2 * pNFS functions to call and manage layout drivers.
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
8 * Dean Hildebrand <dhildebz@umich.edu>
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_page.h>
32 #include <linux/module.h>
33 #include <linux/sort.h>
37 #include "nfs4trace.h"
38 #include "delegation.h"
42 #define NFSDBG_FACILITY NFSDBG_PNFS
43 #define PNFS_LAYOUTGET_RETRY_TIMEOUT (120*HZ)
48 * protects pnfs_modules_tbl.
50 static DEFINE_SPINLOCK(pnfs_spinlock);
53 * pnfs_modules_tbl holds all pnfs modules
55 static LIST_HEAD(pnfs_modules_tbl);
57 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo);
58 static void pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
59 struct list_head *free_me,
60 const struct pnfs_layout_range *range,
62 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
63 struct list_head *tmp_list);
65 /* Return the registered pnfs layout driver module matching given id */
66 static struct pnfs_layoutdriver_type *
67 find_pnfs_driver_locked(u32 id)
69 struct pnfs_layoutdriver_type *local;
71 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
76 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
80 static struct pnfs_layoutdriver_type *
81 find_pnfs_driver(u32 id)
83 struct pnfs_layoutdriver_type *local;
85 spin_lock(&pnfs_spinlock);
86 local = find_pnfs_driver_locked(id);
87 if (local != NULL && !try_module_get(local->owner)) {
88 dprintk("%s: Could not grab reference on module\n", __func__);
91 spin_unlock(&pnfs_spinlock);
95 const struct pnfs_layoutdriver_type *pnfs_find_layoutdriver(u32 id)
97 return find_pnfs_driver(id);
100 void pnfs_put_layoutdriver(const struct pnfs_layoutdriver_type *ld)
103 module_put(ld->owner);
107 unset_pnfs_layoutdriver(struct nfs_server *nfss)
109 if (nfss->pnfs_curr_ld) {
110 if (nfss->pnfs_curr_ld->clear_layoutdriver)
111 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
112 /* Decrement the MDS count. Purge the deviceid cache if zero */
113 if (atomic_dec_and_test(&nfss->nfs_client->cl_mds_count))
114 nfs4_deviceid_purge_client(nfss->nfs_client);
115 module_put(nfss->pnfs_curr_ld->owner);
117 nfss->pnfs_curr_ld = NULL;
121 * When the server sends a list of layout types, we choose one in the order
122 * given in the list below.
124 * FIXME: should this list be configurable in some fashion? module param?
125 * mount option? something else?
127 static const u32 ld_prefs[] = {
132 LAYOUT_NFSV4_1_FILES,
137 ld_cmp(const void *e1, const void *e2)
139 u32 ld1 = *((u32 *)e1);
140 u32 ld2 = *((u32 *)e2);
143 for (i = 0; ld_prefs[i] != 0; i++) {
144 if (ld1 == ld_prefs[i])
147 if (ld2 == ld_prefs[i])
154 * Try to set the server's pnfs module to the pnfs layout type specified by id.
155 * Currently only one pNFS layout driver per filesystem is supported.
157 * @ids array of layout types supported by MDS.
160 set_pnfs_layoutdriver(struct nfs_server *server, const struct nfs_fh *mntfh,
161 struct nfs_fsinfo *fsinfo)
163 struct pnfs_layoutdriver_type *ld_type = NULL;
167 if (fsinfo->nlayouttypes == 0)
169 if (!(server->nfs_client->cl_exchange_flags &
170 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
171 printk(KERN_ERR "NFS: %s: cl_exchange_flags 0x%x\n",
172 __func__, server->nfs_client->cl_exchange_flags);
176 sort(fsinfo->layouttype, fsinfo->nlayouttypes,
177 sizeof(*fsinfo->layouttype), ld_cmp, NULL);
179 for (i = 0; i < fsinfo->nlayouttypes; i++) {
180 id = fsinfo->layouttype[i];
181 ld_type = find_pnfs_driver(id);
183 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX,
185 ld_type = find_pnfs_driver(id);
192 dprintk("%s: No pNFS module found!\n", __func__);
196 server->pnfs_curr_ld = ld_type;
197 if (ld_type->set_layoutdriver
198 && ld_type->set_layoutdriver(server, mntfh)) {
199 printk(KERN_ERR "NFS: %s: Error initializing pNFS layout "
200 "driver %u.\n", __func__, id);
201 module_put(ld_type->owner);
204 /* Bump the MDS count */
205 atomic_inc(&server->nfs_client->cl_mds_count);
207 dprintk("%s: pNFS module for %u set\n", __func__, id);
211 dprintk("%s: Using NFSv4 I/O\n", __func__);
212 server->pnfs_curr_ld = NULL;
216 pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
218 int status = -EINVAL;
219 struct pnfs_layoutdriver_type *tmp;
221 if (ld_type->id == 0) {
222 printk(KERN_ERR "NFS: %s id 0 is reserved\n", __func__);
225 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
226 printk(KERN_ERR "NFS: %s Layout driver must provide "
227 "alloc_lseg and free_lseg.\n", __func__);
231 spin_lock(&pnfs_spinlock);
232 tmp = find_pnfs_driver_locked(ld_type->id);
234 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
236 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
239 printk(KERN_ERR "NFS: %s Module with id %d already loaded!\n",
240 __func__, ld_type->id);
242 spin_unlock(&pnfs_spinlock);
246 EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
249 pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
251 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
252 spin_lock(&pnfs_spinlock);
253 list_del(&ld_type->pnfs_tblid);
254 spin_unlock(&pnfs_spinlock);
256 EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
259 * pNFS client layout cache
262 /* Need to hold i_lock if caller does not already hold reference */
264 pnfs_get_layout_hdr(struct pnfs_layout_hdr *lo)
266 refcount_inc(&lo->plh_refcount);
269 static struct pnfs_layout_hdr *
270 pnfs_alloc_layout_hdr(struct inode *ino, gfp_t gfp_flags)
272 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
273 return ld->alloc_layout_hdr(ino, gfp_flags);
277 pnfs_free_layout_hdr(struct pnfs_layout_hdr *lo)
279 struct nfs_server *server = NFS_SERVER(lo->plh_inode);
280 struct pnfs_layoutdriver_type *ld = server->pnfs_curr_ld;
282 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
283 struct nfs_client *clp = server->nfs_client;
285 spin_lock(&clp->cl_lock);
286 list_del_rcu(&lo->plh_layouts);
287 spin_unlock(&clp->cl_lock);
289 put_cred(lo->plh_lc_cred);
290 return ld->free_layout_hdr(lo);
294 pnfs_detach_layout_hdr(struct pnfs_layout_hdr *lo)
296 struct nfs_inode *nfsi = NFS_I(lo->plh_inode);
297 dprintk("%s: freeing layout cache %p\n", __func__, lo);
299 /* Reset MDS Threshold I/O counters */
305 pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
308 unsigned long i_state;
312 inode = lo->plh_inode;
313 pnfs_layoutreturn_before_put_layout_hdr(lo);
315 if (refcount_dec_and_lock(&lo->plh_refcount, &inode->i_lock)) {
316 if (!list_empty(&lo->plh_segs))
317 WARN_ONCE(1, "NFS: BUG unfreed layout segments.\n");
318 pnfs_detach_layout_hdr(lo);
319 i_state = inode->i_state;
320 spin_unlock(&inode->i_lock);
321 pnfs_free_layout_hdr(lo);
322 /* Notify pnfs_destroy_layout_final() that we're done */
323 if (i_state & (I_FREEING | I_CLEAR))
328 static struct inode *
329 pnfs_grab_inode_layout_hdr(struct pnfs_layout_hdr *lo)
331 struct inode *inode = igrab(lo->plh_inode);
334 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
339 * Compare 2 layout stateid sequence ids, to see which is newer,
340 * taking into account wraparound issues.
342 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
344 return (s32)(s1 - s2) > 0;
347 static void pnfs_barrier_update(struct pnfs_layout_hdr *lo, u32 newseq)
349 if (pnfs_seqid_is_newer(newseq, lo->plh_barrier) || !lo->plh_barrier)
350 lo->plh_barrier = newseq;
354 pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
357 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
359 lo->plh_return_iomode = iomode;
360 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
362 * We must set lo->plh_return_seq to avoid livelocks with
363 * pnfs_layout_need_return()
366 seq = be32_to_cpu(lo->plh_stateid.seqid);
367 if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
368 lo->plh_return_seq = seq;
369 pnfs_barrier_update(lo, seq);
373 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
375 struct pnfs_layout_segment *lseg;
376 lo->plh_return_iomode = 0;
377 lo->plh_return_seq = 0;
378 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
379 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
380 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
382 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
386 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
388 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
389 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
390 smp_mb__after_atomic();
391 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
392 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
396 pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
397 struct list_head *free_me)
399 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
400 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
401 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
402 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
403 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
404 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
408 * Update the seqid of a layout stateid after receiving
409 * NFS4ERR_OLD_STATEID
411 bool nfs4_layout_refresh_old_stateid(nfs4_stateid *dst,
412 struct pnfs_layout_range *dst_range,
415 struct pnfs_layout_hdr *lo;
416 struct pnfs_layout_range range = {
417 .iomode = IOMODE_ANY,
419 .length = NFS4_MAX_UINT64,
425 spin_lock(&inode->i_lock);
426 lo = NFS_I(inode)->layout;
427 if (lo && pnfs_layout_is_valid(lo) &&
428 nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
429 /* Is our call using the most recent seqid? If so, bump it */
430 if (!nfs4_stateid_is_newer(&lo->plh_stateid, dst)) {
431 nfs4_stateid_seqid_inc(dst);
435 /* Try to update the seqid to the most recent */
436 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
438 dst->seqid = lo->plh_stateid.seqid;
444 spin_unlock(&inode->i_lock);
445 pnfs_free_lseg_list(&head);
450 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
452 * In order to continue using the pnfs_layout_hdr, a full recovery
454 * Note that caller must hold inode->i_lock.
