i_size = i_size_read(ino);
829 lgp->args.minlength = PAGE_CACHE_SIZE;
830 if (lgp->args.minlength > range->length)
831 lgp->args.minlength = range->length;
832 if (range->iomode == IOMODE_READ) {
833 if (range->offset >= i_size)
834 lgp->args.minlength = 0;
835 else if (i_size - range->offset < lgp->args.minlength)
836 lgp->args.minlength = i_size - range->offset;
838 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
839 pnfs_copy_range(&lgp->args.range, range);
840 lgp->args.type = server->pnfs_curr_ld->id;
841 lgp->args.inode = ino;
842 lgp->args.ctx = get_nfs_open_context(ctx);
843 lgp->gfp_flags = gfp_flags;
844 lgp->cred = lo->plh_lc_cred;
846 lseg = nfs4_proc_layoutget(lgp, gfp_flags);
847 } while (lseg == ERR_PTR(-EAGAIN));
849 if (IS_ERR(lseg) && !nfs_error_is_fatal(PTR_ERR(lseg)))
852 pnfs_layout_clear_fail_bit(lo,
853 pnfs_iomode_to_fail_bit(range->iomode));
858 static void pnfs_clear_layoutcommit(struct inode *inode,
859 struct list_head *head)
861 struct nfs_inode *nfsi = NFS_I(inode);
862 struct pnfs_layout_segment *lseg, *tmp;
864 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
866 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
867 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
869 pnfs_lseg_dec_and_remove_zero(lseg, head);
873 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
875 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
876 smp_mb__after_atomic();
877 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
878 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
882 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
884 if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
886 lo->plh_return_iomode = 0;
887 pnfs_get_layout_hdr(lo);
888 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
893 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
894 enum pnfs_iomode iomode, bool sync)
896 struct inode *ino = lo->plh_inode;
897 struct nfs4_layoutreturn *lrp;
900 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
901 if (unlikely(lrp == NULL)) {
903 spin_lock(&ino->i_lock);
904 pnfs_clear_layoutreturn_waitbit(lo);
905 spin_unlock(&ino->i_lock);
906 pnfs_put_layout_hdr(lo);
910 nfs4_stateid_copy(&lrp->args.stateid, stateid);
911 lrp->args.layout_type = NFS_SERVER(ino)->pnfs_curr_ld->id;
912 lrp->args.inode = ino;
913 lrp->args.range.iomode = iomode;
914 lrp->args.range.offset = 0;
915 lrp->args.range.length = NFS4_MAX_UINT64;
916 lrp->args.layout = lo;
917 lrp->clp = NFS_SERVER(ino)->nfs_client;
918 lrp->cred = lo->plh_lc_cred;
920 status = nfs4_proc_layoutreturn(lrp, sync);
922 dprintk("<-- %s status: %d\n", __func__, status);
926 /* Return true if layoutreturn is needed */
928 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
930 struct pnfs_layout_segment *s;
932 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
935 /* Defer layoutreturn until all lsegs are done */
936 list_for_each_entry(s, &lo->plh_segs, pls_list) {
937 if (test_bit(NFS_LSEG_LAYOUTRETURN, &s->pls_flags))
944 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
946 struct inode *inode= lo->plh_inode;
948 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
950 spin_lock(&inode->i_lock);
951 if (pnfs_layout_need_return(lo)) {
952 nfs4_stateid stateid;
953 enum pnfs_iomode iomode;
956 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
957 iomode = lo->plh_return_iomode;
958 send = pnfs_prepare_layoutreturn(lo);
959 spin_unlock(&inode->i_lock);
961 /* Send an async layoutreturn so we dont deadlock */
962 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
965 spin_unlock(&inode->i_lock);
969 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
970 * when the layout segment list is empty.
972 * Note that a pnfs_layout_hdr can exist with an empty layout segment
973 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
974 * deviceid is marked invalid.
