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 (!list_empty(&lo->plh_layouts)) {
283 struct nfs_client *clp = server->nfs_client;
285 spin_lock(&clp->cl_lock);
286 list_del_init(&lo->plh_layouts);
287 spin_unlock(&clp->cl_lock);
289 put_rpccred(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))
329 pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
332 if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
334 lo->plh_return_iomode = iomode;
335 set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
337 WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
338 lo->plh_return_seq = seq;
343 pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
345 struct pnfs_layout_segment *lseg;
346 lo->plh_return_iomode = 0;
347 lo->plh_return_seq = 0;
348 clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
349 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
350 if (!test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
352 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
356 static void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
358 clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
359 clear_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags);
360 smp_mb__after_atomic();
361 wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
362 rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
366 pnfs_clear_lseg_state(struct pnfs_layout_segment *lseg,
367 struct list_head *free_me)
369 clear_bit(NFS_LSEG_ROC, &lseg->pls_flags);
370 clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
371 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags))
372 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
373 if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
374 pnfs_lseg_dec_and_remove_zero(lseg, free_me);
378 * Update the seqid of a layout stateid
380 bool nfs4_layoutreturn_refresh_stateid(nfs4_stateid *dst,
381 struct pnfs_layout_range *dst_range,
384 struct pnfs_layout_hdr *lo;
385 struct pnfs_layout_range range = {
386 .iomode = IOMODE_ANY,
388 .length = NFS4_MAX_UINT64,
394 spin_lock(&inode->i_lock);
395 lo = NFS_I(inode)->layout;
396 if (lo && nfs4_stateid_match_other(dst, &lo->plh_stateid)) {
397 err = pnfs_mark_matching_lsegs_return(lo, &head, &range, 0);
399 dst->seqid = lo->plh_stateid.seqid;
404 spin_unlock(&inode->i_lock);
405 pnfs_free_lseg_list(&head);
410 * Mark a pnfs_layout_hdr and all associated layout segments as invalid
412 * In order to continue using the pnfs_layout_hdr, a full recovery
414 * Note that caller must hold inode->i_lock.
417 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
418 struct list_head *lseg_list)
420 struct pnfs_layout_range range = {
421 .iomode = IOMODE_ANY,
423 .length = NFS4_MAX_UINT64,
425 struct pnfs_layout_segment *lseg, *next;
427 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
428 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
429 pnfs_clear_lseg_state(lseg, lseg_list);
430 pnfs_clear_layoutreturn_info(lo);
431 pnfs_free_returned_lsegs(lo, lseg_list, &range, 0);
432 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags) &&
433 !test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
434 pnfs_clear_layoutreturn_waitbit(lo);
435 return !list_empty(&lo->plh_segs);
439 pnfs_iomode_to_fail_bit(u32 iomode)
441 return iomode == IOMODE_RW ?
442 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
446 pnfs_layout_set_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
448 lo->plh_retry_timestamp = jiffies;
449 if (!test_and_set_bit(fail_bit, &lo->plh_flags))
450 refcount_inc(&lo->plh_refcount);
454 pnfs_layout_clear_fail_bit(struct pnfs_layout_hdr *lo, int fail_bit)
456 if (test_and_clear_bit(fail_bit, &lo->plh_flags))
457 refcount_dec(&lo->plh_refcount);
461 pnfs_layout_io_set_failed(struct pnfs_layout_hdr *lo, u32 iomode)
463 struct inode *inode = lo->plh_inode;
464 struct pnfs_layout_range range = {
467 .length = NFS4_MAX_UINT64,
471 spin_lock(&inode->i_lock);
472 pnfs_layout_set_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
473 pnfs_mark_matching_lsegs_invalid(lo, &head, &range, 0);
474 spin_unlock(&inode->i_lock);
475 pnfs_free_lseg_list(&head);
476 dprintk("%s Setting layout IOMODE_%s fail bit\n", __func__,
477 iomode == IOMODE_RW ? "RW" : "READ");
481 pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
483 unsigned long start, end;
484 int fail_bit = pnfs_iomode_to_fail_bit(iomode);
486 if (test_bit(fail_bit, &lo->plh_flags) == 0)
489 start = end - PNFS_LAYOUTGET_RETRY_TIMEOUT;
490 if (!time_in_range(lo->plh_retry_timestamp, start, end)) {
491 /* It is time to retry the failed layoutgets */
492 pnfs_layout_clear_fail_bit(lo, fail_bit);
499 pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
500 const struct pnfs_layout_range *range,
501 const nfs4_stateid *stateid)
503 INIT_LIST_HEAD(&lseg->pls_list);
504 INIT_LIST_HEAD(&lseg->pls_lc_list);
505 refcount_set(&lseg->pls_refcount, 1);
506 set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
507 lseg->pls_layout = lo;
508 lseg->pls_range = *range;
509 lseg->pls_seq = be32_to_cpu(stateid->seqid);
512 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
515 struct inode *inode = lseg->pls_layout->plh_inode;
516 NFS_SERVER(inode)->pnfs_curr_ld->free_lseg(lseg);
521 pnfs_layout_remove_lseg(struct pnfs_layout_hdr *lo,
522 struct pnfs_layout_segment *lseg)
524 WARN_ON(test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
525 list_del_init(&lseg->pls_list);
526 /* Matched by pnfs_get_layout_hdr in pnfs_layout_insert_lseg */
527 refcount_dec(&lo->plh_refcount);
528 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
530 if (list_empty(&lo->plh_segs) &&
531 !test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
532 !test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
533 if (atomic_read(&lo->plh_outstanding) == 0)
534 set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
535 clear_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
540 pnfs_cache_lseg_for_layoutreturn(struct pnfs_layout_hdr *lo,
541 struct pnfs_layout_segment *lseg)
543 if (test_and_clear_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) &&
544 pnfs_layout_is_valid(lo)) {
545 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
546 list_move_tail(&lseg->pls_list, &lo->plh_return_segs);
553 pnfs_put_lseg(struct pnfs_layout_segment *lseg)
555 struct pnfs_layout_hdr *lo;
561 dprintk("%s: lseg %p ref %d valid %d\n", __func__, lseg,
562 refcount_read(&lseg->pls_refcount),
563 test_bit(NFS_LSEG_VALID, &lseg->pls_flags));
565 lo = lseg->pls_layout;
566 inode = lo->plh_inode;
568 if (refcount_dec_and_lock(&lseg->pls_refcount, &inode->i_lock)) {
569 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
570 spin_unlock(&inode->i_lock);
573 pnfs_get_layout_hdr(lo);
574 pnfs_layout_remove_lseg(lo, lseg);
575 if (pnfs_cache_lseg_for_layoutreturn(lo, lseg))
577 spin_unlock(&inode->i_lock);
578 pnfs_free_lseg(lseg);
579 pnfs_put_layout_hdr(lo);
582 EXPORT_SYMBOL_GPL(pnfs_put_lseg);
585 * is l2 fully contained in l1?
587 * [----------------------------------)
592 pnfs_lseg_range_contained(const struct pnfs_layout_range *l1,
593 const struct pnfs_layout_range *l2)
595 u64 start1 = l1->offset;
596 u64 end1 = pnfs_end_offset(start1, l1->length);
597 u64 start2 = l2->offset;
598 u64 end2 = pnfs_end_offset(start2, l2->length);
600 return (start1 <= start2) && (end1 >= end2);
603 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
604 struct list_head *tmp_list)
606 if (!refcount_dec_and_test(&lseg->pls_refcount))
608 pnfs_layout_remove_lseg(lseg->pls_layout, lseg);
609 list_add(&lseg->pls_list, tmp_list);
613 /* Returns 1 if lseg is removed from list, 0 otherwise */
614 static int mark_lseg_invalid(struct pnfs_layout_segment *lseg,
615 struct list_head *tmp_list)
619 if (test_and_clear_bit(NFS_LSEG_VALID, &lseg->pls_flags)) {
620 /* Remove the reference keeping the lseg in the
621 * list. It will now be removed when all
622 * outstanding io is finished.
624 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
625 refcount_read(&lseg->pls_refcount));
626 if (pnfs_lseg_dec_and_remove_zero(lseg, tmp_list))
633 * Compare 2 layout stateid sequence ids, to see which is newer,
634 * taking into account wraparound issues.
636 static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
638 return (s32)(s1 - s2) > 0;
642 pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
643 const struct pnfs_layout_range *recall_range)
645 return (recall_range->iomode == IOMODE_ANY ||
646 lseg_range->iomode == recall_range->iomode) &&
647 pnfs_lseg_range_intersecting(lseg_range, recall_range);
651 pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
652 const struct pnfs_layout_range *recall_range,
655 if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
657 if (recall_range == NULL)
659 return pnfs_should_free_range(&lseg->pls_range, recall_range);
663 * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
664 * @lo: layout header containing the lsegs
665 * @tmp_list: list head where doomed lsegs should go
666 * @recall_range: optional recall range argument to match (may be NULL)
667 * @seq: only invalidate lsegs obtained prior to this sequence (may be 0)
669 * Walk the list of lsegs in the layout header, and tear down any that should
670 * be destroyed. If "recall_range" is specified then the segment must match
671 * that range. If "seq" is non-zero, then only match segments that were handed
672 * out at or before that sequence.
674 * Returns number of matching invalid lsegs remaining in list after scanning
675 * it and purging them.