457 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
458 struct list_head *lseg_list)
460 struct pnfs_layout_range range = {
461 .iomode = IOMODE_ANY,
463 .length = NFS4_MAX_UINT64,
465 struct pnfs_layout_segment *lseg, *next;
467 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
468 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
469 pnfs_clear_lseg_state(lseg, lseg_list);
470 pnfs_clear_layoutreturn_info(lo);
471 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
472 set_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags);
473 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
474 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
475 pnfs_clear_layoutreturn_waitbit(lo);
476 return !list_empty(&lo->plh_segs);
480 pnfs_iomode_to_fail_bit(u32 iomode)
482 return iomode == IOMODE_RW ?
483 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
487 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
489 lo->plh_retry_timestamp = jiffies;
490 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
491 refcount_inc(&lo->plh_refcount);
495 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
497 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
498 refcount_dec(&lo->plh_refcount);
502 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
504 struct inode *inode = lo->plh_inode;
505 struct pnfs_layout_range range = {
508 .length = NFS4_MAX_UINT64,
512 spin_lock(&inode->i_lock);
513 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
514 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
515 spin_unlock(&inode->i_lock);
516 pnfs_free_lseg_list(&head);
517 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
518 iomode == IOMODE_RW ? "RW" : "READ");
522 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
524 unsigned long start, end;
525 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
527 if (test_bit(fail_bit, &lo->plh_flags) == 0)
530 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
531 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
532 /* It is time to retry the failed layoutgets */
533 pnfs_layout_clear_fail_bit(lo, fail_bit);
540 pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
541 const struct pnfs_layout_range *range,
542 const nfs4_stateid *stateid)
544 INIT_LIST_HEAD(&lseg->pls_list);
545 INIT_LIST_HEAD(&lseg->pls_lc_list);
546 INIT_LIST_HEAD(&lseg->pls_commits);
547 refcount_set(&lseg->pls_refcount, 1);
548 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
549 lseg->pls_layout = lo;
550 lseg->pls_range = *range;
551 lseg->pls_seq = be32_to_cpu(stateid->seqid);
554 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
557 struct inode *inode = lseg->pls_layout->plh_inode;
558 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
563 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
564 struct pnfs_layout_segment *lseg)
566 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
567 list_del_init(&lseg->pls_list);
568 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
569 refcount_dec(&lo->plh_refcount);
570 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
572 if (list_empty(&lo->plh_segs) &&
573 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
574 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
575 if (atomic_read(&lo->plh_outstanding) == 0)
576 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
577 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
582 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
583 struct pnfs_layout_segment *lseg)
585 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
586 pnfs_layout_is_valid(lo)) {
587 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
588 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
595 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
597 struct pnfs_layout_hdr *lo;
603 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
604 refcount_read(&lseg->pls_refcount),
605 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
607 lo = lseg->pls_layout;
608 inode = lo->plh_inode;
610 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
611 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
612 spin_unlock(&inode->i_lock);
615 pnfs_get_layout_hdr(lo);
616 pnfs_layout_remove_lseg(lo, lseg);
617 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
619 spin_unlock(&inode->i_lock);
620 pnfs_free_lseg(lseg);
621 pnfs_put_layout_hdr(lo);
624 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
627 * is l2 fully contained in l1?
629 * [----------------------------------)
634 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
635 const struct pnfs_layout_range *l2)
637 u64 start1 = l1->offset;
638 u64 end1 = pnfs_end_offset(start1, l1->length);
639 u64 start2 = l2->offset;
640 u64 end2 = pnfs_end_offset(start2, l2->length);
642 return (start1 <= start2) && (end1 >= end2);
645 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
646 struct list_head *tmp_list)
648 if (!refcount_dec_and_test(&lseg->pls_refcount))
650 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
651 list_add(&lseg->pls_list, tmp_list);
655 /* Returns 1 if lseg is removed from list, 0 otherwise */
656 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
657 struct list_head *tmp_list)
661 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
662 /* Remove the reference keeping the lseg in the
663 * list. It will now be removed when all
664 * outstanding io is finished.
666 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
667 refcount_read(&lseg->pls_refcount));
668 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
675 pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
676 const struct pnfs_layout_range *recall_range)
678 return (recall_range->iomode == IOMODE_ANY ||
679 lseg_range->iomode == recall_range->iomode) &&
680 pnfs_lseg_range_intersecting(lseg_range, recall_range);
684 pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
685 const struct pnfs_layout_range *recall_range,
688 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
690 if (recall_range == NULL)
692 return pnfs_should_free_range(&lseg->pls_range, recall_range);
696 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
697 * @lo: layout header containing the lsegs
698 * @tmp_list: list head where doomed lsegs should go
699 * @recall_range: optional recall range argument to match (may be NULL)
700 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
702 * Walk the list of lsegs in the layout header, and tear down any that should
703 * be destroyed. If "recall_range" is specified then the segment must match
704 * that range. If "seq" is non-zero, then only match segments that were handed
705 * out at or before that sequence.
707 * Returns number of matching invalid lsegs remaining in list after scanning
708 * it and purging them.
711 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
712 struct list_head *tmp_list,
713 const struct pnfs_layout_range *recall_range,
716 struct pnfs_layout_segment *lseg, *next;
719 dprintk("%s:Begin lo %p\n", __func__, lo);
721 if (list_empty(&lo->plh_segs))
723 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
724 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
725 dprintk("%s: freeing lseg %p iomode %d seq %u "
726 "offset %llu length %llu\n", __func__,
727 lseg, lseg->pls_range.iomode, lseg->pls_seq,
728 lseg->pls_range.offset, lseg->pls_range.length);
729 if (!mark_lseg_invalid(lseg, tmp_list))
732 dprintk("%s:Return %i\n", __func__, remaining);
737 pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
738 struct list_head *free_me,
739 const struct pnfs_layout_range *range,
742 struct pnfs_layout_segment *lseg, *next;
744 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
745 if (pnfs_match_lseg_recall(lseg, range, seq))
746 list_move_tail(&lseg->pls_list, free_me);
750 /* note free_me must contain lsegs from a single layout_hdr */
752 pnfs_free_lseg_list(struct list_head *free_me)
754 struct pnfs_layout_segment *lseg, *tmp;
756 if (list_empty(free_me))
759 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
760 list_del(&lseg->pls_list);
761 pnfs_free_lseg(lseg);
765 static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
767 struct pnfs_layout_hdr *lo;
770 spin_lock(&nfsi->vfs_inode.i_lock);
773 pnfs_get_layout_hdr(lo);
774 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
775 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
776 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
777 spin_unlock(&nfsi->vfs_inode.i_lock);
778 pnfs_free_lseg_list(&tmp_list);
779 nfs_commit_inode(&nfsi->vfs_inode, 0);
780 pnfs_put_layout_hdr(lo);
782 spin_unlock(&nfsi->vfs_inode.i_lock);
786 void pnfs_destroy_layout(struct nfs_inode *nfsi)
788 __pnfs_destroy_layout(nfsi);
790 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
792 static bool pnfs_layout_removed(struct nfs_inode *nfsi,
793 struct pnfs_layout_hdr *lo)
797 spin_lock(&nfsi->vfs_inode.i_lock);
798 ret = nfsi->layout != lo;
799 spin_unlock(&nfsi->vfs_inode.i_lock);
803 void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
805 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
808 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
812 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
813 struct list_head *layout_list)
815 struct pnfs_layout_hdr *lo;
818 spin_lock(&inode->i_lock);
819 lo = NFS_I(inode)->layout;
820 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
821 pnfs_get_layout_hdr(lo);
822 list_add(&lo->plh_bulk_destroy, layout_list);
825 spin_unlock(&inode->i_lock);
829 /* Caller must hold rcu_read_lock and clp->cl_lock */
831 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
832 struct nfs_server *server,
833 struct list_head *layout_list)
834 __must_hold(&clp->cl_lock)
837 struct pnfs_layout_hdr *lo, *next;
840 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
841 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
842 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
843 !list_empty(&lo->plh_bulk_destroy))
845 /* If the sb is being destroyed, just bail */
846 if (!nfs_sb_active(server->super))
848 inode = pnfs_grab_inode_layout_hdr(lo);
850 if (test_and_clear_bit(NFS_LAYOUT_HASHED, &lo->plh_flags))
851 list_del_rcu(&lo->plh_layouts);
852 if (pnfs_layout_add_bulk_destroy_list(inode,
856 spin_unlock(&clp->cl_lock);
860 spin_unlock(&clp->cl_lock);
862 nfs_sb_deactive(server->super);
863 spin_lock(&clp->cl_lock);
871 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
874 struct pnfs_layout_hdr *lo;
876 LIST_HEAD(lseg_list);
879 while (!list_empty(layout_list)) {
880 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
882 dprintk("%s freeing layout for inode %lu\n", __func__,
883 lo->plh_inode->i_ino);
884 inode = lo->plh_inode;
886 pnfs_layoutcommit_inode(inode, false);
888 spin_lock(&inode->i_lock);
889 list_del_init(&lo->plh_bulk_destroy);
890 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
892 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
895 spin_unlock(&inode->i_lock);
896 pnfs_free_lseg_list(&lseg_list);
897 /* Free all lsegs that are attached to commit buckets */
898 nfs_commit_inode(inode, 0);
899 pnfs_put_layout_hdr(lo);
900 nfs_iput_and_deactive(inode);
906 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
907 struct nfs_fsid *fsid,
910 struct nfs_server *server;
911 LIST_HEAD(layout_list);
913 spin_lock(&clp->cl_lock);
916 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
917 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
919 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
925 spin_unlock(&clp->cl_lock);
927 if (list_empty(&layout_list))
929 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
933 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
936 struct nfs_server *server;
937 LIST_HEAD(layout_list);
939 spin_lock(&clp->cl_lock);
942 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
943 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
949 spin_unlock(&clp->cl_lock);
951 if (list_empty(&layout_list))
953 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
957 * Called by the state manager to remove all layouts established under an
961 pnfs_destroy_all_layouts(struct nfs_client *clp)
963 nfs4_deviceid_mark_client_invalid(clp);
964 nfs4_deviceid_purge_client(clp);
966 pnfs_destroy_layouts_byclid(clp, false);
970 pnfs_set_layout_cred(struct pnfs_layout_hdr *lo, const struct cred *cred)
972 const struct cred *old;
974 if (cred && cred_fscmp(lo->plh_lc_cred, cred) != 0) {
975 old = xchg(&lo->plh_lc_cred, get_cred(cred));
980 /* update lo->plh_stateid with new if is more recent */
982 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
983 const struct cred *cred, bool update_barrier)
985 u32 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
986 u32 newseq = be32_to_cpu(new->seqid);
988 if (!pnfs_layout_is_valid(lo)) {
989 pnfs_set_layout_cred(lo, cred);
990 nfs4_stateid_copy(&lo->plh_stateid, new);
991 lo->plh_barrier = newseq;
992 pnfs_clear_layoutreturn_info(lo);
993 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
997 if (pnfs_seqid_is_newer(newseq, oldseq))
998 nfs4_stateid_copy(&lo->plh_stateid, new);
1000 if (update_barrier) {
1001 pnfs_barrier_update(lo, newseq);
1005 * Because of wraparound, we want to keep the barrier
1006 * "close" to the current seqids. We really only want to
1007 * get here from a layoutget call.