977 _pnfs_return_layout(struct inode *ino)
979 struct pnfs_layout_hdr *lo = NULL;
980 struct nfs_inode *nfsi = NFS_I(ino);
982 nfs4_stateid stateid;
983 int status = 0, empty;
986 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
988 spin_lock(&ino->i_lock);
991 spin_unlock(&ino->i_lock);
992 dprintk("NFS: %s no layout to return\n", __func__);
995 nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid);
996 /* Reference matched in nfs4_layoutreturn_release */
997 pnfs_get_layout_hdr(lo);
998 empty = list_empty(&lo->plh_segs);
999 pnfs_clear_layoutcommit(ino, &tmp_list);
1000 pnfs_mark_matching_lsegs_invalid(lo, &tmp_list, NULL);
1002 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
1003 struct pnfs_layout_range range = {
1004 .iomode = IOMODE_ANY,
1006 .length = NFS4_MAX_UINT64,
1008 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1011 /* Don't send a LAYOUTRETURN if list was initially empty */
1013 spin_unlock(&ino->i_lock);
1014 dprintk("NFS: %s no layout segments to return\n", __func__);
1015 goto out_put_layout_hdr;
1018 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1019 send = pnfs_prepare_layoutreturn(lo);
1020 spin_unlock(&ino->i_lock);
1021 pnfs_free_lseg_list(&tmp_list);
1023 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1025 pnfs_put_layout_hdr(lo);
1027 dprintk("<-- %s status: %d\n", __func__, status);
1030 EXPORT_SYMBOL_GPL(_pnfs_return_layout);
1033 pnfs_commit_and_return_layout(struct inode *inode)
1035 struct pnfs_layout_hdr *lo;
1038 spin_lock(&inode->i_lock);
1039 lo = NFS_I(inode)->layout;
1041 spin_unlock(&inode->i_lock);
1044 pnfs_get_layout_hdr(lo);
1045 /* Block new layoutgets and read/write to ds */
1046 lo->plh_block_lgets++;
1047 spin_unlock(&inode->i_lock);
1048 filemap_fdatawait(inode->i_mapping);
1049 ret = pnfs_layoutcommit_inode(inode, true);
1051 ret = _pnfs_return_layout(inode);
1052 spin_lock(&inode->i_lock);
1053 lo->plh_block_lgets--;
1054 spin_unlock(&inode->i_lock);
1055 pnfs_put_layout_hdr(lo);
1059 bool pnfs_roc(struct inode *ino)
1061 struct nfs_inode *nfsi = NFS_I(ino);
1062 struct nfs_open_context *ctx;
1063 struct nfs4_state *state;
1064 struct pnfs_layout_hdr *lo;
1065 struct pnfs_layout_segment *lseg, *tmp;
1066 nfs4_stateid stateid;
1067 LIST_HEAD(tmp_list);
1068 bool found = false, layoutreturn = false, roc = false;
1070 spin_lock(&ino->i_lock);
1072 if (!lo || test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags))
1075 /* no roc if we hold a delegation */
1076 if (nfs4_check_delegation(ino, FMODE_READ))
1079 list_for_each_entry(ctx, &nfsi->open_files, list) {
1081 /* Don't return layout if there is open file state */
1082 if (state != NULL && state->state != 0)
1086 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1087 /* always send layoutreturn if being marked so */
1088 if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED,
1090 layoutreturn = pnfs_prepare_layoutreturn(lo);
1092 list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
1093 /* If we are sending layoutreturn, invalidate all valid lsegs */
1094 if (layoutreturn || test_bit(NFS_LSEG_ROC, &lseg->pls_flags)) {
1095 mark_lseg_invalid(lseg, &tmp_list);
1098 /* ROC in two conditions:
1099 * 1. there are ROC lsegs
1100 * 2. we don't send layoutreturn
1102 if (found && !layoutreturn) {
1103 /* lo ref dropped in pnfs_roc_release() */
1104 pnfs_get_layout_hdr(lo);
1109 spin_unlock(&ino->i_lock);
1110 pnfs_free_lseg_list(&tmp_list);
1111 pnfs_layoutcommit_inode(ino, true);
1113 pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1117 void pnfs_roc_release(struct inode *ino)
1119 struct pnfs_layout_hdr *lo;
1121 spin_lock(&ino->i_lock);
1122 lo = NFS_I(ino)->layout;
1123 pnfs_clear_layoutreturn_waitbit(lo);
1124 if (atomic_dec_and_test(&lo->plh_refcount)) {
1125 pnfs_detach_layout_hdr(lo);
1126 spin_unlock(&ino->i_lock);
1127 pnfs_free_layout_hdr(lo);
1129 spin_unlock(&ino->i_lock);
1132 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
1134 struct pnfs_layout_hdr *lo;
1136 spin_lock(&ino->i_lock);
1137 lo = NFS_I(ino)->layout;
1138 pnfs_mark_layout_returned_if_empty(lo);
1139 if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
1140 lo->plh_barrier = barrier;
1141 spin_unlock(&ino->i_lock);
1142 trace_nfs4_layoutreturn_on_close(ino, 0);
1145 void pnfs_roc_get_barrier(struct inode *ino, u32 *barrier)
1147 struct nfs_inode *nfsi = NFS_I(ino);
1148 struct pnfs_layout_hdr *lo;
1151 spin_lock(&ino->i_lock);
1153 current_seqid = be32_to_cpu(lo->plh_stateid.seqid);
1155 /* Since close does not return a layout stateid for use as
1156 * a barrier, we choose the worst-case barrier.