678 pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
679 struct list_head *tmp_list,
680 const struct pnfs_layout_range *recall_range,
683 struct pnfs_layout_segment *lseg, *next;
686 dprintk("%s:Begin lo %p\n", __func__, lo);
688 if (list_empty(&lo->plh_segs))
690 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
691 if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
692 dprintk("%s: freeing lseg %p iomode %d seq %u "
693 "offset %llu length %llu\n", __func__,
694 lseg, lseg->pls_range.iomode, lseg->pls_seq,
695 lseg->pls_range.offset, lseg->pls_range.length);
696 if (!mark_lseg_invalid(lseg, tmp_list))
699 dprintk("%s:Return %i\n", __func__, remaining);
704 pnfs_free_returned_lsegs(struct pnfs_layout_hdr *lo,
705 struct list_head *free_me,
706 const struct pnfs_layout_range *range,
709 struct pnfs_layout_segment *lseg, *next;
711 list_for_each_entry_safe(lseg, next, &lo->plh_return_segs, pls_list) {
712 if (pnfs_match_lseg_recall(lseg, range, seq))
713 list_move_tail(&lseg->pls_list, free_me);
717 /* note free_me must contain lsegs from a single layout_hdr */
719 pnfs_free_lseg_list(struct list_head *free_me)
721 struct pnfs_layout_segment *lseg, *tmp;
723 if (list_empty(free_me))
726 list_for_each_entry_safe(lseg, tmp, free_me, pls_list) {
727 list_del(&lseg->pls_list);
728 pnfs_free_lseg(lseg);
732 static struct pnfs_layout_hdr *__pnfs_destroy_layout(struct nfs_inode *nfsi)
734 struct pnfs_layout_hdr *lo;
737 spin_lock(&nfsi->vfs_inode.i_lock);
740 pnfs_get_layout_hdr(lo);
741 pnfs_mark_layout_stateid_invalid(lo, &tmp_list);
742 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RO_FAILED);
743 pnfs_layout_clear_fail_bit(lo, NFS_LAYOUT_RW_FAILED);
744 spin_unlock(&nfsi->vfs_inode.i_lock);
745 pnfs_free_lseg_list(&tmp_list);
746 nfs_commit_inode(&nfsi->vfs_inode, 0);
747 pnfs_put_layout_hdr(lo);
749 spin_unlock(&nfsi->vfs_inode.i_lock);
753 void pnfs_destroy_layout(struct nfs_inode *nfsi)
755 __pnfs_destroy_layout(nfsi);
757 EXPORT_SYMBOL_GPL(pnfs_destroy_layout);
759 static bool pnfs_layout_removed(struct nfs_inode *nfsi,
760 struct pnfs_layout_hdr *lo)
764 spin_lock(&nfsi->vfs_inode.i_lock);
765 ret = nfsi->layout != lo;
766 spin_unlock(&nfsi->vfs_inode.i_lock);
770 void pnfs_destroy_layout_final(struct nfs_inode *nfsi)
772 struct pnfs_layout_hdr *lo = __pnfs_destroy_layout(nfsi);
775 wait_var_event(lo, pnfs_layout_removed(nfsi, lo));
779 pnfs_layout_add_bulk_destroy_list(struct inode *inode,
780 struct list_head *layout_list)
782 struct pnfs_layout_hdr *lo;
785 spin_lock(&inode->i_lock);
786 lo = NFS_I(inode)->layout;
787 if (lo != NULL && list_empty(&lo->plh_bulk_destroy)) {
788 pnfs_get_layout_hdr(lo);
789 list_add(&lo->plh_bulk_destroy, layout_list);
792 spin_unlock(&inode->i_lock);
796 /* Caller must hold rcu_read_lock and clp->cl_lock */
798 pnfs_layout_bulk_destroy_byserver_locked(struct nfs_client *clp,
799 struct nfs_server *server,
800 struct list_head *layout_list)
801 __must_hold(&clp->cl_lock)
804 struct pnfs_layout_hdr *lo, *next;
807 list_for_each_entry_safe(lo, next, &server->layouts, plh_layouts) {
808 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) ||
809 test_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags) ||
810 !list_empty(&lo->plh_bulk_destroy))
812 /* If the sb is being destroyed, just bail */
813 if (!nfs_sb_active(server->super))
815 inode = igrab(lo->plh_inode);
817 list_del_init(&lo->plh_layouts);
818 if (pnfs_layout_add_bulk_destroy_list(inode,
822 spin_unlock(&clp->cl_lock);
826 spin_unlock(&clp->cl_lock);
827 set_bit(NFS_LAYOUT_INODE_FREEING, &lo->plh_flags);
829 nfs_sb_deactive(server->super);
830 spin_lock(&clp->cl_lock);
838 pnfs_layout_free_bulk_destroy_list(struct list_head *layout_list,
841 struct pnfs_layout_hdr *lo;
843 LIST_HEAD(lseg_list);
846 while (!list_empty(layout_list)) {
847 lo = list_entry(layout_list->next, struct pnfs_layout_hdr,
849 dprintk("%s freeing layout for inode %lu\n", __func__,
850 lo->plh_inode->i_ino);
851 inode = lo->plh_inode;
853 pnfs_layoutcommit_inode(inode, false);
855 spin_lock(&inode->i_lock);
856 list_del_init(&lo->plh_bulk_destroy);
857 if (pnfs_mark_layout_stateid_invalid(lo, &lseg_list)) {
859 set_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
862 spin_unlock(&inode->i_lock);
863 pnfs_free_lseg_list(&lseg_list);
864 /* Free all lsegs that are attached to commit buckets */
865 nfs_commit_inode(inode, 0);
866 pnfs_put_layout_hdr(lo);
867 nfs_iput_and_deactive(inode);
873 pnfs_destroy_layouts_byfsid(struct nfs_client *clp,
874 struct nfs_fsid *fsid,
877 struct nfs_server *server;
878 LIST_HEAD(layout_list);
880 spin_lock(&clp->cl_lock);
883 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
884 if (memcmp(&server->fsid, fsid, sizeof(*fsid)) != 0)
886 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
892 spin_unlock(&clp->cl_lock);
894 if (list_empty(&layout_list))
896 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
900 pnfs_destroy_layouts_byclid(struct nfs_client *clp,
903 struct nfs_server *server;
904 LIST_HEAD(layout_list);
906 spin_lock(&clp->cl_lock);
909 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
910 if (pnfs_layout_bulk_destroy_byserver_locked(clp,
916 spin_unlock(&clp->cl_lock);
918 if (list_empty(&layout_list))
920 return pnfs_layout_free_bulk_destroy_list(&layout_list, is_recall);
924 * Called by the state manger to remove all layouts established under an
928 pnfs_destroy_all_layouts(struct nfs_client *clp)
930 nfs4_deviceid_mark_client_invalid(clp);
931 nfs4_deviceid_purge_client(clp);
933 pnfs_destroy_layouts_byclid(clp, false);
936 /* update lo->plh_stateid with new if is more recent */
938 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
941 u32 oldseq, newseq, new_barrier = 0;
943 oldseq = be32_to_cpu(lo->plh_stateid.seqid);
944 newseq = be32_to_cpu(new->seqid);
946 if (!pnfs_layout_is_valid(lo)) {
947 nfs4_stateid_copy(&lo->plh_stateid, new);
948 lo->plh_barrier = newseq;
949 pnfs_clear_layoutreturn_info(lo);
950 clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
953 if (pnfs_seqid_is_newer(newseq, oldseq)) {
954 nfs4_stateid_copy(&lo->plh_stateid, new);
956 * Because of wraparound, we want to keep the barrier
957 * "close" to the current seqids.
959 new_barrier = newseq - atomic_read(&lo->plh_outstanding);
962 new_barrier = be32_to_cpu(new->seqid);
963 else if (new_barrier == 0)
965 if (pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
966 lo->plh_barrier = new_barrier;
970 pnfs_layout_stateid_blocked(const struct pnfs_layout_hdr *lo,
971 const nfs4_stateid *stateid)
973 u32 seqid = be32_to_cpu(stateid->seqid);
975 return !pnfs_seqid_is_newer(seqid, lo->plh_barrier);
978 /* lget is set to 1 if called from inside send_layoutget call chain */
980 pnfs_layoutgets_blocked(const struct pnfs_layout_hdr *lo)
982 return lo->plh_block_lgets ||
983 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags);
986 static struct nfs_server *
987 pnfs_find_server(struct inode *inode, struct nfs_open_context *ctx)
989 struct nfs_server *server;
992 server = NFS_SERVER(inode);
994 struct dentry *parent_dir = dget_parent(ctx->dentry);
995 server = NFS_SERVER(parent_dir->d_inode);
1001 static void nfs4_free_pages(struct page **pages, size_t size)
1008 for (i = 0; i < size; i++) {
1011 __free_page(pages[i]);
1016 static struct page **nfs4_alloc_pages(size_t size, gfp_t gfp_flags)
1018 struct page **pages;
1021 pages = kcalloc(size, sizeof(struct page *), gfp_flags);
1023 dprintk("%s: can't alloc array of %zu pages\n", __func__, size);
1027 for (i = 0; i < size; i++) {
1028 pages[i] = alloc_page(gfp_flags);
1030 dprintk("%s: failed to allocate page\n", __func__);
1031 nfs4_free_pages(pages, size);
1039 static struct nfs4_layoutget *
1040 pnfs_alloc_init_layoutget_args(struct inode *ino,
1041 struct nfs_open_context *ctx,
1042 const nfs4_stateid *stateid,
1043 const struct pnfs_layout_range *range,
1046 struct nfs_server *server = pnfs_find_server(ino, ctx);
1047 size_t max_pages = max_response_pages(server);
1048 struct nfs4_layoutget *lgp;
1050 dprintk("--> %s\n", __func__);
1052 lgp = kzalloc(sizeof(*lgp), gfp_flags);
1056 lgp->args.layout.pages = nfs4_alloc_pages(max_pages, gfp_flags);
1057 if (!lgp->args.layout.pages) {
1061 lgp->args.layout.pglen = max_pages * PAGE_SIZE;
1062 lgp->res.layoutp = &lgp->args.layout;
1064 /* Don't confuse uninitialised result and success */
1065 lgp->res.status = -NFS4ERR_DELAY;
1067 lgp->args.minlength = PAGE_SIZE;
1068 if (lgp->args.minlength > range->length)
1069 lgp->args.minlength = range->length;
1071 loff_t i_size = i_size_read(ino);
1073 if (range->iomode == IOMODE_READ) {
1074 if (range->offset >= i_size)
1075 lgp->args.minlength = 0;
1076 else if (i_size - range->offset < lgp->args.minlength)
1077 lgp->args.minlength = i_size - range->offset;
1080 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
1081 pnfs_copy_range(&lgp->args.range, range);
1082 lgp->args.type = server->pnfs_curr_ld->id;
1083 lgp->args.inode = ino;