1009 if (atomic_read(&lo->plh_outstanding) == 1)
1010 pnfs_barrier_update(lo, be32_to_cpu(lo->plh_stateid.seqid));
1014 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
1015 const nfs4_stateid *stateid)
1017 u32 seqid = be32_to_cpu(stateid->seqid);
1019 return lo->plh_barrier && pnfs_seqid_is_newer(lo->plh_barrier, seqid);
1022 /* lget is set to 1 if called from inside send_layoutget call chain */
1024 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
1026 return lo->plh_block_lgets ||
1027 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
1030 static struct nfs_server *
1031 pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
1033 struct nfs_server *server;
1036 server = NFS_SERVER(inode);
1038 struct dentry *parent_dir = dget_parent(ctx->dentry);
1039 server = NFS_SERVER(parent_dir->d_inode);
1045 static void nfs4_free_pages(struct page **pages, size_t size)
1052 for (i = 0; i < size; i++) {
1055 __free_page(pages[i]);
1060 static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1062 struct page **pages;
1065 pages = kmalloc_array(size, sizeof(struct page *), gfp_flags);
1067 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1071 for (i = 0; i < size; i++) {
1072 pages[i] = alloc_page(gfp_flags);
1074 dprintk("%s: failed to allocate page\n", __func__);
1075 nfs4_free_pages(pages, i);
1083 static struct nfs4_layoutget *
1084 pnfs_alloc_init_layoutget_args(struct inode *ino,
1085 struct nfs_open_context *ctx,
1086 const nfs4_stateid *stateid,
1087 const struct pnfs_layout_range *range,
1090 struct nfs_server *server = pnfs_find_server(ino, ctx);
1091 size_t max_reply_sz = server->pnfs_curr_ld->max_layoutget_response;
1092 size_t max_pages = max_response_pages(server);
1093 struct nfs4_layoutget *lgp;
1095 dprintk("--> %s\n", __func__);
1097 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1102 size_t npages = (max_reply_sz + PAGE_SIZE - 1) >> PAGE_SHIFT;
1103 if (npages < max_pages)
1107 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1108 if (!lgp->args.layout.pages) {
1112 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1113 lgp->res.layoutp = &lgp->args.layout;
1115 /* Don't confuse uninitialised result and success */
1116 lgp->res.status = -NFS4ERR_DELAY;
1118 lgp->args.minlength = PAGE_SIZE;
1119 if (lgp->args.minlength > range->length)
1120 lgp->args.minlength = range->length;
1122 loff_t i_size = i_size_read(ino);
1124 if (range->iomode == IOMODE_READ) {
1125 if (range->offset >= i_size)
1126 lgp->args.minlength = 0;
1127 else if (i_size - range->offset < lgp->args.minlength)
1128 lgp->args.minlength = i_size - range->offset;
1131 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1132 pnfs_copy_range(&lgp->args.range, range);
1133 lgp->args.type = server->pnfs_curr_ld->id;
1134 lgp->args.inode = ino;
1135 lgp->args.ctx = get_nfs_open_context(ctx);
1136 nfs4_stateid_copy(&lgp->args.stateid, stateid);
1137 lgp->gfp_flags = gfp_flags;
1138 lgp->cred = ctx->cred;
1142 void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1144 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1146 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1147 if (lgp->args.inode)
1148 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1149 put_nfs_open_context(lgp->args.ctx);
1153 static void pnfs_clear_layoutcommit(struct inode *inode,
1154 struct list_head *head)
1156 struct nfs_inode *nfsi = NFS_I(inode);
1157 struct pnfs_layout_segment *lseg, *tmp;
1159 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1161 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1162 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1164 pnfs_lseg_dec_and_remove_zero(lseg, head);
1168 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1169 const nfs4_stateid *arg_stateid,
1170 const struct pnfs_layout_range *range,
1171 const nfs4_stateid *stateid)
1173 struct inode *inode = lo->plh_inode;
1176 spin_lock(&inode->i_lock);
1177 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1178 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1181 u32 seq = be32_to_cpu(arg_stateid->seqid);
1183 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1184 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1185 pnfs_set_layout_stateid(lo, stateid, NULL, true);
1187 pnfs_mark_layout_stateid_invalid(lo, &freeme);
1189 pnfs_clear_layoutreturn_waitbit(lo);
1190 spin_unlock(&inode->i_lock);
1191 pnfs_free_lseg_list(&freeme);
1196 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1197 nfs4_stateid *stateid,
1198 const struct cred **cred,
1199 enum pnfs_iomode *iomode)
1201 /* Serialise LAYOUTGET/LAYOUTRETURN */
1202 if (atomic_read(&lo->plh_outstanding) != 0)
1204 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1206 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1207 pnfs_get_layout_hdr(lo);
1208 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1209 *cred = get_cred(lo->plh_lc_cred);
1210 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1211 if (lo->plh_return_seq != 0)
1212 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1214 *iomode = lo->plh_return_iomode;
1215 pnfs_clear_layoutreturn_info(lo);
1216 } else if (iomode != NULL)
1217 *iomode = IOMODE_ANY;
1218 pnfs_barrier_update(lo, be32_to_cpu(stateid->seqid));
1223 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1224 struct pnfs_layout_hdr *lo,
1225 const nfs4_stateid *stateid,
1226 enum pnfs_iomode iomode)
1228 struct inode *inode = lo->plh_inode;
1230 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1231 args->inode = inode;
1232 args->range.iomode = iomode;
1233 args->range.offset = 0;
1234 args->range.length = NFS4_MAX_UINT64;
1236 nfs4_stateid_copy(&args->stateid, stateid);
1240 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo,
1241 const nfs4_stateid *stateid,
1242 const struct cred **pcred,
1243 enum pnfs_iomode iomode,
1246 struct inode *ino = lo->plh_inode;
1247 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1248 struct nfs4_layoutreturn *lrp;
1249 const struct cred *cred = *pcred;
1253 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
1254 if (unlikely(lrp == NULL)) {
1256 spin_lock(&ino->i_lock);
1257 pnfs_clear_layoutreturn_waitbit(lo);
1258 spin_unlock(&ino->i_lock);
1260 pnfs_put_layout_hdr(lo);
1264 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1265 lrp->args.ld_private = &lrp->ld_private;
1266 lrp->clp = NFS_SERVER(ino)->nfs_client;
1268 if (ld->prepare_layoutreturn)
1269 ld->prepare_layoutreturn(&lrp->args);
1271 status = nfs4_proc_layoutreturn(lrp, sync);
1273 dprintk("<-- %s status: %d\n", __func__, status);
1278 pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1279 enum pnfs_iomode iomode,
1282 struct pnfs_layout_range recall_range = {
1283 .length = NFS4_MAX_UINT64,
1286 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1287 &recall_range, seq) != -EBUSY;
1290 /* Return true if layoutreturn is needed */
1292 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1294 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1296 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1297 lo->plh_return_seq);
1300 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1302 struct inode *inode= lo->plh_inode;
1304 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1306 spin_lock(&inode->i_lock);
1307 if (pnfs_layout_need_return(lo)) {
1308 const struct cred *cred;
1309 nfs4_stateid stateid;
1310 enum pnfs_iomode iomode;
1313 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
1314 spin_unlock(&inode->i_lock);
1316 /* Send an async layoutreturn so we dont deadlock */
1317 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
1320 spin_unlock(&inode->i_lock);
1324 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1325 * when the layout segment list is empty.