1158 *barrier = current_seqid + atomic_read(&lo->plh_outstanding);
1159 spin_unlock(&ino->i_lock);
1162 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1164 struct nfs_inode *nfsi = NFS_I(ino);
1165 struct pnfs_layout_hdr *lo;
1168 /* we might not have grabbed lo reference. so need to check under
1170 spin_lock(&ino->i_lock);
1172 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
1174 spin_unlock(&ino->i_lock);
1177 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1183 * Compare two layout segments for sorting into layout cache.
1184 * We want to preferentially return RW over RO layouts, so ensure those
1188 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1189 const struct pnfs_layout_range *l2)
1193 /* high offset > low offset */
1194 d = l1->offset - l2->offset;
1198 /* short length > long length */
1199 d = l2->length - l1->length;
1203 /* read > read/write */
1204 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1208 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1209 const struct pnfs_layout_range *l2)
1211 return pnfs_lseg_range_cmp(l1, l2) > 0;
1215 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1216 struct pnfs_layout_segment *old)
1222 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1223 struct pnfs_layout_segment *lseg,
1224 bool (*is_after)(const struct pnfs_layout_range *,
1225 const struct pnfs_layout_range *),
1226 bool (*do_merge)(struct pnfs_layout_segment *,
1227 struct pnfs_layout_segment *),
1228 struct list_head *free_me)
1230 struct pnfs_layout_segment *lp, *tmp;
1232 dprintk("%s:Begin\n", __func__);
1234 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1235 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1237 if (do_merge(lseg, lp)) {
1238 mark_lseg_invalid(lp, free_me);
1241 if (is_after(&lseg->pls_range, &lp->pls_range))
1243 list_add_tail(&lseg->pls_list, &lp->pls_list);
1244 dprintk("%s: inserted lseg %p "
1245 "iomode %d offset %llu length %llu before "
1246 "lp %p iomode %d offset %llu length %llu\n",
1247 __func__, lseg, lseg->pls_range.iomode,
1248 lseg->pls_range.offset, lseg->pls_range.length,
1249 lp, lp->pls_range.iomode, lp->pls_range.offset,
1250 lp->pls_range.length);
1253 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1254 dprintk("%s: inserted lseg %p "
1255 "iomode %d offset %llu length %llu at tail\n",
1256 __func__, lseg, lseg->pls_range.iomode,
1257 lseg->pls_range.offset, lseg->pls_range.length);
1259 pnfs_get_layout_hdr(lo);
1261 dprintk("%s:Return\n", __func__);
1263 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1266 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1267 struct pnfs_layout_segment *lseg,
1268 struct list_head *free_me)
1270 struct inode *inode = lo->plh_inode;
1271 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1273 if (ld->add_lseg != NULL)
1274 ld->add_lseg(lo, lseg, free_me);
1276 pnfs_generic_layout_insert_lseg(lo, lseg,
1277 pnfs_lseg_range_is_after,
1282 static struct pnfs_layout_hdr *
1283 alloc_init_layout_hdr(struct inode *ino,
1284 struct nfs_open_context *ctx,
1287 struct pnfs_layout_hdr *lo;
1289 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1292 atomic_set(&lo->plh_refcount, 1);
1293 INIT_LIST_HEAD(&lo->plh_layouts);
1294 INIT_LIST_HEAD(&lo->plh_segs);
1295 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1296 lo->plh_inode = ino;
1297 lo->plh_lc_cred = get_rpccred(ctx->cred);
1301 static struct pnfs_layout_hdr *
1302 pnfs_find_alloc_layout(struct inode *ino,
1303 struct nfs_open_context *ctx,
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:
1339 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1340 const struct pnfs_layout_range *range)
1342 struct pnfs_layout_range range1;
1344 if ((range->iomode == IOMODE_RW &&
1345 ls_range->iomode != IOMODE_RW) ||
1346 !pnfs_lseg_range_intersecting(ls_range, range))
1349 /* range1 covers only the first byte in the range */
1352 return pnfs_lseg_range_contained(ls_range, &range1);
1356 * lookup range in layout
1358 static struct pnfs_layout_segment *
1359 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1360 struct pnfs_layout_range *range)
1362 struct pnfs_layout_segment *lseg, *ret = NULL;
1364 dprintk("%s:Begin\n", __func__);
1366 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1367 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1368 !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
1369 pnfs_lseg_range_match(&lseg->pls_range, range)) {
1370 ret = pnfs_get_lseg(lseg);
1375 dprintk("%s:Return lseg %p ref %d\n",
1376 __func__, ret, ret ? atomic_read(&ret->pls_refcount) : 0);
1381 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1382 * to the MDS or over pNFS
1384 * The nfs_inode read_io and write_io fields are cumulative counters reset
1385 * when there are no layout segments. Note that in pnfs_update_layout iomode
1386 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1389 * A return of true means use MDS I/O.