1084 lgp->args.ctx = get_nfs_open_context(ctx);
1085 nfs4_stateid_copy(&lgp->args.stateid, stateid);
1086 lgp->gfp_flags = gfp_flags;
1087 lgp->cred = get_rpccred(ctx->cred);
1091 void pnfs_layoutget_free(struct nfs4_layoutget *lgp)
1093 size_t max_pages = lgp->args.layout.pglen / PAGE_SIZE;
1095 nfs4_free_pages(lgp->args.layout.pages, max_pages);
1096 if (lgp->args.inode)
1097 pnfs_put_layout_hdr(NFS_I(lgp->args.inode)->layout);
1098 put_rpccred(lgp->cred);
1099 put_nfs_open_context(lgp->args.ctx);
1103 static void pnfs_clear_layoutcommit(struct inode *inode,
1104 struct list_head *head)
1106 struct nfs_inode *nfsi = NFS_I(inode);
1107 struct pnfs_layout_segment *lseg, *tmp;
1109 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
1111 list_for_each_entry_safe(lseg, tmp, &nfsi->layout->plh_segs, pls_list) {
1112 if (!test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
1114 pnfs_lseg_dec_and_remove_zero(lseg, head);
1118 void pnfs_layoutreturn_free_lsegs(struct pnfs_layout_hdr *lo,
1119 const nfs4_stateid *arg_stateid,
1120 const struct pnfs_layout_range *range,
1121 const nfs4_stateid *stateid)
1123 struct inode *inode = lo->plh_inode;
1126 spin_lock(&inode->i_lock);
1127 if (!pnfs_layout_is_valid(lo) || !arg_stateid ||
1128 !nfs4_stateid_match_other(&lo->plh_stateid, arg_stateid))
1131 u32 seq = be32_to_cpu(arg_stateid->seqid);
1133 pnfs_mark_matching_lsegs_invalid(lo, &freeme, range, seq);
1134 pnfs_free_returned_lsegs(lo, &freeme, range, seq);
1135 pnfs_set_layout_stateid(lo, stateid, true);
1137 pnfs_mark_layout_stateid_invalid(lo, &freeme);
1139 pnfs_clear_layoutreturn_waitbit(lo);
1140 spin_unlock(&inode->i_lock);
1141 pnfs_free_lseg_list(&freeme);
1146 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
1147 nfs4_stateid *stateid,
1148 enum pnfs_iomode *iomode)
1150 /* Serialise LAYOUTGET/LAYOUTRETURN */
1151 if (atomic_read(&lo->plh_outstanding) != 0)
1153 if (test_and_set_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags))
1155 set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags);
1156 pnfs_get_layout_hdr(lo);
1157 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
1158 if (stateid != NULL) {
1159 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1160 if (lo->plh_return_seq != 0)
1161 stateid->seqid = cpu_to_be32(lo->plh_return_seq);
1164 *iomode = lo->plh_return_iomode;
1165 pnfs_clear_layoutreturn_info(lo);
1168 if (stateid != NULL)
1169 nfs4_stateid_copy(stateid, &lo->plh_stateid);
1171 *iomode = IOMODE_ANY;
1176 pnfs_init_layoutreturn_args(struct nfs4_layoutreturn_args *args,
1177 struct pnfs_layout_hdr *lo,
1178 const nfs4_stateid *stateid,
1179 enum pnfs_iomode iomode)
1181 struct inode *inode = lo->plh_inode;
1183 args->layout_type = NFS_SERVER(inode)->pnfs_curr_ld->id;
1184 args->inode = inode;
1185 args->range.iomode = iomode;
1186 args->range.offset = 0;
1187 args->range.length = NFS4_MAX_UINT64;
1189 nfs4_stateid_copy(&args->stateid, stateid);
1193 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, const nfs4_stateid *stateid,
1194 enum pnfs_iomode iomode, bool sync)
1196 struct inode *ino = lo->plh_inode;
1197 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1198 struct nfs4_layoutreturn *lrp;
1201 lrp = kzalloc(sizeof(*lrp), GFP_NOFS);
1202 if (unlikely(lrp == NULL)) {
1204 spin_lock(&ino->i_lock);
1205 pnfs_clear_layoutreturn_waitbit(lo);
1206 spin_unlock(&ino->i_lock);
1207 pnfs_put_layout_hdr(lo);
1211 pnfs_init_layoutreturn_args(&lrp->args, lo, stateid, iomode);
1212 lrp->args.ld_private = &lrp->ld_private;
1213 lrp->clp = NFS_SERVER(ino)->nfs_client;
1214 lrp->cred = lo->plh_lc_cred;
1215 if (ld->prepare_layoutreturn)
1216 ld->prepare_layoutreturn(&lrp->args);
1218 status = nfs4_proc_layoutreturn(lrp, sync);
1220 dprintk("<-- %s status: %d\n", __func__, status);
1225 pnfs_layout_segments_returnable(struct pnfs_layout_hdr *lo,
1226 enum pnfs_iomode iomode,
1229 struct pnfs_layout_range recall_range = {
1230 .length = NFS4_MAX_UINT64,
1233 return pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
1234 &recall_range, seq) != -EBUSY;
1237 /* Return true if layoutreturn is needed */
1239 pnfs_layout_need_return(struct pnfs_layout_hdr *lo)
1241 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1243 return pnfs_layout_segments_returnable(lo, lo->plh_return_iomode,
1244 lo->plh_return_seq);
1247 static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
1249 struct inode *inode= lo->plh_inode;
1251 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1253 spin_lock(&inode->i_lock);
1254 if (pnfs_layout_need_return(lo)) {
1255 nfs4_stateid stateid;
1256 enum pnfs_iomode iomode;
1259 send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1260 spin_unlock(&inode->i_lock);
1262 /* Send an async layoutreturn so we dont deadlock */
1263 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
1266 spin_unlock(&inode->i_lock);
1270 * Initiates a LAYOUTRETURN(FILE), and removes the pnfs_layout_hdr
1271 * when the layout segment list is empty.
1273 * Note that a pnfs_layout_hdr can exist with an empty layout segment
1274 * list when LAYOUTGET has failed, or when LAYOUTGET succeeded, but the
1275 * deviceid is marked invalid.
1278 _pnfs_return_layout(struct inode *ino)
1280 struct pnfs_layout_hdr *lo = NULL;
1281 struct nfs_inode *nfsi = NFS_I(ino);
1282 struct pnfs_layout_range range = {
1283 .iomode = IOMODE_ANY,
1285 .length = NFS4_MAX_UINT64,
1287 LIST_HEAD(tmp_list);
1288 nfs4_stateid stateid;
1290 bool send, valid_layout;
1292 dprintk("NFS: %s for inode %lu\n", __func__, ino->i_ino);
1294 spin_lock(&ino->i_lock);
1297 spin_unlock(&ino->i_lock);
1298 dprintk("NFS: %s no layout to return\n", __func__);
1301 /* Reference matched in nfs4_layoutreturn_release */
1302 pnfs_get_layout_hdr(lo);
1303 /* Is there an outstanding layoutreturn ? */
1304 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1305 spin_unlock(&ino->i_lock);
1306 if (wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1307 TASK_UNINTERRUPTIBLE))
1308 goto out_put_layout_hdr;
1309 spin_lock(&ino->i_lock);
1311 valid_layout = pnfs_layout_is_valid(lo);
1312 pnfs_clear_layoutcommit(ino, &tmp_list);
1313 pnfs_mark_matching_lsegs_return(lo, &tmp_list, &range, 0);
1315 if (NFS_SERVER(ino)->pnfs_curr_ld->return_range)
1316 NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo, &range);
1318 /* Don't send a LAYOUTRETURN if list was initially empty */
1319 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) ||
1321 spin_unlock(&ino->i_lock);
1322 dprintk("NFS: %s no layout segments to return\n", __func__);
1323 goto out_put_layout_hdr;
1326 send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
1327 spin_unlock(&ino->i_lock);
1329 status = pnfs_send_layoutreturn(lo, &stateid, IOMODE_ANY, true);
1331 pnfs_free_lseg_list(&tmp_list);
1332 pnfs_put_layout_hdr(lo);
1334 dprintk("<-- %s status: %d\n", __func__, status);
1339 pnfs_commit_and_return_layout(struct inode *inode)
1341 struct pnfs_layout_hdr *lo;
1344 spin_lock(&inode->i_lock);
1345 lo = NFS_I(inode)->layout;
1347 spin_unlock(&inode->i_lock);
1350 pnfs_get_layout_hdr(lo);
1351 /* Block new layoutgets and read/write to ds */
1352 lo->plh_block_lgets++;
1353 spin_unlock(&inode->i_lock);
1354 filemap_fdatawait(inode->i_mapping);
1355 ret = pnfs_layoutcommit_inode(inode, true);
1357 ret = _pnfs_return_layout(inode);
1358 spin_lock(&inode->i_lock);
1359 lo->plh_block_lgets--;
1360 spin_unlock(&inode->i_lock);
1361 pnfs_put_layout_hdr(lo);
1365 bool pnfs_roc(struct inode *ino,
1366 struct nfs4_layoutreturn_args *args,
1367 struct nfs4_layoutreturn_res *res,
1368 const struct rpc_cred *cred)
1370 struct nfs_inode *nfsi = NFS_I(ino);
1371 struct nfs_open_context *ctx;
1372 struct nfs4_state *state;
1373 struct pnfs_layout_hdr *lo;
1374 struct pnfs_layout_segment *lseg, *next;
1375 nfs4_stateid stateid;
1376 enum pnfs_iomode iomode = 0;
1377 bool layoutreturn = false, roc = false;
1378 bool skip_read = false;
1380 if (!nfs_have_layout(ino))
1383 spin_lock(&ino->i_lock);
1385 if (!lo || !pnfs_layout_is_valid(lo) ||
1386 test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1390 pnfs_get_layout_hdr(lo);
1391 if (test_bit(NFS_LAYOUT_RETURN_LOCK, &lo->plh_flags)) {
1392 spin_unlock(&ino->i_lock);
1393 wait_on_bit(&lo->plh_flags, NFS_LAYOUT_RETURN,
1394 TASK_UNINTERRUPTIBLE);
1395 pnfs_put_layout_hdr(lo);
1399 /* no roc if we hold a delegation */
1400 if (nfs4_check_delegation(ino, FMODE_READ)) {
1401 if (nfs4_check_delegation(ino, FMODE_WRITE))
1406 list_for_each_entry(ctx, &nfsi->open_files, list) {
1410 /* Don't return layout if there is open file state */
1411 if (state->state & FMODE_WRITE)
1413 if (state->state & FMODE_READ)
1418 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list) {
1419 if (skip_read && lseg->pls_range.iomode == IOMODE_READ)
1421 /* If we are sending layoutreturn, invalidate all valid lsegs */
1422 if (!test_and_clear_bit(NFS_LSEG_ROC, &lseg->pls_flags))
1425 * Note: mark lseg for return so pnfs_layout_remove_lseg
1426 * doesn't invalidate the layout for us.