1327 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1328 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1329 * deviceid is marked invalid.
1332 _pnfs_return_layout(struct inode *ino)
1334 struct pnfs_layout_hdr *lo = NULL;
1335 struct nfs_inode *nfsi = NFS_I(ino);
1336 struct pnfs_layout_range range = {
1337 .iomode = IOMODE_ANY,
1339 .length = NFS4_MAX_UINT64,
1341 LIST_HEAD(tmp_list);
1342 const struct cred *cred;
1343 nfs4_stateid stateid;
1345 bool send, valid_layout;
1347 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1349 spin_lock(&ino->i_lock);
1352 spin_unlock(&ino->i_lock);
1353 dprintk("NFS: %s no layout to return\n", __func__);
1356 /* Reference matched in nfs4_layoutreturn_release */
1357 pnfs_get_layout_hdr(lo);
1358 /* Is there an outstanding layoutreturn ? */
1359 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1360 spin_unlock(&ino->i_lock);
1361 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1362 TASK_UNINTERRUPTIBLE))
1363 goto out_put_layout_hdr;
1364 spin_lock(&ino->i_lock);
1366 valid_layout = pnfs_layout_is_valid(lo);
1367 pnfs_clear_layoutcommit(ino, &tmp_list);
1368 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1370 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
1371 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1373 /* Don't send a LAYOUTRETURN if list was initially empty */
1374 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1376 spin_unlock(&ino->i_lock);
1377 dprintk("NFS: %s no layout segments to return\n", __func__);
1378 goto out_wait_layoutreturn;
1381 send = pnfs_prepare_layoutreturn(lo, &stateid, &cred, NULL);
1382 spin_unlock(&ino->i_lock);
1384 status = pnfs_send_layoutreturn(lo, &stateid, &cred, IOMODE_ANY, true);
1385 out_wait_layoutreturn:
1386 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN, TASK_UNINTERRUPTIBLE);
1388 pnfs_free_lseg_list(&tmp_list);
1389 pnfs_put_layout_hdr(lo);
1391 dprintk("<-- %s status: %d\n", __func__, status);
1396 pnfs_commit_and_return_layout(struct inode *inode)
1398 struct pnfs_layout_hdr *lo;
1401 spin_lock(&inode->i_lock);
1402 lo = NFS_I(inode)->layout;
1404 spin_unlock(&inode->i_lock);
1407 pnfs_get_layout_hdr(lo);
1408 /* Block new layoutgets and read/write to ds */
1409 lo->plh_block_lgets++;
1410 spin_unlock(&inode->i_lock);
1411 filemap_fdatawait(inode->i_mapping);
1412 ret = pnfs_layoutcommit_inode(inode, true);
1414 ret = _pnfs_return_layout(inode);
1415 spin_lock(&inode->i_lock);
1416 lo->plh_block_lgets--;
1417 spin_unlock(&inode->i_lock);
1418 pnfs_put_layout_hdr(lo);
1422 bool pnfs_roc(struct inode *ino,
1423 struct nfs4_layoutreturn_args *args,
1424 struct nfs4_layoutreturn_res *res,
1425 const struct cred *cred)
1427 struct nfs_inode *nfsi = NFS_I(ino);
1428 struct nfs_open_context *ctx;
1429 struct nfs4_state *state;
1430 struct pnfs_layout_hdr *lo;
1431 struct pnfs_layout_segment *lseg, *next;
1432 const struct cred *lc_cred;
1433 nfs4_stateid stateid;
1434 enum pnfs_iomode iomode = 0;
1435 bool layoutreturn = false, roc = false;
1436 bool skip_read = false;
1438 if (!nfs_have_layout(ino))
1442 spin_lock(&ino->i_lock);
1444 if (!lo || !pnfs_layout_is_valid(lo) ||
1445 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1449 pnfs_get_layout_hdr(lo);
1450 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1451 spin_unlock(&ino->i_lock);
1453 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1454 TASK_UNINTERRUPTIBLE);
1455 pnfs_put_layout_hdr(lo);
1459 /* no roc if we hold a delegation */
1460 if (nfs4_check_delegation(ino, FMODE_READ)) {
1461 if (nfs4_check_delegation(ino, FMODE_WRITE))
1466 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1470 /* Don't return layout if there is open file state */
1471 if (state->state & FMODE_WRITE)
1473 if (state->state & FMODE_READ)
1478 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1479 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1481 /* If we are sending layoutreturn, invalidate all valid lsegs */
1482 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1485 * Note: mark lseg for return so pnfs_layout_remove_lseg
1486 * doesn't invalidate the layout for us.
1488 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1489 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1491 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1494 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1497 /* ROC in two conditions:
1498 * 1. there are ROC lsegs
1499 * 2. we don't send layoutreturn
1501 /* lo ref dropped in pnfs_roc_release() */
1502 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &lc_cred, &iomode);
1503 /* If the creds don't match, we can't compound the layoutreturn */
1504 if (!layoutreturn || cred_fscmp(cred, lc_cred) != 0)
1508 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1509 res->lrs_present = 0;
1510 layoutreturn = false;
1514 spin_unlock(&ino->i_lock);
1516 pnfs_layoutcommit_inode(ino, true);
1518 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1519 if (ld->prepare_layoutreturn)
1520 ld->prepare_layoutreturn(args);
1521 pnfs_put_layout_hdr(lo);
1525 pnfs_send_layoutreturn(lo, &stateid, &lc_cred, iomode, true);
1526 pnfs_put_layout_hdr(lo);
1530 int pnfs_roc_done(struct rpc_task *task, struct nfs4_layoutreturn_args **argpp,
1531 struct nfs4_layoutreturn_res **respp, int *ret)
1533 struct nfs4_layoutreturn_args *arg = *argpp;
1534 int retval = -EAGAIN;
1538 /* Handle Layoutreturn errors */
1543 case -NFS4ERR_NOMATCHING_LAYOUT:
1544 /* Was there an RPC level error? If not, retry */
1545 if (task->tk_rpc_status == 0)
1547 /* If the call was not sent, let caller handle it */
1548 if (!RPC_WAS_SENT(task))
1551 * Otherwise, assume the call succeeded and
1552 * that we need to release the layout
1555 (*respp)->lrs_present = 0;
1558 case -NFS4ERR_DELAY:
1559 /* Let the caller handle the retry */
1560 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1562 case -NFS4ERR_OLD_STATEID:
1563 if (!nfs4_layout_refresh_old_stateid(&arg->stateid,
1564 &arg->range, arg->inode))
1566 *ret = -NFS4ERR_NOMATCHING_LAYOUT;
1574 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1575 struct nfs4_layoutreturn_res *res,
1578 struct pnfs_layout_hdr *lo = args->layout;
1579 struct inode *inode = args->inode;
1580 const nfs4_stateid *arg_stateid = NULL;
1581 const nfs4_stateid *res_stateid = NULL;
1582 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1585 case -NFS4ERR_NOMATCHING_LAYOUT:
1586 spin_lock(&inode->i_lock);
1587 if (pnfs_layout_is_valid(lo) &&
1588 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1589 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
1590 spin_unlock(&inode->i_lock);
1593 if (res->lrs_present)
1594 res_stateid = &res->stateid;
1597 arg_stateid = &args->stateid;
1599 trace_nfs4_layoutreturn_on_close(args->inode, &args->stateid, ret);
1600 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1602 if (ld_private && ld_private->ops && ld_private->ops->free)
1603 ld_private->ops->free(ld_private);
1604 pnfs_put_layout_hdr(lo);
1607 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1609 struct nfs_inode *nfsi = NFS_I(ino);
1610 struct pnfs_layout_hdr *lo;
1613 /* we might not have grabbed lo reference. so need to check under
1615 spin_lock(&ino->i_lock);
1617 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1618 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1621 spin_unlock(&ino->i_lock);
1626 * Compare two layout segments for sorting into layout cache.