1392 * If a file's size is smaller than the file size threshold, data accesses
1393 * SHOULD be sent to the metadata server. If an I/O request has a length that
1394 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1395 * server. If both file size and I/O size are provided, the client SHOULD
1396 * reach or exceed both thresholds before sending its read or write
1397 * requests to the data server.
1399 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1400 struct inode *ino, int iomode)
1402 struct nfs4_threshold *t = ctx->mdsthreshold;
1403 struct nfs_inode *nfsi = NFS_I(ino);
1404 loff_t fsize = i_size_read(ino);
1405 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1410 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1411 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1415 if (t->bm & THRESHOLD_RD) {
1416 dprintk("%s fsize %llu\n", __func__, fsize);
1418 if (fsize < t->rd_sz)
1421 if (t->bm & THRESHOLD_RD_IO) {
1422 dprintk("%s nfsi->read_io %llu\n", __func__,
1425 if (nfsi->read_io < t->rd_io_sz)
1430 if (t->bm & THRESHOLD_WR) {
1431 dprintk("%s fsize %llu\n", __func__, fsize);
1433 if (fsize < t->wr_sz)
1436 if (t->bm & THRESHOLD_WR_IO) {
1437 dprintk("%s nfsi->write_io %llu\n", __func__,
1440 if (nfsi->write_io < t->wr_io_sz)
1445 if (size_set && io_set) {
1448 } else if (size || io)
1451 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1455 static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1458 * send layoutcommit as it can hold up layoutreturn due to lseg
1461 pnfs_layoutcommit_inode(lo->plh_inode, false);
1462 return !wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1463 nfs_wait_bit_killable,
1464 TASK_UNINTERRUPTIBLE);
1467 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1469 unsigned long *bitlock = &lo->plh_flags;
1471 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1472 smp_mb__after_atomic();
1473 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1477 * Layout segment is retreived from the server if not cached.
1478 * The appropriate layout segment is referenced and returned to the caller.
1480 struct pnfs_layout_segment *
1481 pnfs_update_layout(struct inode *ino,
1482 struct nfs_open_context *ctx,
1485 enum pnfs_iomode iomode,
1488 struct pnfs_layout_range arg = {
1494 struct nfs_server *server = NFS_SERVER(ino);
1495 struct nfs_client *clp = server->nfs_client;
1496 struct pnfs_layout_hdr *lo;
1497 struct pnfs_layout_segment *lseg = NULL;
1500 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1501 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1502 PNFS_UPDATE_LAYOUT_NO_PNFS);
1506 if (iomode == IOMODE_READ && i_size_read(ino) == 0) {
1507 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1508 PNFS_UPDATE_LAYOUT_RD_ZEROLEN);
1512 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1513 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1514 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1520 spin_lock(&ino->i_lock);
1521 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1523 spin_unlock(&ino->i_lock);
1524 trace_pnfs_update_layout(ino, pos, count, iomode, NULL,
1525 PNFS_UPDATE_LAYOUT_NOMEM);
1529 /* Do we even need to bother with this? */
1530 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1531 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1532 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1533 dprintk("%s matches recall, use MDS\n", __func__);
1537 /* if LAYOUTGET already failed once we don't try again */
1538 if (pnfs_layout_io_test_failed(lo, iomode)) {
1539 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1540 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1544 first = list_empty(&lo->plh_segs);
1546 /* The first layoutget for the file. Need to serialize per
1547 * RFC 5661 Errata 3208.
1549 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1551 spin_unlock(&ino->i_lock);
1552 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_FIRST_LAYOUTGET,
1553 TASK_UNINTERRUPTIBLE);
1554 pnfs_put_layout_hdr(lo);
1558 /* Check to see if the layout for the given range
1561 lseg = pnfs_find_lseg(lo, &arg);
1563 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1564 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1570 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1571 * for LAYOUTRETURN even if first is true.