1428 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
1429 if (!mark_lseg_invalid(lseg, &lo->plh_return_segs))
1431 pnfs_set_plh_return_info(lo, lseg->pls_range.iomode, 0);
1434 if (!test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
1437 /* ROC in two conditions:
1438 * 1. there are ROC lsegs
1439 * 2. we don't send layoutreturn
1441 /* lo ref dropped in pnfs_roc_release() */
1442 layoutreturn = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
1443 /* If the creds don't match, we can't compound the layoutreturn */
1444 if (!layoutreturn || cred != lo->plh_lc_cred)
1448 pnfs_init_layoutreturn_args(args, lo, &stateid, iomode);
1449 res->lrs_present = 0;
1450 layoutreturn = false;
1453 spin_unlock(&ino->i_lock);
1454 pnfs_layoutcommit_inode(ino, true);
1456 struct pnfs_layoutdriver_type *ld = NFS_SERVER(ino)->pnfs_curr_ld;
1457 if (ld->prepare_layoutreturn)
1458 ld->prepare_layoutreturn(args);
1459 pnfs_put_layout_hdr(lo);
1463 pnfs_send_layoutreturn(lo, &stateid, iomode, true);
1464 pnfs_put_layout_hdr(lo);
1468 void pnfs_roc_release(struct nfs4_layoutreturn_args *args,
1469 struct nfs4_layoutreturn_res *res,
1472 struct pnfs_layout_hdr *lo = args->layout;
1473 struct inode *inode = args->inode;
1474 const nfs4_stateid *arg_stateid = NULL;
1475 const nfs4_stateid *res_stateid = NULL;
1476 struct nfs4_xdr_opaque_data *ld_private = args->ld_private;
1479 case -NFS4ERR_NOMATCHING_LAYOUT:
1480 spin_lock(&inode->i_lock);
1481 if (pnfs_layout_is_valid(lo) &&
1482 nfs4_stateid_match_other(&args->stateid, &lo->plh_stateid))
1483 pnfs_set_plh_return_info(lo, args->range.iomode, 0);
1484 spin_unlock(&inode->i_lock);
1487 if (res->lrs_present)
1488 res_stateid = &res->stateid;
1491 arg_stateid = &args->stateid;
1493 pnfs_layoutreturn_free_lsegs(lo, arg_stateid, &args->range,
1495 if (ld_private && ld_private->ops && ld_private->ops->free)
1496 ld_private->ops->free(ld_private);
1497 pnfs_put_layout_hdr(lo);
1498 trace_nfs4_layoutreturn_on_close(args->inode, 0);
1501 bool pnfs_wait_on_layoutreturn(struct inode *ino, struct rpc_task *task)
1503 struct nfs_inode *nfsi = NFS_I(ino);
1504 struct pnfs_layout_hdr *lo;
1507 /* we might not have grabbed lo reference. so need to check under
1509 spin_lock(&ino->i_lock);
1511 if (lo && test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1512 rpc_sleep_on(&NFS_SERVER(ino)->roc_rpcwaitq, task, NULL);
1515 spin_unlock(&ino->i_lock);
1520 * Compare two layout segments for sorting into layout cache.
1521 * We want to preferentially return RW over RO layouts, so ensure those
1525 pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
1526 const struct pnfs_layout_range *l2)
1530 /* high offset > low offset */
1531 d = l1->offset - l2->offset;
1535 /* short length > long length */
1536 d = l2->length - l1->length;
1540 /* read > read/write */
1541 return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
1545 pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
1546 const struct pnfs_layout_range *l2)
1548 return pnfs_lseg_range_cmp(l1, l2) > 0;
1552 pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
1553 struct pnfs_layout_segment *old)
1559 pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1560 struct pnfs_layout_segment *lseg,
1561 bool (*is_after)(const struct pnfs_layout_range *,
1562 const struct pnfs_layout_range *),
1563 bool (*do_merge)(struct pnfs_layout_segment *,
1564 struct pnfs_layout_segment *),
1565 struct list_head *free_me)
1567 struct pnfs_layout_segment *lp, *tmp;
1569 dprintk("%s:Begin\n", __func__);
1571 list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
1572 if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
1574 if (do_merge(lseg, lp)) {
1575 mark_lseg_invalid(lp, free_me);
1578 if (is_after(&lseg->pls_range, &lp->pls_range))
1580 list_add_tail(&lseg->pls_list, &lp->pls_list);
1581 dprintk("%s: inserted lseg %p "
1582 "iomode %d offset %llu length %llu before "
1583 "lp %p iomode %d offset %llu length %llu\n",
1584 __func__, lseg, lseg->pls_range.iomode,
1585 lseg->pls_range.offset, lseg->pls_range.length,
1586 lp, lp->pls_range.iomode, lp->pls_range.offset,
1587 lp->pls_range.length);
1590 list_add_tail(&lseg->pls_list, &lo->plh_segs);
1591 dprintk("%s: inserted lseg %p "
1592 "iomode %d offset %llu length %llu at tail\n",
1593 __func__, lseg, lseg->pls_range.iomode,
1594 lseg->pls_range.offset, lseg->pls_range.length);
1596 pnfs_get_layout_hdr(lo);
1598 dprintk("%s:Return\n", __func__);
1600 EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
1603 pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
1604 struct pnfs_layout_segment *lseg,
1605 struct list_head *free_me)
1607 struct inode *inode = lo->plh_inode;
1608 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
1610 if (ld->add_lseg != NULL)
1611 ld->add_lseg(lo, lseg, free_me);
1613 pnfs_generic_layout_insert_lseg(lo, lseg,
1614 pnfs_lseg_range_is_after,
1619 static struct pnfs_layout_hdr *
1620 alloc_init_layout_hdr(struct inode *ino,
1621 struct nfs_open_context *ctx,
1624 struct pnfs_layout_hdr *lo;
1626 lo = pnfs_alloc_layout_hdr(ino, gfp_flags);
1629 refcount_set(&lo->plh_refcount, 1);
1630 INIT_LIST_HEAD(&lo->plh_layouts);
1631 INIT_LIST_HEAD(&lo->plh_segs);
1632 INIT_LIST_HEAD(&lo->plh_return_segs);
1633 INIT_LIST_HEAD(&lo->plh_bulk_destroy);
1634 lo->plh_inode = ino;
1635 lo->plh_lc_cred = get_rpccred(ctx->cred);
1636 lo->plh_flags |= 1 << NFS_LAYOUT_INVALID_STID;
1640 static struct pnfs_layout_hdr *
1641 pnfs_find_alloc_layout(struct inode *ino,
1642 struct nfs_open_context *ctx,
1644 __releases(&ino->i_lock)
1645 __acquires(&ino->i_lock)
1647 struct nfs_inode *nfsi = NFS_I(ino);
1648 struct pnfs_layout_hdr *new = NULL;
1650 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
1652 if (nfsi->layout != NULL)
1654 spin_unlock(&ino->i_lock);
1655 new = alloc_init_layout_hdr(ino, ctx, gfp_flags);
1656 spin_lock(&ino->i_lock);
1658 if (likely(nfsi->layout == NULL)) { /* Won the race? */
1661 } else if (new != NULL)
1662 pnfs_free_layout_hdr(new);
1664 pnfs_get_layout_hdr(nfsi->layout);
1665 return nfsi->layout;
1669 * iomode matching rules:
1670 * iomode lseg strict match
1672 * ----- ----- ------ -----
1677 * READ READ N/A true
1678 * READ RW true false
1679 * READ RW false true
1682 pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range,
1683 const struct pnfs_layout_range *range,
1686 struct pnfs_layout_range range1;
1688 if ((range->iomode == IOMODE_RW &&
1689 ls_range->iomode != IOMODE_RW) ||
1690 (range->iomode != ls_range->iomode &&
1692 !pnfs_lseg_range_intersecting(ls_range, range))
1695 /* range1 covers only the first byte in the range */
1698 return pnfs_lseg_range_contained(ls_range, &range1);
1702 * lookup range in layout
1704 static struct pnfs_layout_segment *
1705 pnfs_find_lseg(struct pnfs_layout_hdr *lo,
1706 struct pnfs_layout_range *range,
1709 struct pnfs_layout_segment *lseg, *ret = NULL;
1711 dprintk("%s:Begin\n", __func__);
1713 list_for_each_entry(lseg, &lo->plh_segs, pls_list) {
1714 if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) &&
1715 pnfs_lseg_range_match(&lseg->pls_range, range,
1717 ret = pnfs_get_lseg(lseg);
1722 dprintk("%s:Return lseg %p ref %d\n",
1723 __func__, ret, ret ? refcount_read(&ret->pls_refcount) : 0);
1728 * Use mdsthreshold hints set at each OPEN to determine if I/O should go
1729 * to the MDS or over pNFS
1731 * The nfs_inode read_io and write_io fields are cumulative counters reset
1732 * when there are no layout segments. Note that in pnfs_update_layout iomode
1733 * is set to IOMODE_READ for a READ request, and set to IOMODE_RW for a
1736 * A return of true means use MDS I/O.
1739 * If a file's size is smaller than the file size threshold, data accesses
1740 * SHOULD be sent to the metadata server. If an I/O request has a length that
1741 * is below the I/O size threshold, the I/O SHOULD be sent to the metadata
1742 * server. If both file size and I/O size are provided, the client SHOULD
1743 * reach or exceed both thresholds before sending its read or write
1744 * requests to the data server.