1627 * We want to preferentially return RW over RO layouts, so ensure those
1631 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1632 const struct pnfs_layout_range *l2)
1636 /* high offset > low offset */
1637 d = l1->offset - l2->offset;
1641 /* short length > long length */
1642 d = l2->length - l1->length;
1646 /* read > read/write */
1647 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1651 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1652 const struct pnfs_layout_range *l2)
1654 return pnfs_lseg_range_cmp(l1, l2) > 0;
1658 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1659 struct pnfs_layout_segment *old)
1665 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1666 struct pnfs_layout_segment *lseg,
1667 bool (*is_after)(const struct pnfs_layout_range *,
1668 const struct pnfs_layout_range *),
1669 bool (*do_merge)(struct pnfs_layout_segment *,
1670 struct pnfs_layout_segment *),
1671 struct list_head *free_me)
1673 struct pnfs_layout_segment *lp, *tmp;
1675 dprintk("%s:Begin\n", __func__);
1677 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1678 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1680 if (do_merge(lseg, lp)) {
1681 mark_lseg_invalid(lp, free_me);
1684 if (is_after(&lseg->pls_range, &lp->pls_range))
1686 list_add_tail(&lseg->pls_list, &lp->pls_list);
1687 dprintk("%s: inserted lseg %p "
1688 "iomode %d offset %llu length %llu before "
1689 "lp %p iomode %d offset %llu length %llu\n",
1690 __func__, lseg, lseg->pls_range.iomode,
1691 lseg->pls_range.offset, lseg->pls_range.length,
1692 lp, lp->pls_range.iomode, lp->pls_range.offset,
1693 lp->pls_range.length);
1696 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1697 dprintk("%s: inserted lseg %p "
1698 "iomode %d offset %llu length %llu at tail\n",
1699 __func__, lseg, lseg->pls_range.iomode,
1700 lseg->pls_range.offset, lseg->pls_range.length);
1702 pnfs_get_layout_hdr(lo);
1704 dprintk("%s:Return\n", __func__);
1706 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1709 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1710 struct pnfs_layout_segment *lseg,
1711 struct list_head *free_me)
1713 struct inode *inode = lo->plh_inode;
1714 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1716 if (ld->add_lseg != NULL)
1717 ld->add_lseg(lo, lseg, free_me);
1719 pnfs_generic_layout_insert_lseg(lo, lseg,
1720 pnfs_lseg_range_is_after,
1725 static struct pnfs_layout_hdr *
1726 alloc_init_layout_hdr(struct inode *ino,
1727 struct nfs_open_context *ctx,
1730 struct pnfs_layout_hdr *lo;
1732 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1735 refcount_set(&lo->plh_refcount, 1);
1736 INIT_LIST_HEAD(&lo->plh_layouts);
1737 INIT_LIST_HEAD(&lo->plh_segs);
1738 INIT_LIST_HEAD(&lo->plh_return_segs);
1739 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1740 lo->plh_inode = ino;
1741 lo->plh_lc_cred = get_cred(ctx->cred);
1742 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1746 static struct pnfs_layout_hdr *
1747 pnfs_find_alloc_layout(struct inode *ino,
1748 struct nfs_open_context *ctx,
1750 __releases(&ino->i_lock)
1751 __acquires(&ino->i_lock)
1753 struct nfs_inode *nfsi = NFS_I(ino);
1754 struct pnfs_layout_hdr *new = NULL;
1756 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1758 if (nfsi->layout != NULL)
1760 spin_unlock(&ino->i_lock);
1761 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1762 spin_lock(&ino->i_lock);
1764 if (likely(nfsi->layout == NULL)) { /* Won the race? */
1767 } else if (new != NULL)
1768 pnfs_free_layout_hdr(new);
1770 pnfs_get_layout_hdr(nfsi->layout);
1771 return nfsi->layout;
1775 * iomode matching rules:
1776 * iomode lseg strict match
1778 * ----- ----- ------ -----
1783 * READ READ N/A true
1784 * READ RW true false
1785 * READ RW false true
1788 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1789 const struct pnfs_layout_range *range,
1792 struct pnfs_layout_range range1;
1794 if ((range->iomode == IOMODE_RW &&
1795 ls_range->iomode != IOMODE_RW) ||
1796 (range->iomode != ls_range->iomode &&
1798 !pnfs_lseg_range_intersecting(ls_range, range))
1801 /* range1 covers only the first byte in the range */
1804 return pnfs_lseg_range_contained(ls_range, &range1);
1808 * lookup range in layout
1810 static struct pnfs_layout_segment *
1811 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1812 struct pnfs_layout_range *range,
1815 struct pnfs_layout_segment *lseg, *ret = NULL;
1817 dprintk("%s:Begin\n", __func__);
1819 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1820 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1821 pnfs_lseg_range_match(&lseg->pls_range, range,
1823 ret = pnfs_get_lseg(lseg);
1828 dprintk("%s:Return lseg %p ref %d\n",
1829 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1834 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1835 * to the MDS or over pNFS
1837 * The nfs_inode read_io and write_io fields are cumulative counters reset
1838 * when there are no layout segments. Note that in pnfs_update_layout iomode
1839 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1842 * A return of true means use MDS I/O.
1845 * If a file's size is smaller than the file size threshold, data accesses
1846 * SHOULD be sent to the metadata server. If an I/O request has a length that
1847 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1848 * server. If both file size and I/O size are provided, the client SHOULD
1849 * reach or exceed both thresholds before sending its read or write
1850 * requests to the data server.
1852 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1853 struct inode *ino, int iomode)
1855 struct nfs4_threshold *t = ctx->mdsthreshold;
1856 struct nfs_inode *nfsi = NFS_I(ino);
1857 loff_t fsize = i_size_read(ino);
1858 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1863 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1864 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1868 if (t->bm & THRESHOLD_RD) {
1869 dprintk("%s fsize %llu\n", __func__, fsize);
1871 if (fsize < t->rd_sz)
1874 if (t->bm & THRESHOLD_RD_IO) {
1875 dprintk("%s nfsi->read_io %llu\n", __func__,
1878 if (nfsi->read_io < t->rd_io_sz)
1883 if (t->bm & THRESHOLD_WR) {
1884 dprintk("%s fsize %llu\n", __func__, fsize);
1886 if (fsize < t->wr_sz)
1889 if (t->bm & THRESHOLD_WR_IO) {
1890 dprintk("%s nfsi->write_io %llu\n", __func__,
1893 if (nfsi->write_io < t->wr_io_sz)
1898 if (size_set && io_set) {
1901 } else if (size || io)
1904 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1908 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1911 * send layoutcommit as it can hold up layoutreturn due to lseg
1914 pnfs_layoutcommit_inode(lo->plh_inode, false);
1915 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1916 nfs_wait_bit_killable,
1920 static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1922 atomic_inc(&lo->plh_outstanding);
1925 static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1927 if (atomic_dec_and_test(&lo->plh_outstanding) &&
1928 test_and_clear_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags))
1929 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN);
1932 static bool pnfs_is_first_layoutget(struct pnfs_layout_hdr *lo)
1934 return test_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags);
1937 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1939 unsigned long *bitlock = &lo->plh_flags;
1941 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1942 smp_mb__after_atomic();
1943 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1946 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1947 struct nfs_server *server)
1949 if (!test_and_set_bit(NFS_LAYOUT_HASHED, &lo->plh_flags)) {
1950 struct nfs_client *clp = server->nfs_client;
1952 /* The lo must be on the clp list if there is any
1953 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1955 spin_lock(&clp->cl_lock);
1956 list_add_tail_rcu(&lo->plh_layouts, &server->layouts);
1957 spin_unlock(&clp->cl_lock);
1962 * Layout segment is retreived from the server if not cached.
1963 * The appropriate layout segment is referenced and returned to the caller.
1965 struct pnfs_layout_segment *
1966 pnfs_update_layout(struct inode *ino,
1967 struct nfs_open_context *ctx,
1970 enum pnfs_iomode iomode,
1974 struct pnfs_layout_range arg = {
1980 struct nfs_server *server = NFS_SERVER(ino);
1981 struct nfs_client *clp = server->nfs_client;
1982 struct pnfs_layout_hdr *lo = NULL;
1983 struct pnfs_layout_segment *lseg = NULL;
1984 struct nfs4_layoutget *lgp;
1985 nfs4_stateid stateid;
1987 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1990 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1991 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1992 PNFS_UPDATE_LAYOUT_NO_PNFS);
1996 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1997 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1998 PNFS_UPDATE_LAYOUT_MDSTHRESH);
2003 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
2007 spin_lock(&ino->i_lock);
2008 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
2010 spin_unlock(&ino->i_lock);
2011 lseg = ERR_PTR(-ENOMEM);
2012 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2013 PNFS_UPDATE_LAYOUT_NOMEM);
2017 /* Do we even need to bother with this? */
2018 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
2019 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2020 PNFS_UPDATE_LAYOUT_BULK_RECALL);
2021 dprintk("%s matches recall, use MDS\n", __func__);
2025 /* if LAYOUTGET already failed once we don't try again */
2026 if (pnfs_layout_io_test_failed(lo, iomode)) {
2027 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2028 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
2033 * If the layout segment list is empty, but there are outstanding
2034 * layoutget calls, then they might be subject to a layoutrecall.
2036 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2037 atomic_read(&lo->plh_outstanding) != 0) {
2038 spin_unlock(&ino->i_lock);
2039 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags, NFS_LAYOUT_DRAIN,
2042 goto out_put_layout_hdr;
2043 pnfs_put_layout_hdr(lo);
2048 * Because we free lsegs when sending LAYOUTRETURN, we need to wait
2051 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
2052 spin_unlock(&ino->i_lock);
2053 dprintk("%s wait for layoutreturn\n", __func__);
2054 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
2055 if (!IS_ERR(lseg)) {
2056 pnfs_put_layout_hdr(lo);
2057 dprintk("%s retrying\n", __func__);
2058 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
2060 PNFS_UPDATE_LAYOUT_RETRY);
2063 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2064 PNFS_UPDATE_LAYOUT_RETURN);
2065 goto out_put_layout_hdr;
2068 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
2070 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2071 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
2076 * Choose a stateid for the LAYOUTGET. If we don't have a layout
2077 * stateid, or it has been invalidated, then we must use the open
2080 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
2084 * The first layoutget for the file. Need to serialize per
2085 * RFC 5661 Errata 3208.