1573 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1574 spin_unlock(&ino->i_lock);
1575 dprintk("%s wait for layoutreturn\n", __func__);
1576 if (pnfs_prepare_to_retry_layoutget(lo)) {
1578 pnfs_clear_first_layoutget(lo);
1579 pnfs_put_layout_hdr(lo);
1580 dprintk("%s retrying\n", __func__);
1583 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1584 PNFS_UPDATE_LAYOUT_RETURN);
1585 goto out_put_layout_hdr;
1588 if (pnfs_layoutgets_blocked(lo)) {
1589 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1590 PNFS_UPDATE_LAYOUT_BLOCKED);
1593 atomic_inc(&lo->plh_outstanding);
1594 spin_unlock(&ino->i_lock);
1596 if (list_empty(&lo->plh_layouts)) {
1597 /* The lo must be on the clp list if there is any
1598 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1600 spin_lock(&clp->cl_lock);
1601 if (list_empty(&lo->plh_layouts))
1602 list_add_tail(&lo->plh_layouts, &server->layouts);
1603 spin_unlock(&clp->cl_lock);
1606 pg_offset = arg.offset & ~PAGE_CACHE_MASK;
1608 arg.offset -= pg_offset;
1609 arg.length += pg_offset;
1611 if (arg.length != NFS4_MAX_UINT64)
1612 arg.length = PAGE_CACHE_ALIGN(arg.length);
1614 lseg = send_layoutget(lo, ctx, &arg, gfp_flags);
1615 atomic_dec(&lo->plh_outstanding);
1616 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1617 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
1620 pnfs_clear_first_layoutget(lo);
1621 pnfs_put_layout_hdr(lo);
1623 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
1624 "(%s, offset: %llu, length: %llu)\n",
1625 __func__, ino->i_sb->s_id,
1626 (unsigned long long)NFS_FILEID(ino),
1627 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
1628 iomode==IOMODE_RW ? "read/write" : "read-only",
1629 (unsigned long long)pos,
1630 (unsigned long long)count);
1633 spin_unlock(&ino->i_lock);
1634 goto out_put_layout_hdr;
1636 EXPORT_SYMBOL_GPL(pnfs_update_layout);
1639 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
1641 switch (range->iomode) {
1648 if (range->offset == NFS4_MAX_UINT64)
1650 if (range->length == 0)
1652 if (range->length != NFS4_MAX_UINT64 &&
1653 range->length > NFS4_MAX_UINT64 - range->offset)
1658 struct pnfs_layout_segment *
1659 pnfs_layout_process(struct nfs4_layoutget *lgp)
1661 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
1662 struct nfs4_layoutget_res *res = &lgp->res;
1663 struct pnfs_layout_segment *lseg;
1664 struct inode *ino = lo->plh_inode;
1666 int status = -EINVAL;
1668 if (!pnfs_sanity_check_layout_range(&res->range))
1671 /* Inject layout blob into I/O device driver */
1672 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
1673 if (!lseg || IS_ERR(lseg)) {
1677 status = PTR_ERR(lseg);
1678 dprintk("%s: Could not allocate layout: error %d\n",
1683 init_lseg(lo, lseg);
1684 lseg->pls_range = res->range;
1686 spin_lock(&ino->i_lock);
1687 if (pnfs_layoutgets_blocked(lo)) {
1688 dprintk("%s forget reply due to state\n", __func__);
1689 goto out_forget_reply;
1692 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
1693 /* existing state ID, make sure the sequence number matches. */
1694 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
1695 dprintk("%s forget reply due to sequence\n", __func__);
1697 goto out_forget_reply;
1699 pnfs_set_layout_stateid(lo, &res->stateid, false);
1702 * We got an entirely new state ID. Mark all segments for the
1703 * inode invalid, and don't bother validating the stateid
1706 pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL);
1708 nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
1709 lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
1712 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
1714 pnfs_get_lseg(lseg);
1715 pnfs_layout_insert_lseg(lo, lseg, &free_me);
1717 if (res->return_on_close)
1718 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
1720 spin_unlock(&ino->i_lock);
1721 pnfs_free_lseg_list(&free_me);
1724 return ERR_PTR(status);
1727 spin_unlock(&ino->i_lock);
1728 lseg->pls_layout = lo;
1729 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
1734 pnfs_set_plh_return_iomode(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode)
1736 if (lo->plh_return_iomode == iomode)
1738 if (lo->plh_return_iomode != 0)
1739 iomode = IOMODE_ANY;
1740 lo->plh_return_iomode = iomode;
1744 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
1745 struct list_head *tmp_list,
1746 const struct pnfs_layout_range *return_range)
1748 struct pnfs_layout_segment *lseg, *next;
1751 dprintk("%s:Begin lo %p\n", __func__, lo);
1753 if (list_empty(&lo->plh_segs))
1756 assert_spin_locked(&lo->plh_inode->i_lock);
1758 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
1759 if (should_free_lseg(&lseg->pls_range, return_range)) {
1760 dprintk("%s: marking lseg %p iomode %d "
1761 "offset %llu length %llu\n", __func__,
1762 lseg, lseg->pls_range.iomode,
1763 lseg->pls_range.offset,
1764 lseg->pls_range.length);
1765 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1766 pnfs_set_plh_return_iomode(lo, return_range->iomode);
1767 if (!mark_lseg_invalid(lseg, tmp_list))
1769 set_bit(NFS_LAYOUT_RETURN_REQUESTED,
1775 void pnfs_error_mark_layout_for_return(struct inode *inode,
1776 struct pnfs_layout_segment *lseg)
1778 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
1779 struct pnfs_layout_range range = {
1780 .iomode = lseg->pls_range.iomode,
1782 .length = NFS4_MAX_UINT64,