1746 static bool pnfs_within_mdsthreshold(struct nfs_open_context *ctx,
1747 struct inode *ino, int iomode)
1749 struct nfs4_threshold *t = ctx->mdsthreshold;
1750 struct nfs_inode *nfsi = NFS_I(ino);
1751 loff_t fsize = i_size_read(ino);
1752 bool size = false, size_set = false, io = false, io_set = false, ret = false;
1757 dprintk("%s bm=0x%x rd_sz=%llu wr_sz=%llu rd_io=%llu wr_io=%llu\n",
1758 __func__, t->bm, t->rd_sz, t->wr_sz, t->rd_io_sz, t->wr_io_sz);
1762 if (t->bm & THRESHOLD_RD) {
1763 dprintk("%s fsize %llu\n", __func__, fsize);
1765 if (fsize < t->rd_sz)
1768 if (t->bm & THRESHOLD_RD_IO) {
1769 dprintk("%s nfsi->read_io %llu\n", __func__,
1772 if (nfsi->read_io < t->rd_io_sz)
1777 if (t->bm & THRESHOLD_WR) {
1778 dprintk("%s fsize %llu\n", __func__, fsize);
1780 if (fsize < t->wr_sz)
1783 if (t->bm & THRESHOLD_WR_IO) {
1784 dprintk("%s nfsi->write_io %llu\n", __func__,
1787 if (nfsi->write_io < t->wr_io_sz)
1792 if (size_set && io_set) {
1795 } else if (size || io)
1798 dprintk("<-- %s size %d io %d ret %d\n", __func__, size, io, ret);
1802 static int pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
1805 * send layoutcommit as it can hold up layoutreturn due to lseg
1808 pnfs_layoutcommit_inode(lo->plh_inode, false);
1809 return wait_on_bit_action(&lo->plh_flags, NFS_LAYOUT_RETURN,
1810 nfs_wait_bit_killable,
1814 static void nfs_layoutget_begin(struct pnfs_layout_hdr *lo)
1816 atomic_inc(&lo->plh_outstanding);
1819 static void nfs_layoutget_end(struct pnfs_layout_hdr *lo)
1821 if (atomic_dec_and_test(&lo->plh_outstanding))
1822 wake_up_var(&lo->plh_outstanding);
1825 static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
1827 unsigned long *bitlock = &lo->plh_flags;
1829 clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
1830 smp_mb__after_atomic();
1831 wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
1834 static void _add_to_server_list(struct pnfs_layout_hdr *lo,
1835 struct nfs_server *server)
1837 if (list_empty(&lo->plh_layouts)) {
1838 struct nfs_client *clp = server->nfs_client;
1840 /* The lo must be on the clp list if there is any
1841 * chance of a CB_LAYOUTRECALL(FILE) coming in.
1843 spin_lock(&clp->cl_lock);
1844 if (list_empty(&lo->plh_layouts))
1845 list_add_tail(&lo->plh_layouts, &server->layouts);
1846 spin_unlock(&clp->cl_lock);
1851 * Layout segment is retreived from the server if not cached.
1852 * The appropriate layout segment is referenced and returned to the caller.
1854 struct pnfs_layout_segment *
1855 pnfs_update_layout(struct inode *ino,
1856 struct nfs_open_context *ctx,
1859 enum pnfs_iomode iomode,
1863 struct pnfs_layout_range arg = {
1869 struct nfs_server *server = NFS_SERVER(ino);
1870 struct nfs_client *clp = server->nfs_client;
1871 struct pnfs_layout_hdr *lo = NULL;
1872 struct pnfs_layout_segment *lseg = NULL;
1873 struct nfs4_layoutget *lgp;
1874 nfs4_stateid stateid;
1876 unsigned long giveup = jiffies + (clp->cl_lease_time << 1);
1879 if (!pnfs_enabled_sb(NFS_SERVER(ino))) {
1880 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1881 PNFS_UPDATE_LAYOUT_NO_PNFS);
1885 if (pnfs_within_mdsthreshold(ctx, ino, iomode)) {
1886 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1887 PNFS_UPDATE_LAYOUT_MDSTHRESH);
1892 lseg = ERR_PTR(nfs4_client_recover_expired_lease(clp));
1896 spin_lock(&ino->i_lock);
1897 lo = pnfs_find_alloc_layout(ino, ctx, gfp_flags);
1899 spin_unlock(&ino->i_lock);
1900 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1901 PNFS_UPDATE_LAYOUT_NOMEM);
1905 /* Do we even need to bother with this? */
1906 if (test_bit(NFS_LAYOUT_BULK_RECALL, &lo->plh_flags)) {
1907 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1908 PNFS_UPDATE_LAYOUT_BULK_RECALL);
1909 dprintk("%s matches recall, use MDS\n", __func__);
1913 /* if LAYOUTGET already failed once we don't try again */
1914 if (pnfs_layout_io_test_failed(lo, iomode)) {
1915 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1916 PNFS_UPDATE_LAYOUT_IO_TEST_FAIL);
1921 * If the layout segment list is empty, but there are outstanding
1922 * layoutget calls, then they might be subject to a layoutrecall.
1924 if (list_empty(&lo->plh_segs) &&
1925 atomic_read(&lo->plh_outstanding) != 0) {
1926 spin_unlock(&ino->i_lock);
1927 lseg = ERR_PTR(wait_var_event_killable(&lo->plh_outstanding,
1928 !atomic_read(&lo->plh_outstanding)));
1930 goto out_put_layout_hdr;
1931 pnfs_put_layout_hdr(lo);
1935 lseg = pnfs_find_lseg(lo, &arg, strict_iomode);
1937 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1938 PNFS_UPDATE_LAYOUT_FOUND_CACHED);
1942 if (!nfs4_valid_open_stateid(ctx->state)) {
1943 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
1944 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1949 * Choose a stateid for the LAYOUTGET. If we don't have a layout
1950 * stateid, or it has been invalidated, then we must use the open
1953 if (test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags)) {
1956 * The first layoutget for the file. Need to serialize per
1957 * RFC 5661 Errata 3208.
1959 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET,
1961 spin_unlock(&ino->i_lock);
1962 lseg = ERR_PTR(wait_on_bit(&lo->plh_flags,
1963 NFS_LAYOUT_FIRST_LAYOUTGET,
1966 goto out_put_layout_hdr;
1967 pnfs_put_layout_hdr(lo);
1968 dprintk("%s retrying\n", __func__);
1973 if (nfs4_select_rw_stateid(ctx->state,
1974 iomode == IOMODE_RW ? FMODE_WRITE : FMODE_READ,
1975 NULL, &stateid, NULL) != 0) {
1976 trace_pnfs_update_layout(ino, pos, count,
1978 PNFS_UPDATE_LAYOUT_INVALID_OPEN);
1982 nfs4_stateid_copy(&stateid, &lo->plh_stateid);
1986 * Because we free lsegs before sending LAYOUTRETURN, we need to wait
1987 * for LAYOUTRETURN even if first is true.
1989 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags)) {
1990 spin_unlock(&ino->i_lock);
1991 dprintk("%s wait for layoutreturn\n", __func__);
1992 lseg = ERR_PTR(pnfs_prepare_to_retry_layoutget(lo));
1993 if (!IS_ERR(lseg)) {
1995 pnfs_clear_first_layoutget(lo);
1996 pnfs_put_layout_hdr(lo);
1997 dprintk("%s retrying\n", __func__);
1998 trace_pnfs_update_layout(ino, pos, count, iomode, lo,
1999 lseg, PNFS_UPDATE_LAYOUT_RETRY);
2002 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2003 PNFS_UPDATE_LAYOUT_RETURN);
2004 goto out_put_layout_hdr;
2007 if (pnfs_layoutgets_blocked(lo)) {
2008 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2009 PNFS_UPDATE_LAYOUT_BLOCKED);
2012 nfs_layoutget_begin(lo);
2013 spin_unlock(&ino->i_lock);
2015 _add_to_server_list(lo, server);
2017 pg_offset = arg.offset & ~PAGE_MASK;
2019 arg.offset -= pg_offset;
2020 arg.length += pg_offset;
2022 if (arg.length != NFS4_MAX_UINT64)
2023 arg.length = PAGE_ALIGN(arg.length);
2025 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, &stateid, &arg, gfp_flags);
2027 trace_pnfs_update_layout(ino, pos, count, iomode, lo, NULL,
2028 PNFS_UPDATE_LAYOUT_NOMEM);
2029 nfs_layoutget_end(lo);
2030 goto out_put_layout_hdr;
2033 lseg = nfs4_proc_layoutget(lgp, &timeout);
2034 trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg,
2035 PNFS_UPDATE_LAYOUT_SEND_LAYOUTGET);
2036 nfs_layoutget_end(lo);
2038 switch(PTR_ERR(lseg)) {
2040 if (time_after(jiffies, giveup))
2043 case -ERECALLCONFLICT:
2047 if (!nfs_error_is_fatal(PTR_ERR(lseg))) {
2048 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2051 goto out_put_layout_hdr;
2055 pnfs_clear_first_layoutget(lo);
2056 trace_pnfs_update_layout(ino, pos, count,
2057 iomode, lo, lseg, PNFS_UPDATE_LAYOUT_RETRY);
2058 pnfs_put_layout_hdr(lo);
2062 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2067 pnfs_clear_first_layoutget(lo);
2068 pnfs_put_layout_hdr(lo);
2070 dprintk("%s: inode %s/%llu pNFS layout segment %s for "
2071 "(%s, offset: %llu, length: %llu)\n",
2072 __func__, ino->i_sb->s_id,
2073 (unsigned long long)NFS_FILEID(ino),
2074 IS_ERR_OR_NULL(lseg) ? "not found" : "found",
2075 iomode==IOMODE_RW ? "read/write" : "read-only",
2076 (unsigned long long)pos,
2077 (unsigned long long)count);
2080 spin_unlock(&ino->i_lock);
2081 goto out_put_layout_hdr;
2083 EXPORT_SYMBOL_GPL(pnfs_update_layout);
2086 pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
2088 switch (range->iomode) {
2095 if (range->offset == NFS4_MAX_UINT64)
2097 if (range->length == 0)
2099 if (range->length != NFS4_MAX_UINT64 &&
2100 range->length > NFS4_MAX_UINT64 - range->offset)
2105 static struct pnfs_layout_hdr *
2106 _pnfs_grab_empty_layout(struct inode *ino, struct nfs_open_context *ctx)
2108 struct pnfs_layout_hdr *lo;
2110 spin_lock(&ino->i_lock);
2111 lo = pnfs_find_alloc_layout(ino, ctx, GFP_KERNEL);
2114 if (!test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags))
2116 if (test_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
2118 if (pnfs_layoutgets_blocked(lo))
2120 if (test_and_set_bit(NFS_LAYOUT_FIRST_LAYOUTGET, &lo->plh_flags))
2122 nfs_layoutget_begin(lo);
2123 spin_unlock(&ino->i_lock);
2124 _add_to_server_list(lo, NFS_SERVER(ino));
2128 spin_unlock(&ino->i_lock);
2129 pnfs_put_layout_hdr(lo);
2133 extern const nfs4_stateid current_stateid;
2135 static void _lgopen_prepare_attached(struct nfs4_opendata *data,
2136 struct nfs_open_context *ctx)
2138 struct inode *ino = data->dentry->d_inode;
2139 struct pnfs_layout_range rng = {
2140 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2141 IOMODE_RW: IOMODE_READ,
2143 .length = NFS4_MAX_UINT64,
2145 struct nfs4_layoutget *lgp;
2146 struct pnfs_layout_hdr *lo;
2148 /* Heuristic: don't send layoutget if we have cached data */
2149 if (rng.iomode == IOMODE_READ &&
2150 (i_size_read(ino) == 0 || ino->i_mapping->nrpages != 0))
2153 lo = _pnfs_grab_empty_layout(ino, ctx);
2156 lgp = pnfs_alloc_init_layoutget_args(ino, ctx, ¤t_stateid,
2159 pnfs_clear_first_layoutget(lo);
2160 nfs_layoutget_end(lo);
2161 pnfs_put_layout_hdr(lo);
2165 data->o_arg.lg_args = &lgp->args;
2166 data->o_res.lg_res = &lgp->res;
2169 static void _lgopen_prepare_floating(struct nfs4_opendata *data,
2170 struct nfs_open_context *ctx)
2172 struct pnfs_layout_range rng = {
2173 .iomode = (data->o_arg.fmode & FMODE_WRITE) ?