2087 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
2089 spin_unlock(&ino->i_lock);
2090 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
2091 NFS_LAYOUT_FIRST_LAYOUTGET,
2094 goto out_put_layout_hdr;
2095 pnfs_put_layout_hdr(lo);
2096 dprintk("%s retrying\n", __func__);
2100 spin_unlock(&ino->i_lock);
2102 status = nfs4_select_rw_stateid(ctx->state,
2103 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
2104 NULL, &stateid, NULL);
2106 lseg = ERR_PTR(status);
2107 trace_pnfs_update_layout(ino, pos, count,
2109 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
2110 nfs4_schedule_stateid_recovery(server, ctx->state);
2111 pnfs_clear_first_layoutget(lo);
2112 pnfs_put_layout_hdr(lo);
2115 spin_lock(&ino->i_lock);
2117 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
2120 if (pnfs_layoutgets_blocked(lo)) {
2121 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2122 PNFS_UPDATE_LAYOUT_BLOCKED);
2125 nfs_layoutget_begin(lo);
2126 spin_unlock(&ino->i_lock);
2128 _add_to_server_list(lo, server);
2130 pg_offset = arg.offset & ~PAGE_MASK;
2132 arg.offset -= pg_offset;
2133 arg.length += pg_offset;
2135 if (arg.length != NFS4_MAX_UINT64)
2136 arg.length = PAGE_ALIGN(arg.length);
2138 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
2140 lseg = ERR_PTR(-ENOMEM);
2141 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2142 PNFS_UPDATE_LAYOUT_NOMEM);
2143 nfs_layoutget_end(lo);
2144 goto out_put_layout_hdr;
2147 lseg = nfs4_proc_layoutget(lgp, &timeout);
2148 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2149 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
2150 nfs_layoutget_end(lo);
2152 switch(PTR_ERR(lseg)) {
2154 if (time_after(jiffies, giveup))
2157 case -ERECALLCONFLICT:
2161 /* The server returned NFS4ERR_LAYOUTUNAVAILABLE */
2162 pnfs_layout_set_fail_bit(
2163 lo, pnfs_iomode_to_fail_bit(iomode));
2165 goto out_put_layout_hdr;
2167 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2168 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2171 goto out_put_layout_hdr;
2175 pnfs_clear_first_layoutget(lo);
2176 trace_pnfs_update_layout(ino, pos, count,
2177 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2178 pnfs_put_layout_hdr(lo);
2182 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2187 pnfs_clear_first_layoutget(lo);
2188 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2189 PNFS_UPDATE_LAYOUT_EXIT);
2190 pnfs_put_layout_hdr(lo);
2192 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2193 "(%s, offset: %llu, length: %llu)\n",
2194 __func__, ino->i_sb->s_id,
2195 (unsigned long long)NFS_FILEID(ino),
2196 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2197 iomode==IOMODE_RW ? "read/write" : "read-only",
2198 (unsigned long long)pos,
2199 (unsigned long long)count);
2202 spin_unlock(&ino->i_lock);
2203 goto out_put_layout_hdr;
2205 EXPORT_SYMBOL_GPL(pnfs_update_layout);
2208 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2210 switch (range->iomode) {
2217 if (range->offset == NFS4_MAX_UINT64)
2219 if (range->length == 0)
2221 if (range->length != NFS4_MAX_UINT64 &&
2222 range->length > NFS4_MAX_UINT64 - range->offset)
2227 static struct pnfs_layout_hdr *
2228 _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2230 struct pnfs_layout_hdr *lo;
2232 spin_lock(&ino->i_lock);
2233 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2236 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2238 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2240 if (pnfs_layoutgets_blocked(lo))
2242 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2244 nfs_layoutget_begin(lo);
2245 spin_unlock(&ino->i_lock);
2246 _add_to_server_list(lo, NFS_SERVER(ino));
2250 spin_unlock(&ino->i_lock);
2251 pnfs_put_layout_hdr(lo);
2255 static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2256 struct nfs_open_context *ctx)
2258 struct inode *ino = data->dentry->d_inode;
2259 struct pnfs_layout_range rng = {
2260 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2261 IOMODE_RW: IOMODE_READ,
2263 .length = NFS4_MAX_UINT64,
2265 struct nfs4_layoutget *lgp;
2266 struct pnfs_layout_hdr *lo;
2268 /* Heuristic: don't send layoutget if we have cached data */
2269 if (rng.iomode == IOMODE_READ &&
2270 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2273 lo = _pnfs_grab_empty_layout(ino, ctx);
2276 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, ¤t_stateid,
2279 pnfs_clear_first_layoutget(lo);
2280 nfs_layoutget_end(lo);
2281 pnfs_put_layout_hdr(lo);
2285 data->o_arg.lg_args = &lgp->args;
2286 data->o_res.lg_res = &lgp->res;
2289 static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2290 struct nfs_open_context *ctx)
2292 struct pnfs_layout_range rng = {
2293 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2294 IOMODE_RW: IOMODE_READ,
2296 .length = NFS4_MAX_UINT64,
2298 struct nfs4_layoutget *lgp;
2300 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, ¤t_stateid,
2305 data->o_arg.lg_args = &lgp->args;
2306 data->o_res.lg_res = &lgp->res;
2309 void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2310 struct nfs_open_context *ctx)
2312 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2314 if (!(pnfs_enabled_sb(server) &&
2315 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2317 /* Could check on max_ops, but currently hardcoded high enough */
2318 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2321 _lgopen_prepare_attached(data, ctx);
2323 _lgopen_prepare_floating(data, ctx);
2326 void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2327 struct nfs_open_context *ctx)
2329 struct pnfs_layout_hdr *lo;
2330 struct pnfs_layout_segment *lseg;
2331 struct nfs_server *srv = NFS_SERVER(ino);
2336 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2337 if (lgp->res.status) {
2338 switch (lgp->res.status) {
2342 * Halt lgopen attempts if the server doesn't recognise
2343 * the "current stateid" value, the layout type, or the
2344 * layoutget operation as being valid.
2345 * Also if it complains about too many ops in the compound
2346 * or of the request/reply being too big.
2348 case -NFS4ERR_BAD_STATEID:
2349 case -NFS4ERR_NOTSUPP:
2350 case -NFS4ERR_REP_TOO_BIG:
2351 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2352 case -NFS4ERR_REQ_TOO_BIG:
2353 case -NFS4ERR_TOO_MANY_OPS:
2354 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2355 srv->caps &= ~NFS_CAP_LGOPEN;
2359 if (!lgp->args.inode) {
2360 lo = _pnfs_grab_empty_layout(ino, ctx);
2363 lgp->args.inode = ino;
2365 lo = NFS_I(lgp->args.inode)->layout;
2367 lseg = pnfs_layout_process(lgp);
2368 if (!IS_ERR(lseg)) {
2369 iomode = lgp->args.range.iomode;
2370 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2371 pnfs_put_lseg(lseg);
2375 void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2378 struct inode *inode = lgp->args.inode;
2380 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2381 pnfs_clear_first_layoutget(lo);
2382 nfs_layoutget_end(lo);
2384 pnfs_layoutget_free(lgp);
2388 struct pnfs_layout_segment *
2389 pnfs_layout_process(struct nfs4_layoutget *lgp)
2391 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2392 struct nfs4_layoutget_res *res = &lgp->res;
2393 struct pnfs_layout_segment *lseg;
2394 struct inode *ino = lo->plh_inode;
2397 if (!pnfs_sanity_check_layout_range(&res->range))
2398 return ERR_PTR(-EINVAL);
2400 /* Inject layout blob into I/O device driver */
2401 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2402 if (IS_ERR_OR_NULL(lseg)) {
2404 lseg = ERR_PTR(-ENOMEM);
2406 dprintk("%s: Could not allocate layout: error %ld\n",
2407 __func__, PTR_ERR(lseg));
2411 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2413 spin_lock(&ino->i_lock);
2414 if (pnfs_layoutgets_blocked(lo)) {
2415 dprintk("%s forget reply due to state\n", __func__);
2419 if (test_bit(NFS_LAYOUT_DRAIN, &lo->plh_flags) &&
2420 !pnfs_is_first_layoutget(lo))
2423 if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2424 /* existing state ID, make sure the sequence number matches. */
2425 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2426 if (!pnfs_layout_is_valid(lo))
2427 lo->plh_barrier = 0;
2428 dprintk("%s forget reply due to sequence\n", __func__);
2431 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, false);
2432 } else if (pnfs_layout_is_valid(lo)) {
2434 * We got an entirely new state ID. Mark all segments for the
2435 * inode invalid, and retry the layoutget
2437 struct pnfs_layout_range range = {
2438 .iomode = IOMODE_ANY,
2439 .length = NFS4_MAX_UINT64,
2441 pnfs_set_plh_return_info(lo, IOMODE_ANY, 0);
2442 pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
2446 /* We have a completely new layout */
2447 pnfs_set_layout_stateid(lo, &res->stateid, lgp->cred, true);
2450 pnfs_get_lseg(lseg);
2451 pnfs_layout_insert_lseg(lo, lseg, &free_me);
2454 if (res->return_on_close)
2455 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2457 spin_unlock(&ino->i_lock);
2458 pnfs_free_lseg_list(&free_me);
2462 spin_unlock(&ino->i_lock);
2463 lseg->pls_layout = lo;
2464 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2465 pnfs_free_lseg_list(&free_me);
2466 return ERR_PTR(-EAGAIN);
2470 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2471 * @lo: pointer to layout header
2472 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2473 * @return_range: describe layout segment ranges to be returned
2474 * @seq: stateid seqid to match
2476 * This function is mainly intended for use by layoutrecall. It attempts
2477 * to free the layout segment immediately, or else to mark it for return
2478 * as soon as its reference count drops to zero.
2481 * - 0: a layoutreturn needs to be scheduled.
2482 * - EBUSY: there are layout segment that are still in use.
2483 * - ENOENT: there are no layout segments that need to be returned.