1785 bool return_now = false;
1787 spin_lock(&inode->i_lock);
1788 pnfs_set_plh_return_iomode(lo, range.iomode);
1790 * mark all matching lsegs so that we are sure to have no live
1791 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
1794 if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range)) {
1795 nfs4_stateid stateid;
1796 enum pnfs_iomode iomode = lo->plh_return_iomode;
1798 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1799 return_now = pnfs_prepare_layoutreturn(lo);
1800 spin_unlock(&inode->i_lock);
1802 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1804 spin_unlock(&inode->i_lock);
1805 nfs_commit_inode(inode, 0);
1807 pnfs_free_lseg_list(&free_me);
1809 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
1812 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
1814 u64 rd_size = req->wb_bytes;
1816 if (pgio->pg_lseg == NULL) {
1817 if (pgio->pg_dreq == NULL)
1818 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
1820 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
1822 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1828 if (IS_ERR(pgio->pg_lseg)) {
1829 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1830 pgio->pg_lseg = NULL;
1834 /* If no lseg, fall back to read through mds */
1835 if (pgio->pg_lseg == NULL)
1836 nfs_pageio_reset_read_mds(pgio);
1839 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
1842 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
1843 struct nfs_page *req, u64 wb_size)
1845 if (pgio->pg_lseg == NULL) {
1846 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1852 if (IS_ERR(pgio->pg_lseg)) {
1853 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1854 pgio->pg_lseg = NULL;
1858 /* If no lseg, fall back to write through mds */
1859 if (pgio->pg_lseg == NULL)
1860 nfs_pageio_reset_write_mds(pgio);
1862 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
1865 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
1867 if (desc->pg_lseg) {
1868 pnfs_put_lseg(desc->pg_lseg);
1869 desc->pg_lseg = NULL;
1872 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
1875 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
1876 * of bytes (maximum @req->wb_bytes) that can be coalesced.
1879 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
1880 struct nfs_page *prev, struct nfs_page *req)
1883 u64 seg_end, req_start, seg_left;
1885 size = nfs_generic_pg_test(pgio, prev, req);
1890 * 'size' contains the number of bytes left in the current page (up
1891 * to the original size asked for in @req->wb_bytes).
1893 * Calculate how many bytes are left in the layout segment
1894 * and if there are less bytes than 'size', return that instead.
1896 * Please also note that 'end_offset' is actually the offset of the
1897 * first byte that lies outside the pnfs_layout_range. FIXME?
1900 if (pgio->pg_lseg) {
1901 seg_end = end_offset(pgio->pg_lseg->pls_range.offset,
1902 pgio->pg_lseg->pls_range.length);
1903 req_start = req_offset(req);
1904 WARN_ON_ONCE(req_start >= seg_end);
1905 /* start of request is past the last byte of this segment */
1906 if (req_start >= seg_end) {
1907 /* reference the new lseg */
1908 if (pgio->pg_ops->pg_cleanup)
1909 pgio->pg_ops->pg_cleanup(pgio);
1910 if (pgio->pg_ops->pg_init)
1911 pgio->pg_ops->pg_init(pgio, req);
1915 /* adjust 'size' iff there are fewer bytes left in the
1916 * segment than what nfs_generic_pg_test returned */
1917 seg_left = seg_end - req_start;
1918 if (seg_left < size)
1919 size = (unsigned int)seg_left;
1924 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
1926 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
1928 struct nfs_pageio_descriptor pgio;
1930 /* Resend all requests through the MDS */
1931 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
1932 hdr->completion_ops);
1933 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
1934 return nfs_pageio_resend(&pgio, hdr);
1936 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
1938 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
1941 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
1942 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
1943 PNFS_LAYOUTRET_ON_ERROR) {
1944 pnfs_return_layout(hdr->inode);
1946 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
1947 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
1951 * Called by non rpc-based layout drivers
1953 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
1955 if (likely(!hdr->pnfs_error)) {
1956 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1957 hdr->mds_offset + hdr->res.count);
1958 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
1960 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
1961 if (unlikely(hdr->pnfs_error))
1962 pnfs_ld_handle_write_error(hdr);
1963 hdr->mds_ops->rpc_release(hdr);
1965 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
1968 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
1969 struct nfs_pgio_header *hdr)
1971 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1973 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1974 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
1975 nfs_pageio_reset_write_mds(desc);
1976 mirror->pg_recoalesce = 1;
1978 nfs_pgio_data_destroy(hdr);
1982 static enum pnfs_try_status
1983 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
1984 const struct rpc_call_ops *call_ops,