2174 IOMODE_RW: IOMODE_READ,
2176 .length = NFS4_MAX_UINT64,
2178 struct nfs4_layoutget *lgp;
2180 lgp = pnfs_alloc_init_layoutget_args(NULL, ctx, ¤t_stateid,
2185 data->o_arg.lg_args = &lgp->args;
2186 data->o_res.lg_res = &lgp->res;
2189 void pnfs_lgopen_prepare(struct nfs4_opendata *data,
2190 struct nfs_open_context *ctx)
2192 struct nfs_server *server = NFS_SERVER(data->dir->d_inode);
2194 if (!(pnfs_enabled_sb(server) &&
2195 server->pnfs_curr_ld->flags & PNFS_LAYOUTGET_ON_OPEN))
2197 /* Could check on max_ops, but currently hardcoded high enough */
2198 if (!nfs_server_capable(data->dir->d_inode, NFS_CAP_LGOPEN))
2201 _lgopen_prepare_attached(data, ctx);
2203 _lgopen_prepare_floating(data, ctx);
2206 void pnfs_parse_lgopen(struct inode *ino, struct nfs4_layoutget *lgp,
2207 struct nfs_open_context *ctx)
2209 struct pnfs_layout_hdr *lo;
2210 struct pnfs_layout_segment *lseg;
2211 struct nfs_server *srv = NFS_SERVER(ino);
2216 dprintk("%s: entered with status %i\n", __func__, lgp->res.status);
2217 if (lgp->res.status) {
2218 switch (lgp->res.status) {
2222 * Halt lgopen attempts if the server doesn't recognise
2223 * the "current stateid" value, the layout type, or the
2224 * layoutget operation as being valid.
2225 * Also if it complains about too many ops in the compound
2226 * or of the request/reply being too big.
2228 case -NFS4ERR_BAD_STATEID:
2229 case -NFS4ERR_NOTSUPP:
2230 case -NFS4ERR_REP_TOO_BIG:
2231 case -NFS4ERR_REP_TOO_BIG_TO_CACHE:
2232 case -NFS4ERR_REQ_TOO_BIG:
2233 case -NFS4ERR_TOO_MANY_OPS:
2234 case -NFS4ERR_UNKNOWN_LAYOUTTYPE:
2235 srv->caps &= ~NFS_CAP_LGOPEN;
2239 if (!lgp->args.inode) {
2240 lo = _pnfs_grab_empty_layout(ino, ctx);
2243 lgp->args.inode = ino;
2245 lo = NFS_I(lgp->args.inode)->layout;
2247 lseg = pnfs_layout_process(lgp);
2248 if (!IS_ERR(lseg)) {
2249 iomode = lgp->args.range.iomode;
2250 pnfs_layout_clear_fail_bit(lo, pnfs_iomode_to_fail_bit(iomode));
2251 pnfs_put_lseg(lseg);
2255 void nfs4_lgopen_release(struct nfs4_layoutget *lgp)
2258 struct inode *inode = lgp->args.inode;
2260 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2261 pnfs_clear_first_layoutget(lo);
2262 nfs_layoutget_end(lo);
2264 pnfs_layoutget_free(lgp);
2268 struct pnfs_layout_segment *
2269 pnfs_layout_process(struct nfs4_layoutget *lgp)
2271 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
2272 struct nfs4_layoutget_res *res = &lgp->res;
2273 struct pnfs_layout_segment *lseg;
2274 struct inode *ino = lo->plh_inode;
2277 if (!pnfs_sanity_check_layout_range(&res->range))
2278 return ERR_PTR(-EINVAL);
2280 /* Inject layout blob into I/O device driver */
2281 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
2282 if (IS_ERR_OR_NULL(lseg)) {
2284 lseg = ERR_PTR(-ENOMEM);
2286 dprintk("%s: Could not allocate layout: error %ld\n",
2287 __func__, PTR_ERR(lseg));
2291 pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
2293 spin_lock(&ino->i_lock);
2294 if (pnfs_layoutgets_blocked(lo)) {
2295 dprintk("%s forget reply due to state\n", __func__);
2299 if (!pnfs_layout_is_valid(lo)) {
2300 /* We have a completely new layout */
2301 pnfs_set_layout_stateid(lo, &res->stateid, true);
2302 } else if (nfs4_stateid_match_other(&lo->plh_stateid, &res->stateid)) {
2303 /* existing state ID, make sure the sequence number matches. */
2304 if (pnfs_layout_stateid_blocked(lo, &res->stateid)) {
2305 dprintk("%s forget reply due to sequence\n", __func__);
2308 pnfs_set_layout_stateid(lo, &res->stateid, false);
2311 * We got an entirely new state ID. Mark all segments for the
2312 * inode invalid, and retry the layoutget
2314 struct pnfs_layout_range range = {
2315 .iomode = IOMODE_ANY,
2316 .length = NFS4_MAX_UINT64,
2318 pnfs_set_plh_return_info(lo, IOMODE_ANY, 0);
2319 pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs,
2324 pnfs_get_lseg(lseg);
2325 pnfs_layout_insert_lseg(lo, lseg, &free_me);
2328 if (res->return_on_close)
2329 set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
2331 spin_unlock(&ino->i_lock);
2332 pnfs_free_lseg_list(&free_me);
2336 spin_unlock(&ino->i_lock);
2337 lseg->pls_layout = lo;
2338 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
2339 pnfs_free_lseg_list(&free_me);
2340 return ERR_PTR(-EAGAIN);
2344 * pnfs_mark_matching_lsegs_return - Free or return matching layout segments
2345 * @lo: pointer to layout header
2346 * @tmp_list: list header to be used with pnfs_free_lseg_list()
2347 * @return_range: describe layout segment ranges to be returned
2348 * @seq: stateid seqid to match
2350 * This function is mainly intended for use by layoutrecall. It attempts
2351 * to free the layout segment immediately, or else to mark it for return
2352 * as soon as its reference count drops to zero.
2355 * - 0: a layoutreturn needs to be scheduled.
2356 * - EBUSY: there are layout segment that are still in use.
2357 * - ENOENT: there are no layout segments that need to be returned.
2360 pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
2361 struct list_head *tmp_list,
2362 const struct pnfs_layout_range *return_range,
2365 struct pnfs_layout_segment *lseg, *next;
2368 dprintk("%s:Begin lo %p\n", __func__, lo);
2370 assert_spin_locked(&lo->plh_inode->i_lock);
2372 if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
2373 tmp_list = &lo->plh_return_segs;
2375 list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
2376 if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
2377 dprintk("%s: marking lseg %p iomode %d "
2378 "offset %llu length %llu\n", __func__,
2379 lseg, lseg->pls_range.iomode,
2380 lseg->pls_range.offset,
2381 lseg->pls_range.length);
2382 if (test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))
2383 tmp_list = &lo->plh_return_segs;
2384 if (mark_lseg_invalid(lseg, tmp_list))
2387 set_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags);
2391 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2395 if (!list_empty(&lo->plh_return_segs)) {
2396 pnfs_set_plh_return_info(lo, return_range->iomode, seq);
2403 void pnfs_error_mark_layout_for_return(struct inode *inode,
2404 struct pnfs_layout_segment *lseg)
2406 struct pnfs_layout_hdr *lo = NFS_I(inode)->layout;
2407 struct pnfs_layout_range range = {
2408 .iomode = lseg->pls_range.iomode,
2410 .length = NFS4_MAX_UINT64,
2412 bool return_now = false;
2414 spin_lock(&inode->i_lock);
2415 if (!pnfs_layout_is_valid(lo)) {
2416 spin_unlock(&inode->i_lock);
2419 pnfs_set_plh_return_info(lo, range.iomode, 0);
2421 * mark all matching lsegs so that we are sure to have no live
2422 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
2425 if (pnfs_mark_matching_lsegs_return(lo, &lo->plh_return_segs, &range, 0) != -EBUSY) {
2426 nfs4_stateid stateid;
2427 enum pnfs_iomode iomode;
2429 return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
2430 spin_unlock(&inode->i_lock);
2432 pnfs_send_layoutreturn(lo, &stateid, iomode, false);
2434 spin_unlock(&inode->i_lock);
2435 nfs_commit_inode(inode, 0);
2438 EXPORT_SYMBOL_GPL(pnfs_error_mark_layout_for_return);
2441 pnfs_generic_pg_check_layout(struct nfs_pageio_descriptor *pgio)
2443 if (pgio->pg_lseg == NULL ||
2444 test_bit(NFS_LSEG_VALID, &pgio->pg_lseg->pls_flags))
2446 pnfs_put_lseg(pgio->pg_lseg);
2447 pgio->pg_lseg = NULL;
2449 EXPORT_SYMBOL_GPL(pnfs_generic_pg_check_layout);
2452 * Check for any intersection between the request and the pgio->pg_lseg,
2453 * and if none, put this pgio->pg_lseg away.