2486 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2487 struct list_head *tmp_list,
2488 const struct pnfs_layout_range *return_range,
2491 struct pnfs_layout_segment *lseg, *next;
2494 dprintk("%s:Begin lo %p\n", __func__, lo);
2496 assert_spin_locked(&lo->plh_inode->i_lock);
2498 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2499 tmp_list = &lo->plh_return_segs;
2501 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2502 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2503 dprintk("%s: marking lseg %p iomode %d "
2504 "offset %llu length %llu\n", __func__,
2505 lseg, lseg->pls_range.iomode,
2506 lseg->pls_range.offset,
2507 lseg->pls_range.length);
2508 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2509 tmp_list = &lo->plh_return_segs;
2510 if (mark_lseg_invalid(lseg, tmp_list))
2513 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2517 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2521 if (!list_empty(&lo->plh_return_segs)) {
2522 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2530 pnfs_mark_layout_for_return(struct inode *inode,
2531 const struct pnfs_layout_range *range)
2533 struct pnfs_layout_hdr *lo;
2534 bool return_now = false;
2536 spin_lock(&inode->i_lock);
2537 lo = NFS_I(inode)->layout;
2538 if (!pnfs_layout_is_valid(lo)) {
2539 spin_unlock(&inode->i_lock);
2542 pnfs_set_plh_return_info(lo, range->iomode, 0);
2544 * mark all matching lsegs so that we are sure to have no live
2545 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2548 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, range, 0) != -EBUSY) {
2549 const struct cred *cred;
2550 nfs4_stateid stateid;
2551 enum pnfs_iomode iomode;
2553 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode);
2554 spin_unlock(&inode->i_lock);
2556 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
2558 spin_unlock(&inode->i_lock);
2559 nfs_commit_inode(inode, 0);
2563 void pnfs_error_mark_layout_for_return(struct inode *inode,
2564 struct pnfs_layout_segment *lseg)
2566 struct pnfs_layout_range range = {
2567 .iomode = lseg->pls_range.iomode,
2569 .length = NFS4_MAX_UINT64,
2572 pnfs_mark_layout_for_return(inode, &range);
2574 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2577 pnfs_layout_can_be_returned(struct pnfs_layout_hdr *lo)
2579 return pnfs_layout_is_valid(lo) &&
2580 !test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) &&
2581 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
2584 static struct pnfs_layout_segment *
2585 pnfs_find_first_lseg(struct pnfs_layout_hdr *lo,
2586 const struct pnfs_layout_range *range,
2587 enum pnfs_iomode iomode)
2589 struct pnfs_layout_segment *lseg;
2591 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
2592 if (!test_bit(NFS_LSEG_VALID, &lseg->pls_flags))
2594 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2596 if (lseg->pls_range.iomode != iomode && iomode != IOMODE_ANY)
2598 if (pnfs_lseg_range_intersecting(&lseg->pls_range, range))
2604 /* Find open file states whose mode matches that of the range */
2606 pnfs_should_return_unused_layout(struct pnfs_layout_hdr *lo,
2607 const struct pnfs_layout_range *range)
2609 struct list_head *head;
2610 struct nfs_open_context *ctx;
2613 if (!pnfs_layout_can_be_returned(lo) ||
2614 !pnfs_find_first_lseg(lo, range, range->iomode))
2617 head = &NFS_I(lo->plh_inode)->open_files;
2618 list_for_each_entry_rcu(ctx, head, list) {
2620 mode |= ctx->state->state & (FMODE_READ|FMODE_WRITE);
2623 switch (range->iomode) {
2627 mode &= ~FMODE_WRITE;
2630 if (pnfs_find_first_lseg(lo, range, IOMODE_READ))
2631 mode &= ~FMODE_READ;
2636 static int pnfs_layout_return_unused_byserver(struct nfs_server *server,
2639 const struct pnfs_layout_range *range = data;
2640 const struct cred *cred;
2641 struct pnfs_layout_hdr *lo;
2642 struct inode *inode;
2643 nfs4_stateid stateid;
2644 enum pnfs_iomode iomode;
2648 list_for_each_entry_rcu(lo, &server->layouts, plh_layouts) {
2649 inode = lo->plh_inode;
2650 if (!inode || !pnfs_layout_can_be_returned(lo) ||
2651 test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2653 spin_lock(&inode->i_lock);
2654 if (!lo->plh_inode ||
2655 !pnfs_should_return_unused_layout(lo, range)) {
2656 spin_unlock(&inode->i_lock);
2659 pnfs_get_layout_hdr(lo);
2660 pnfs_set_plh_return_info(lo, range->iomode, 0);
2661 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
2663 !pnfs_prepare_layoutreturn(lo, &stateid, &cred, &iomode)) {
2664 spin_unlock(&inode->i_lock);
2666 pnfs_put_layout_hdr(lo);
2670 spin_unlock(&inode->i_lock);
2672 pnfs_send_layoutreturn(lo, &stateid, &cred, iomode, false);
2673 pnfs_put_layout_hdr(lo);
2682 pnfs_layout_return_unused_byclid(struct nfs_client *clp,
2683 enum pnfs_iomode iomode)
2685 struct pnfs_layout_range range = {
2688 .length = NFS4_MAX_UINT64,
2691 nfs_client_for_each_server(clp, pnfs_layout_return_unused_byserver,
2696 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2698 if (pgio->pg_lseg == NULL ||
2699 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2701 pnfs_put_lseg(pgio->pg_lseg);
2702 pgio->pg_lseg = NULL;
2704 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2707 * Check for any intersection between the request and the pgio->pg_lseg,
2708 * and if none, put this pgio->pg_lseg away.
2711 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2713 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2714 pnfs_put_lseg(pgio->pg_lseg);
2715 pgio->pg_lseg = NULL;
2718 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_range);
2721 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2723 u64 rd_size = req->wb_bytes;
2725 pnfs_generic_pg_check_layout(pgio);
2726 pnfs_generic_pg_check_range(pgio, req);
2727 if (pgio->pg_lseg == NULL) {
2728 if (pgio->pg_dreq == NULL)
2729 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2731 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2733 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2734 nfs_req_openctx(req),
2740 if (IS_ERR(pgio->pg_lseg)) {
2741 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2742 pgio->pg_lseg = NULL;
2746 /* If no lseg, fall back to read through mds */
2747 if (pgio->pg_lseg == NULL)
2748 nfs_pageio_reset_read_mds(pgio);
2751 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2754 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2755 struct nfs_page *req, u64 wb_size)
2757 pnfs_generic_pg_check_layout(pgio);
2758 pnfs_generic_pg_check_range(pgio, req);
2759 if (pgio->pg_lseg == NULL) {
2760 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2761 nfs_req_openctx(req),
2767 if (IS_ERR(pgio->pg_lseg)) {
2768 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2769 pgio->pg_lseg = NULL;
2773 /* If no lseg, fall back to write through mds */
2774 if (pgio->pg_lseg == NULL)
2775 nfs_pageio_reset_write_mds(pgio);
2777 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2780 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2782 if (desc->pg_lseg) {
2783 pnfs_put_lseg(desc->pg_lseg);
2784 desc->pg_lseg = NULL;
2787 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2790 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2791 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2794 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2795 struct nfs_page *prev, struct nfs_page *req)
2798 u64 seg_end, req_start, seg_left;
2800 size = nfs_generic_pg_test(pgio, prev, req);
2805 * 'size' contains the number of bytes left in the current page (up
2806 * to the original size asked for in @req->wb_bytes).
2808 * Calculate how many bytes are left in the layout segment
2809 * and if there are less bytes than 'size', return that instead.
2811 * Please also note that 'end_offset' is actually the offset of the
2812 * first byte that lies outside the pnfs_layout_range. FIXME?
2815 if (pgio->pg_lseg) {
2816 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2817 pgio->pg_lseg->pls_range.length);
2818 req_start = req_offset(req);
2820 /* start of request is past the last byte of this segment */
2821 if (req_start >= seg_end)
2824 /* adjust 'size' iff there are fewer bytes left in the
2825 * segment than what nfs_generic_pg_test returned */
2826 seg_left = seg_end - req_start;
2827 if (seg_left < size)
2828 size = (unsigned int)seg_left;
2833 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2835 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2837 struct nfs_pageio_descriptor pgio;
2839 /* Resend all requests through the MDS */
2840 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2841 hdr->completion_ops);
2842 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2843 return nfs_pageio_resend(&pgio, hdr);
2845 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2847 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2850 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2851 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2852 PNFS_LAYOUTRET_ON_ERROR) {
2853 pnfs_return_layout(hdr->inode);
2855 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2856 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2860 * Called by non rpc-based layout drivers
2862 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2864 if (likely(!hdr->pnfs_error)) {
2865 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2866 hdr->mds_offset + hdr->res.count);
2867 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2869 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2870 if (unlikely(hdr->pnfs_error))
2871 pnfs_ld_handle_write_error(hdr);
2872 hdr->mds_ops->rpc_release(hdr);
2874 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2877 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2878 struct nfs_pgio_header *hdr)
2880 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2882 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2883 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2884 nfs_pageio_reset_write_mds(desc);
2885 mirror->pg_recoalesce = 1;
2887 hdr->completion_ops->completion(hdr);
2890 static enum pnfs_try_status
2891 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2892 const struct rpc_call_ops *call_ops,
2893 struct pnfs_layout_segment *lseg,
2896 struct inode *inode = hdr->inode;
2897 enum pnfs_try_status trypnfs;
2898 struct nfs_server *nfss = NFS_SERVER(inode);
2900 hdr->mds_ops = call_ops;
2902 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2903 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2904 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2905 if (trypnfs != PNFS_NOT_ATTEMPTED)
2906 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2907 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2912 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2913 struct nfs_pgio_header *hdr, int how)
2915 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2916 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2917 enum pnfs_try_status trypnfs;
2919 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2921 case PNFS_NOT_ATTEMPTED:
2922 pnfs_write_through_mds(desc, hdr);
2923 case PNFS_ATTEMPTED:
2925 case PNFS_TRY_AGAIN:
2926 /* cleanup hdr and prepare to redo pnfs */
2927 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2928 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2929 list_splice_init(&hdr->pages, &mirror->pg_list);
2930 mirror->pg_recoalesce = 1;
2932 hdr->mds_ops->rpc_release(hdr);
2936 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2938 pnfs_put_lseg(hdr->lseg);
2939 nfs_pgio_header_free(hdr);
2943 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2945 struct nfs_pgio_header *hdr;
2948 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2950 desc->pg_error = -ENOMEM;
2951 return desc->pg_error;
2953 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2955 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2956 ret = nfs_generic_pgio(desc, hdr);
2958 pnfs_do_write(desc, hdr, desc->pg_ioflags);
2962 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2964 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2966 struct nfs_pageio_descriptor pgio;
2968 /* Resend all requests through the MDS */
2969 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2970 return nfs_pageio_resend(&pgio, hdr);
2972 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2974 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2976 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2977 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2978 PNFS_LAYOUTRET_ON_ERROR) {
2979 pnfs_return_layout(hdr->inode);
2981 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2982 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2986 * Called by non rpc-based layout drivers
2988 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2990 if (likely(!hdr->pnfs_error))
2991 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2992 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2993 if (unlikely(hdr->pnfs_error))
2994 pnfs_ld_handle_read_error(hdr);
2995 hdr->mds_ops->rpc_release(hdr);
2997 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
3000 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
3001 struct nfs_pgio_header *hdr)
3003 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3005 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3006 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
3007 nfs_pageio_reset_read_mds(desc);
3008 mirror->pg_recoalesce = 1;
3010 hdr->completion_ops->completion(hdr);
3014 * Call the appropriate parallel I/O subsystem read function.