1985 struct pnfs_layout_segment *lseg,
1988 struct inode *inode = hdr->inode;
1989 enum pnfs_try_status trypnfs;
1990 struct nfs_server *nfss = NFS_SERVER(inode);
1992 hdr->mds_ops = call_ops;
1994 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
1995 inode->i_ino, hdr->args.count, hdr->args.offset, how);
1996 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
1997 if (trypnfs != PNFS_NOT_ATTEMPTED)
1998 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
1999 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2004 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2005 struct nfs_pgio_header *hdr, int how)
2007 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2008 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2009 enum pnfs_try_status trypnfs;
2011 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2012 if (trypnfs == PNFS_NOT_ATTEMPTED)
2013 pnfs_write_through_mds(desc, hdr);
2016 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2018 pnfs_put_lseg(hdr->lseg);
2019 nfs_pgio_header_free(hdr);
2023 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2025 struct nfs_pgio_header *hdr;
2028 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2030 desc->pg_error = -ENOMEM;
2031 return desc->pg_error;
2033 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2035 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2036 ret = nfs_generic_pgio(desc, hdr);
2038 pnfs_do_write(desc, hdr, desc->pg_ioflags);
2042 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2044 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2046 struct nfs_pageio_descriptor pgio;
2048 /* Resend all requests through the MDS */
2049 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2050 return nfs_pageio_resend(&pgio, hdr);
2052 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2054 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2056 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2057 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2058 PNFS_LAYOUTRET_ON_ERROR) {
2059 pnfs_return_layout(hdr->inode);
2061 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2062 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2066 * Called by non rpc-based layout drivers
2068 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2070 if (likely(!hdr->pnfs_error)) {
2071 __nfs4_read_done_cb(hdr);
2072 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2074 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2075 if (unlikely(hdr->pnfs_error))
2076 pnfs_ld_handle_read_error(hdr);
2077 hdr->mds_ops->rpc_release(hdr);
2079 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2082 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2083 struct nfs_pgio_header *hdr)
2085 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2087 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2088 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2089 nfs_pageio_reset_read_mds(desc);
2090 mirror->pg_recoalesce = 1;
2092 nfs_pgio_data_destroy(hdr);
2097 * Call the appropriate parallel I/O subsystem read function.
2099 static enum pnfs_try_status
2100 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2101 const struct rpc_call_ops *call_ops,
2102 struct pnfs_layout_segment *lseg)
2104 struct inode *inode = hdr->inode;
2105 struct nfs_server *nfss = NFS_SERVER(inode);
2106 enum pnfs_try_status trypnfs;
2108 hdr->mds_ops = call_ops;
2110 dprintk("%s: Reading ino:%lu %u@%llu\n",
2111 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2113 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2114 if (trypnfs != PNFS_NOT_ATTEMPTED)
2115 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2116 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2120 /* Resend all requests through pnfs. */
2121 int pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2123 struct nfs_pageio_descriptor pgio;
2125 nfs_pageio_init_read(&pgio, hdr->inode, false, hdr->completion_ops);
2126 return nfs_pageio_resend(&pgio, hdr);
2128 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2131 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2133 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2134 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2135 enum pnfs_try_status trypnfs;
2138 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2139 if (trypnfs == PNFS_TRY_AGAIN)
2140 err = pnfs_read_resend_pnfs(hdr);
2141 if (trypnfs == PNFS_NOT_ATTEMPTED || err)
2142 pnfs_read_through_mds(desc, hdr);
2145 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2147 pnfs_put_lseg(hdr->lseg);
2148 nfs_pgio_header_free(hdr);
2152 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2154 struct nfs_pgio_header *hdr;
2157 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2159 desc->pg_error = -ENOMEM;
2160 return desc->pg_error;
2162 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2163 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2164 ret = nfs_generic_pgio(desc, hdr);
2166 pnfs_do_read(desc, hdr);
2169 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2171 static void pnfs_clear_layoutcommitting(struct inode *inode)
2173 unsigned long *bitlock = &NFS_I(inode)->flags;
2175 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2176 smp_mb__after_atomic();
2177 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2181 * There can be multiple RW segments.