2456 pnfs_generic_pg_check_range(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2458 if (pgio->pg_lseg && !pnfs_lseg_request_intersecting(pgio->pg_lseg, req)) {
2459 pnfs_put_lseg(pgio->pg_lseg);
2460 pgio->pg_lseg = NULL;
2465 pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
2467 u64 rd_size = req->wb_bytes;
2469 pnfs_generic_pg_check_layout(pgio);
2470 pnfs_generic_pg_check_range(pgio, req);
2471 if (pgio->pg_lseg == NULL) {
2472 if (pgio->pg_dreq == NULL)
2473 rd_size = i_size_read(pgio->pg_inode) - req_offset(req);
2475 rd_size = nfs_dreq_bytes_left(pgio->pg_dreq);
2477 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2484 if (IS_ERR(pgio->pg_lseg)) {
2485 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2486 pgio->pg_lseg = NULL;
2490 /* If no lseg, fall back to read through mds */
2491 if (pgio->pg_lseg == NULL)
2492 nfs_pageio_reset_read_mds(pgio);
2495 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_read);
2498 pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio,
2499 struct nfs_page *req, u64 wb_size)
2501 pnfs_generic_pg_check_layout(pgio);
2502 pnfs_generic_pg_check_range(pgio, req);
2503 if (pgio->pg_lseg == NULL) {
2504 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
2511 if (IS_ERR(pgio->pg_lseg)) {
2512 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
2513 pgio->pg_lseg = NULL;
2517 /* If no lseg, fall back to write through mds */
2518 if (pgio->pg_lseg == NULL)
2519 nfs_pageio_reset_write_mds(pgio);
2521 EXPORT_SYMBOL_GPL(pnfs_generic_pg_init_write);
2524 pnfs_generic_pg_cleanup(struct nfs_pageio_descriptor *desc)
2526 if (desc->pg_lseg) {
2527 pnfs_put_lseg(desc->pg_lseg);
2528 desc->pg_lseg = NULL;
2531 EXPORT_SYMBOL_GPL(pnfs_generic_pg_cleanup);
2534 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number
2535 * of bytes (maximum @req->wb_bytes) that can be coalesced.
2538 pnfs_generic_pg_test(struct nfs_pageio_descriptor *pgio,
2539 struct nfs_page *prev, struct nfs_page *req)
2542 u64 seg_end, req_start, seg_left;
2544 size = nfs_generic_pg_test(pgio, prev, req);
2549 * 'size' contains the number of bytes left in the current page (up
2550 * to the original size asked for in @req->wb_bytes).
2552 * Calculate how many bytes are left in the layout segment
2553 * and if there are less bytes than 'size', return that instead.
2555 * Please also note that 'end_offset' is actually the offset of the
2556 * first byte that lies outside the pnfs_layout_range. FIXME?
2559 if (pgio->pg_lseg) {
2560 seg_end = pnfs_end_offset(pgio->pg_lseg->pls_range.offset,
2561 pgio->pg_lseg->pls_range.length);
2562 req_start = req_offset(req);
2564 /* start of request is past the last byte of this segment */
2565 if (req_start >= seg_end)
2568 /* adjust 'size' iff there are fewer bytes left in the
2569 * segment than what nfs_generic_pg_test returned */
2570 seg_left = seg_end - req_start;
2571 if (seg_left < size)
2572 size = (unsigned int)seg_left;
2577 EXPORT_SYMBOL_GPL(pnfs_generic_pg_test);
2579 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *hdr)
2581 struct nfs_pageio_descriptor pgio;
2583 /* Resend all requests through the MDS */
2584 nfs_pageio_init_write(&pgio, hdr->inode, FLUSH_STABLE, true,
2585 hdr->completion_ops);
2586 set_bit(NFS_CONTEXT_RESEND_WRITES, &hdr->args.context->flags);
2587 return nfs_pageio_resend(&pgio, hdr);
2589 EXPORT_SYMBOL_GPL(pnfs_write_done_resend_to_mds);
2591 static void pnfs_ld_handle_write_error(struct nfs_pgio_header *hdr)
2594 dprintk("pnfs write error = %d\n", hdr->pnfs_error);
2595 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2596 PNFS_LAYOUTRET_ON_ERROR) {
2597 pnfs_return_layout(hdr->inode);
2599 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2600 hdr->task.tk_status = pnfs_write_done_resend_to_mds(hdr);
2604 * Called by non rpc-based layout drivers
2606 void pnfs_ld_write_done(struct nfs_pgio_header *hdr)
2608 if (likely(!hdr->pnfs_error)) {
2609 pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
2610 hdr->mds_offset + hdr->res.count);
2611 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2613 trace_nfs4_pnfs_write(hdr, hdr->pnfs_error);
2614 if (unlikely(hdr->pnfs_error))
2615 pnfs_ld_handle_write_error(hdr);
2616 hdr->mds_ops->rpc_release(hdr);
2618 EXPORT_SYMBOL_GPL(pnfs_ld_write_done);
2621 pnfs_write_through_mds(struct nfs_pageio_descriptor *desc,
2622 struct nfs_pgio_header *hdr)
2624 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2626 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2627 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2628 nfs_pageio_reset_write_mds(desc);
2629 mirror->pg_recoalesce = 1;
2631 hdr->completion_ops->completion(hdr);
2634 static enum pnfs_try_status
2635 pnfs_try_to_write_data(struct nfs_pgio_header *hdr,
2636 const struct rpc_call_ops *call_ops,
2637 struct pnfs_layout_segment *lseg,
2640 struct inode *inode = hdr->inode;
2641 enum pnfs_try_status trypnfs;
2642 struct nfs_server *nfss = NFS_SERVER(inode);
2644 hdr->mds_ops = call_ops;
2646 dprintk("%s: Writing ino:%lu %u@%llu (how %d)\n", __func__,
2647 inode->i_ino, hdr->args.count, hdr->args.offset, how);
2648 trypnfs = nfss->pnfs_curr_ld->write_pagelist(hdr, how);
2649 if (trypnfs != PNFS_NOT_ATTEMPTED)
2650 nfs_inc_stats(inode, NFSIOS_PNFS_WRITE);
2651 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2656 pnfs_do_write(struct nfs_pageio_descriptor *desc,
2657 struct nfs_pgio_header *hdr, int how)
2659 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2660 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2661 enum pnfs_try_status trypnfs;
2663 trypnfs = pnfs_try_to_write_data(hdr, call_ops, lseg, how);
2665 case PNFS_NOT_ATTEMPTED:
2666 pnfs_write_through_mds(desc, hdr);
2667 case PNFS_ATTEMPTED:
2669 case PNFS_TRY_AGAIN:
2670 /* cleanup hdr and prepare to redo pnfs */
2671 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2672 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2673 list_splice_init(&hdr->pages, &mirror->pg_list);
2674 mirror->pg_recoalesce = 1;
2676 hdr->mds_ops->rpc_release(hdr);
2680 static void pnfs_writehdr_free(struct nfs_pgio_header *hdr)
2682 pnfs_put_lseg(hdr->lseg);
2683 nfs_pgio_header_free(hdr);
2687 pnfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
2689 struct nfs_pgio_header *hdr;
2692 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2694 desc->pg_error = -ENOMEM;
2695 return desc->pg_error;
2697 nfs_pgheader_init(desc, hdr, pnfs_writehdr_free);
2699 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2700 ret = nfs_generic_pgio(desc, hdr);
2702 pnfs_do_write(desc, hdr, desc->pg_ioflags);
2706 EXPORT_SYMBOL_GPL(pnfs_generic_pg_writepages);
2708 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *hdr)
2710 struct nfs_pageio_descriptor pgio;
2712 /* Resend all requests through the MDS */
2713 nfs_pageio_init_read(&pgio, hdr->inode, true, hdr->completion_ops);
2714 return nfs_pageio_resend(&pgio, hdr);
2716 EXPORT_SYMBOL_GPL(pnfs_read_done_resend_to_mds);
2718 static void pnfs_ld_handle_read_error(struct nfs_pgio_header *hdr)
2720 dprintk("pnfs read error = %d\n", hdr->pnfs_error);
2721 if (NFS_SERVER(hdr->inode)->pnfs_curr_ld->flags &
2722 PNFS_LAYOUTRET_ON_ERROR) {
2723 pnfs_return_layout(hdr->inode);
2725 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags))
2726 hdr->task.tk_status = pnfs_read_done_resend_to_mds(hdr);
2730 * Called by non rpc-based layout drivers
2732 void pnfs_ld_read_done(struct nfs_pgio_header *hdr)
2734 if (likely(!hdr->pnfs_error))
2735 hdr->mds_ops->rpc_call_done(&hdr->task, hdr);
2736 trace_nfs4_pnfs_read(hdr, hdr->pnfs_error);
2737 if (unlikely(hdr->pnfs_error))
2738 pnfs_ld_handle_read_error(hdr);
2739 hdr->mds_ops->rpc_release(hdr);
2741 EXPORT_SYMBOL_GPL(pnfs_ld_read_done);
2744 pnfs_read_through_mds(struct nfs_pageio_descriptor *desc,
2745 struct nfs_pgio_header *hdr)
2747 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2749 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2750 list_splice_tail_init(&hdr->pages, &mirror->pg_list);
2751 nfs_pageio_reset_read_mds(desc);
2752 mirror->pg_recoalesce = 1;
2754 hdr->completion_ops->completion(hdr);
2758 * Call the appropriate parallel I/O subsystem read function.