3016 static enum pnfs_try_status
3017 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
3018 const struct rpc_call_ops *call_ops,
3019 struct pnfs_layout_segment *lseg)
3021 struct inode *inode = hdr->inode;
3022 struct nfs_server *nfss = NFS_SERVER(inode);
3023 enum pnfs_try_status trypnfs;
3025 hdr->mds_ops = call_ops;
3027 dprintk("%s: Reading ino:%lu %u@%llu\n",
3028 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
3030 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
3031 if (trypnfs != PNFS_NOT_ATTEMPTED)
3032 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
3033 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
3037 /* Resend all requests through pnfs. */
3038 void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr,
3039 unsigned int mirror_idx)
3041 struct nfs_pageio_descriptor pgio;
3043 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3044 /* Prevent deadlocks with layoutreturn! */
3045 pnfs_put_lseg(hdr->lseg);
3048 nfs_pageio_init_read(&pgio, hdr->inode, false,
3049 hdr->completion_ops);
3050 pgio.pg_mirror_idx = mirror_idx;
3051 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
3054 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
3057 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
3059 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
3060 struct pnfs_layout_segment *lseg = desc->pg_lseg;
3061 enum pnfs_try_status trypnfs;
3063 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
3065 case PNFS_NOT_ATTEMPTED:
3066 pnfs_read_through_mds(desc, hdr);
3067 case PNFS_ATTEMPTED:
3069 case PNFS_TRY_AGAIN:
3070 /* cleanup hdr and prepare to redo pnfs */
3071 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
3072 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
3073 list_splice_init(&hdr->pages, &mirror->pg_list);
3074 mirror->pg_recoalesce = 1;
3076 hdr->mds_ops->rpc_release(hdr);
3080 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
3082 pnfs_put_lseg(hdr->lseg);
3083 nfs_pgio_header_free(hdr);
3087 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
3089 struct nfs_pgio_header *hdr;
3092 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
3094 desc->pg_error = -ENOMEM;
3095 return desc->pg_error;
3097 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
3098 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
3099 ret = nfs_generic_pgio(desc, hdr);
3101 pnfs_do_read(desc, hdr);
3104 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
3106 static void pnfs_clear_layoutcommitting(struct inode *inode)
3108 unsigned long *bitlock = &NFS_I(inode)->flags;
3110 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
3111 smp_mb__after_atomic();
3112 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
3116 * There can be multiple RW segments.
3118 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
3120 struct pnfs_layout_segment *lseg;
3122 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
3123 if (lseg->pls_range.iomode == IOMODE_RW &&
3124 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
3125 list_add(&lseg->pls_lc_list, listp);
3129 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
3131 struct pnfs_layout_segment *lseg, *tmp;
3133 /* Matched by references in pnfs_set_layoutcommit */
3134 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
3135 list_del_init(&lseg->pls_lc_list);
3136 pnfs_put_lseg(lseg);
3139 pnfs_clear_layoutcommitting(inode);
3142 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
3144 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
3146 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
3149 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
3152 struct nfs_inode *nfsi = NFS_I(inode);
3153 bool mark_as_dirty = false;
3155 spin_lock(&inode->i_lock);
3156 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
3157 nfsi->layout->plh_lwb = end_pos;
3158 mark_as_dirty = true;
3159 dprintk("%s: Set layoutcommit for inode %lu ",
3160 __func__, inode->i_ino);
3161 } else if (end_pos > nfsi->layout->plh_lwb)
3162 nfsi->layout->plh_lwb = end_pos;
3163 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
3164 /* references matched in nfs4_layoutcommit_release */
3165 pnfs_get_lseg(lseg);
3167 spin_unlock(&inode->i_lock);
3168 dprintk("%s: lseg %p end_pos %llu\n",
3169 __func__, lseg, nfsi->layout->plh_lwb);
3171 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
3172 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
3174 mark_inode_dirty_sync(inode);
3176 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
3178 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
3180 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
3182 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
3183 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
3184 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
3188 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
3189 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
3190 * data to disk to allow the server to recover the data if it crashes.
3191 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
3192 * is off, and a COMMIT is sent to a data server, or
3193 * if WRITEs to a data server return NFS_DATA_SYNC.
3196 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
3198 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3199 struct nfs4_layoutcommit_data *data;
3200 struct nfs_inode *nfsi = NFS_I(inode);
3204 if (!pnfs_layoutcommit_outstanding(inode))
3207 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
3210 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
3213 status = wait_on_bit_lock_action(&nfsi->flags,
3214 NFS_INO_LAYOUTCOMMITTING,
3215 nfs_wait_bit_killable,
3222 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
3223 data = kzalloc(sizeof(*data), GFP_NOFS);
3225 goto clear_layoutcommitting;
3228 spin_lock(&inode->i_lock);
3229 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
3232 INIT_LIST_HEAD(&data->lseg_list);
3233 pnfs_list_write_lseg(inode, &data->lseg_list);
3235 end_pos = nfsi->layout->plh_lwb;
3237 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
3238 data->cred = get_cred(nfsi->layout->plh_lc_cred);
3239 spin_unlock(&inode->i_lock);
3241 data->args.inode = inode;
3242 nfs_fattr_init(&data->fattr);
3243 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
3244 data->res.fattr = &data->fattr;
3246 data->args.lastbytewritten = end_pos - 1;
3248 data->args.lastbytewritten = U64_MAX;
3249 data->res.server = NFS_SERVER(inode);
3251 if (ld->prepare_layoutcommit) {
3252 status = ld->prepare_layoutcommit(&data->args);
3254 put_cred(data->cred);
3255 spin_lock(&inode->i_lock);
3256 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
3257 if (end_pos > nfsi->layout->plh_lwb)
3258 nfsi->layout->plh_lwb = end_pos;
3264 status = nfs4_proc_layoutcommit(data, sync);
3267 mark_inode_dirty_sync(inode);
3268 dprintk("<-- %s status %d\n", __func__, status);
3271 spin_unlock(&inode->i_lock);
3273 clear_layoutcommitting:
3274 pnfs_clear_layoutcommitting(inode);
3277 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
3280 pnfs_generic_sync(struct inode *inode, bool datasync)
3282 return pnfs_layoutcommit_inode(inode, true);
3284 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3286 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3288 struct nfs4_threshold *thp;
3290 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3292 dprintk("%s mdsthreshold allocation failed\n", __func__);
3298 #if IS_ENABLED(CONFIG_NFS_V4_2)
3300 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
3302 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3303 struct nfs_server *server = NFS_SERVER(inode);
3304 struct nfs_inode *nfsi = NFS_I(inode);
3305 struct nfs42_layoutstat_data *data;
3306 struct pnfs_layout_hdr *hdr;
3309 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3312 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3315 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3318 spin_lock(&inode->i_lock);
3319 if (!NFS_I(inode)->layout) {
3320 spin_unlock(&inode->i_lock);
3321 goto out_clear_layoutstats;
3323 hdr = NFS_I(inode)->layout;
3324 pnfs_get_layout_hdr(hdr);
3325 spin_unlock(&inode->i_lock);
3327 data = kzalloc(sizeof(*data), gfp_flags);
3333 data->args.fh = NFS_FH(inode);
3334 data->args.inode = inode;
3335 status = ld->prepare_layoutstats(&data->args);
3339 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3342 dprintk("%s returns %d\n", __func__, status);
3348 pnfs_put_layout_hdr(hdr);
3349 out_clear_layoutstats:
3350 smp_mb__before_atomic();
3351 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3352 smp_mb__after_atomic();
3355 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3358 unsigned int layoutstats_timer;
3359 module_param(layoutstats_timer, uint, 0644);
3360 EXPORT_SYMBOL_GPL(layoutstats_timer);