2183 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2185 struct pnfs_layout_segment *lseg;
2187 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2188 if (lseg->pls_range.iomode == IOMODE_RW &&
2189 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2190 list_add(&lseg->pls_lc_list, listp);
2194 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2196 struct pnfs_layout_segment *lseg, *tmp;
2198 /* Matched by references in pnfs_set_layoutcommit */
2199 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2200 list_del_init(&lseg->pls_lc_list);
2201 pnfs_put_lseg(lseg);
2204 pnfs_clear_layoutcommitting(inode);
2207 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2209 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2211 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2214 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2217 struct nfs_inode *nfsi = NFS_I(inode);
2218 bool mark_as_dirty = false;
2220 spin_lock(&inode->i_lock);
2221 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2222 nfsi->layout->plh_lwb = end_pos;
2223 mark_as_dirty = true;
2224 dprintk("%s: Set layoutcommit for inode %lu ",
2225 __func__, inode->i_ino);
2226 } else if (end_pos > nfsi->layout->plh_lwb)
2227 nfsi->layout->plh_lwb = end_pos;
2228 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2229 /* references matched in nfs4_layoutcommit_release */
2230 pnfs_get_lseg(lseg);
2232 spin_unlock(&inode->i_lock);
2233 dprintk("%s: lseg %p end_pos %llu\n",
2234 __func__, lseg, nfsi->layout->plh_lwb);
2236 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2237 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2239 mark_inode_dirty_sync(inode);
2241 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2243 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2245 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2247 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2248 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2249 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2253 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2254 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2255 * data to disk to allow the server to recover the data if it crashes.
2256 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2257 * is off, and a COMMIT is sent to a data server, or
2258 * if WRITEs to a data server return NFS_DATA_SYNC.
2261 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2263 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2264 struct nfs4_layoutcommit_data *data;
2265 struct nfs_inode *nfsi = NFS_I(inode);
2269 if (!pnfs_layoutcommit_outstanding(inode))
2272 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2275 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2278 status = wait_on_bit_lock_action(&nfsi->flags,
2279 NFS_INO_LAYOUTCOMMITTING,
2280 nfs_wait_bit_killable,
2287 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2288 data = kzalloc(sizeof(*data), GFP_NOFS);
2290 goto clear_layoutcommitting;
2293 spin_lock(&inode->i_lock);
2294 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2297 INIT_LIST_HEAD(&data->lseg_list);
2298 pnfs_list_write_lseg(inode, &data->lseg_list);
2300 end_pos = nfsi->layout->plh_lwb;
2302 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2303 spin_unlock(&inode->i_lock);
2305 data->args.inode = inode;
2306 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
2307 nfs_fattr_init(&data->fattr);
2308 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2309 data->res.fattr = &data->fattr;
2310 data->args.lastbytewritten = end_pos - 1;
2311 data->res.server = NFS_SERVER(inode);
2313 if (ld->prepare_layoutcommit) {
2314 status = ld->prepare_layoutcommit(&data->args);
2316 put_rpccred(data->cred);
2317 spin_lock(&inode->i_lock);
2318 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2319 if (end_pos > nfsi->layout->plh_lwb)
2320 nfsi->layout->plh_lwb = end_pos;
2326 status = nfs4_proc_layoutcommit(data, sync);
2329 mark_inode_dirty_sync(inode);
2330 dprintk("<-- %s status %d\n", __func__, status);
2333 spin_unlock(&inode->i_lock);
2335 clear_layoutcommitting:
2336 pnfs_clear_layoutcommitting(inode);
2339 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
2342 pnfs_generic_sync(struct inode *inode, bool datasync)
2344 return pnfs_layoutcommit_inode(inode, true);
2346 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
2348 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
2350 struct nfs4_threshold *thp;
2352 thp = kzalloc(sizeof(*thp), GFP_NOFS);
2354 dprintk("%s mdsthreshold allocation failed\n", __func__);
2360 #if IS_ENABLED(CONFIG_NFS_V4_2)
2362 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
2364 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2365 struct nfs_server *server = NFS_SERVER(inode);
2366 struct nfs_inode *nfsi = NFS_I(inode);
2367 struct nfs42_layoutstat_data *data;
2368 struct pnfs_layout_hdr *hdr;
2371 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
2374 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
2377 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
2380 spin_lock(&inode->i_lock);
2381 if (!NFS_I(inode)->layout) {
2382 spin_unlock(&inode->i_lock);
2385 hdr = NFS_I(inode)->layout;
2386 pnfs_get_layout_hdr(hdr);
2387 spin_unlock(&inode->i_lock);
2389 data = kzalloc(sizeof(*data), gfp_flags);
2395 data->args.fh = NFS_FH(inode);
2396 data->args.inode = inode;
2397 nfs4_stateid_copy(&data->args.stateid, &hdr->plh_stateid);
2398 status = ld->prepare_layoutstats(&data->args);
2402 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
2405 dprintk("%s returns %d\n", __func__, status);
2411 pnfs_put_layout_hdr(hdr);
2412 smp_mb__before_atomic();
2413 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
2414 smp_mb__after_atomic();
2417 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
2420 unsigned int layoutstats_timer;
2421 module_param(layoutstats_timer, uint, 0644);
2422 EXPORT_SYMBOL_GPL(layoutstats_timer);