2760 static enum pnfs_try_status
2761 pnfs_try_to_read_data(struct nfs_pgio_header *hdr,
2762 const struct rpc_call_ops *call_ops,
2763 struct pnfs_layout_segment *lseg)
2765 struct inode *inode = hdr->inode;
2766 struct nfs_server *nfss = NFS_SERVER(inode);
2767 enum pnfs_try_status trypnfs;
2769 hdr->mds_ops = call_ops;
2771 dprintk("%s: Reading ino:%lu %u@%llu\n",
2772 __func__, inode->i_ino, hdr->args.count, hdr->args.offset);
2774 trypnfs = nfss->pnfs_curr_ld->read_pagelist(hdr);
2775 if (trypnfs != PNFS_NOT_ATTEMPTED)
2776 nfs_inc_stats(inode, NFSIOS_PNFS_READ);
2777 dprintk("%s End (trypnfs:%d)\n", __func__, trypnfs);
2781 /* Resend all requests through pnfs. */
2782 void pnfs_read_resend_pnfs(struct nfs_pgio_header *hdr)
2784 struct nfs_pageio_descriptor pgio;
2786 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2787 /* Prevent deadlocks with layoutreturn! */
2788 pnfs_put_lseg(hdr->lseg);
2791 nfs_pageio_init_read(&pgio, hdr->inode, false,
2792 hdr->completion_ops);
2793 hdr->task.tk_status = nfs_pageio_resend(&pgio, hdr);
2796 EXPORT_SYMBOL_GPL(pnfs_read_resend_pnfs);
2799 pnfs_do_read(struct nfs_pageio_descriptor *desc, struct nfs_pgio_header *hdr)
2801 const struct rpc_call_ops *call_ops = desc->pg_rpc_callops;
2802 struct pnfs_layout_segment *lseg = desc->pg_lseg;
2803 enum pnfs_try_status trypnfs;
2805 trypnfs = pnfs_try_to_read_data(hdr, call_ops, lseg);
2807 case PNFS_NOT_ATTEMPTED:
2808 pnfs_read_through_mds(desc, hdr);
2809 case PNFS_ATTEMPTED:
2811 case PNFS_TRY_AGAIN:
2812 /* cleanup hdr and prepare to redo pnfs */
2813 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
2814 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
2815 list_splice_init(&hdr->pages, &mirror->pg_list);
2816 mirror->pg_recoalesce = 1;
2818 hdr->mds_ops->rpc_release(hdr);
2822 static void pnfs_readhdr_free(struct nfs_pgio_header *hdr)
2824 pnfs_put_lseg(hdr->lseg);
2825 nfs_pgio_header_free(hdr);
2829 pnfs_generic_pg_readpages(struct nfs_pageio_descriptor *desc)
2831 struct nfs_pgio_header *hdr;
2834 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
2836 desc->pg_error = -ENOMEM;
2837 return desc->pg_error;
2839 nfs_pgheader_init(desc, hdr, pnfs_readhdr_free);
2840 hdr->lseg = pnfs_get_lseg(desc->pg_lseg);
2841 ret = nfs_generic_pgio(desc, hdr);
2843 pnfs_do_read(desc, hdr);
2846 EXPORT_SYMBOL_GPL(pnfs_generic_pg_readpages);
2848 static void pnfs_clear_layoutcommitting(struct inode *inode)
2850 unsigned long *bitlock = &NFS_I(inode)->flags;
2852 clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
2853 smp_mb__after_atomic();
2854 wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
2858 * There can be multiple RW segments.
2860 static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp)
2862 struct pnfs_layout_segment *lseg;
2864 list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) {
2865 if (lseg->pls_range.iomode == IOMODE_RW &&
2866 test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags))
2867 list_add(&lseg->pls_lc_list, listp);
2871 static void pnfs_list_write_lseg_done(struct inode *inode, struct list_head *listp)
2873 struct pnfs_layout_segment *lseg, *tmp;
2875 /* Matched by references in pnfs_set_layoutcommit */
2876 list_for_each_entry_safe(lseg, tmp, listp, pls_lc_list) {
2877 list_del_init(&lseg->pls_lc_list);
2878 pnfs_put_lseg(lseg);
2881 pnfs_clear_layoutcommitting(inode);
2884 void pnfs_set_lo_fail(struct pnfs_layout_segment *lseg)
2886 pnfs_layout_io_set_failed(lseg->pls_layout, lseg->pls_range.iomode);
2888 EXPORT_SYMBOL_GPL(pnfs_set_lo_fail);
2891 pnfs_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg,
2894 struct nfs_inode *nfsi = NFS_I(inode);
2895 bool mark_as_dirty = false;
2897 spin_lock(&inode->i_lock);
2898 if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
2899 nfsi->layout->plh_lwb = end_pos;
2900 mark_as_dirty = true;
2901 dprintk("%s: Set layoutcommit for inode %lu ",
2902 __func__, inode->i_ino);
2903 } else if (end_pos > nfsi->layout->plh_lwb)
2904 nfsi->layout->plh_lwb = end_pos;
2905 if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) {
2906 /* references matched in nfs4_layoutcommit_release */
2907 pnfs_get_lseg(lseg);
2909 spin_unlock(&inode->i_lock);
2910 dprintk("%s: lseg %p end_pos %llu\n",
2911 __func__, lseg, nfsi->layout->plh_lwb);
2913 /* if pnfs_layoutcommit_inode() runs between inode locks, the next one
2914 * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */
2916 mark_inode_dirty_sync(inode);
2918 EXPORT_SYMBOL_GPL(pnfs_set_layoutcommit);
2920 void pnfs_cleanup_layoutcommit(struct nfs4_layoutcommit_data *data)
2922 struct nfs_server *nfss = NFS_SERVER(data->args.inode);
2924 if (nfss->pnfs_curr_ld->cleanup_layoutcommit)
2925 nfss->pnfs_curr_ld->cleanup_layoutcommit(data);
2926 pnfs_list_write_lseg_done(data->args.inode, &data->lseg_list);
2930 * For the LAYOUT4_NFSV4_1_FILES layout type, NFS_DATA_SYNC WRITEs and
2931 * NFS_UNSTABLE WRITEs with a COMMIT to data servers must store enough
2932 * data to disk to allow the server to recover the data if it crashes.
2933 * LAYOUTCOMMIT is only needed when the NFL4_UFLG_COMMIT_THRU_MDS flag
2934 * is off, and a COMMIT is sent to a data server, or
2935 * if WRITEs to a data server return NFS_DATA_SYNC.
2938 pnfs_layoutcommit_inode(struct inode *inode, bool sync)
2940 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
2941 struct nfs4_layoutcommit_data *data;
2942 struct nfs_inode *nfsi = NFS_I(inode);
2946 if (!pnfs_layoutcommit_outstanding(inode))
2949 dprintk("--> %s inode %lu\n", __func__, inode->i_ino);
2952 if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
2955 status = wait_on_bit_lock_action(&nfsi->flags,
2956 NFS_INO_LAYOUTCOMMITTING,
2957 nfs_wait_bit_killable,
2964 /* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
2965 data = kzalloc(sizeof(*data), GFP_NOFS);
2967 goto clear_layoutcommitting;
2970 spin_lock(&inode->i_lock);
2971 if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
2974 INIT_LIST_HEAD(&data->lseg_list);
2975 pnfs_list_write_lseg(inode, &data->lseg_list);
2977 end_pos = nfsi->layout->plh_lwb;
2979 nfs4_stateid_copy(&data->args.stateid, &nfsi->layout->plh_stateid);
2980 spin_unlock(&inode->i_lock);
2982 data->args.inode = inode;
2983 data->cred = get_rpccred(nfsi->layout->plh_lc_cred);
2984 nfs_fattr_init(&data->fattr);
2985 data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask;
2986 data->res.fattr = &data->fattr;
2988 data->args.lastbytewritten = end_pos - 1;
2990 data->args.lastbytewritten = U64_MAX;
2991 data->res.server = NFS_SERVER(inode);
2993 if (ld->prepare_layoutcommit) {
2994 status = ld->prepare_layoutcommit(&data->args);
2996 put_rpccred(data->cred);
2997 spin_lock(&inode->i_lock);
2998 set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags);
2999 if (end_pos > nfsi->layout->plh_lwb)
3000 nfsi->layout->plh_lwb = end_pos;
3006 status = nfs4_proc_layoutcommit(data, sync);
3009 mark_inode_dirty_sync(inode);
3010 dprintk("<-- %s status %d\n", __func__, status);
3013 spin_unlock(&inode->i_lock);
3015 clear_layoutcommitting:
3016 pnfs_clear_layoutcommitting(inode);
3019 EXPORT_SYMBOL_GPL(pnfs_layoutcommit_inode);
3022 pnfs_generic_sync(struct inode *inode, bool datasync)
3024 return pnfs_layoutcommit_inode(inode, true);
3026 EXPORT_SYMBOL_GPL(pnfs_generic_sync);
3028 struct nfs4_threshold *pnfs_mdsthreshold_alloc(void)
3030 struct nfs4_threshold *thp;
3032 thp = kzalloc(sizeof(*thp), GFP_NOFS);
3034 dprintk("%s mdsthreshold allocation failed\n", __func__);
3040 #if IS_ENABLED(CONFIG_NFS_V4_2)
3042 pnfs_report_layoutstat(struct inode *inode, gfp_t gfp_flags)
3044 struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
3045 struct nfs_server *server = NFS_SERVER(inode);
3046 struct nfs_inode *nfsi = NFS_I(inode);
3047 struct nfs42_layoutstat_data *data;
3048 struct pnfs_layout_hdr *hdr;
3051 if (!pnfs_enabled_sb(server) || !ld->prepare_layoutstats)
3054 if (!nfs_server_capable(inode, NFS_CAP_LAYOUTSTATS))
3057 if (test_and_set_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags))
3060 spin_lock(&inode->i_lock);
3061 if (!NFS_I(inode)->layout) {
3062 spin_unlock(&inode->i_lock);
3063 goto out_clear_layoutstats;
3065 hdr = NFS_I(inode)->layout;
3066 pnfs_get_layout_hdr(hdr);
3067 spin_unlock(&inode->i_lock);
3069 data = kzalloc(sizeof(*data), gfp_flags);
3075 data->args.fh = NFS_FH(inode);
3076 data->args.inode = inode;
3077 status = ld->prepare_layoutstats(&data->args);
3081 status = nfs42_proc_layoutstats_generic(NFS_SERVER(inode), data);
3084 dprintk("%s returns %d\n", __func__, status);
3090 pnfs_put_layout_hdr(hdr);
3091 out_clear_layoutstats:
3092 smp_mb__before_atomic();
3093 clear_bit(NFS_INO_LAYOUTSTATS, &nfsi->flags);
3094 smp_mb__after_atomic();
3097 EXPORT_SYMBOL_GPL(pnfs_report_layoutstat);
3100 unsigned int layoutstats_timer;
3101 module_param(layoutstats_timer, uint, 0644);
3102 EXPORT_SYMBOL_GPL(layoutstats_timer);