1 // SPDX-License-Identifier: GPL-2.0-only
3 * Module for pnfs flexfile layout driver.
5 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
7 * Tao Peng <bergwolf@primarydata.com>
10 #include <linux/nfs_fs.h>
11 #include <linux/nfs_page.h>
12 #include <linux/module.h>
13 #include <linux/sched/mm.h>
15 #include <linux/sunrpc/metrics.h>
17 #include "flexfilelayout.h"
18 #include "../nfs4session.h"
19 #include "../nfs4idmap.h"
20 #include "../internal.h"
21 #include "../delegation.h"
22 #include "../nfs4trace.h"
23 #include "../iostat.h"
27 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
29 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
30 #define FF_LAYOUTRETURN_MAXERR 20
32 static unsigned short io_maxretrans;
34 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
35 struct nfs_pgio_header *hdr);
36 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
37 struct nfs42_layoutstat_devinfo *devinfo,
39 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
40 const struct nfs42_layoutstat_devinfo *devinfo,
41 struct nfs4_ff_layout_mirror *mirror);
43 static struct pnfs_layout_hdr *
44 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
46 struct nfs4_flexfile_layout *ffl;
48 ffl = kzalloc(sizeof(*ffl), gfp_flags);
50 INIT_LIST_HEAD(&ffl->error_list);
51 INIT_LIST_HEAD(&ffl->mirrors);
52 ffl->last_report_time = ktime_get();
53 return &ffl->generic_hdr;
59 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
61 struct nfs4_ff_layout_ds_err *err, *n;
63 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
68 kfree(FF_LAYOUT_FROM_HDR(lo));
71 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
75 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
76 if (unlikely(p == NULL))
78 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
79 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
80 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
81 p[0], p[1], p[2], p[3]);
85 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
89 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
92 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
93 nfs4_print_deviceid(devid);
97 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
101 p = xdr_inline_decode(xdr, 4);
104 fh->size = be32_to_cpup(p++);
105 if (fh->size > sizeof(struct nfs_fh)) {
106 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
111 p = xdr_inline_decode(xdr, fh->size);
114 memcpy(&fh->data, p, fh->size);
115 dprintk("%s: fh len %d\n", __func__, fh->size);
121 * Currently only stringified uids and gids are accepted.
122 * I.e., kerberos is not supported to the DSes, so no pricipals.
124 * That means that one common function will suffice, but when
125 * principals are added, this should be split to accomodate
126 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
129 decode_name(struct xdr_stream *xdr, u32 *id)
134 /* opaque_length(4)*/
135 p = xdr_inline_decode(xdr, 4);
138 len = be32_to_cpup(p++);
142 dprintk("%s: len %u\n", __func__, len);
145 p = xdr_inline_decode(xdr, len);
149 if (!nfs_map_string_to_numeric((char *)p, len, id))
155 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
156 const struct nfs4_ff_layout_mirror *m2)
160 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
162 for (i = 0; i < m1->fh_versions_cnt; i++) {
163 bool found_fh = false;
164 for (j = 0; j < m2->fh_versions_cnt; j++) {
165 if (nfs_compare_fh(&m1->fh_versions[i],
166 &m2->fh_versions[j]) == 0) {
177 static struct nfs4_ff_layout_mirror *
178 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
179 struct nfs4_ff_layout_mirror *mirror)
181 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
182 struct nfs4_ff_layout_mirror *pos;
183 struct inode *inode = lo->plh_inode;
185 spin_lock(&inode->i_lock);
186 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
187 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
189 if (!ff_mirror_match_fh(mirror, pos))
191 if (refcount_inc_not_zero(&pos->ref)) {
192 spin_unlock(&inode->i_lock);
196 list_add(&mirror->mirrors, &ff_layout->mirrors);
198 spin_unlock(&inode->i_lock);
203 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
206 if (mirror->layout == NULL)
208 inode = mirror->layout->plh_inode;
209 spin_lock(&inode->i_lock);
210 list_del(&mirror->mirrors);
211 spin_unlock(&inode->i_lock);
212 mirror->layout = NULL;
215 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
217 struct nfs4_ff_layout_mirror *mirror;
219 mirror = kzalloc(sizeof(*mirror), gfp_flags);
220 if (mirror != NULL) {
221 spin_lock_init(&mirror->lock);
222 refcount_set(&mirror->ref, 1);
223 INIT_LIST_HEAD(&mirror->mirrors);
228 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
230 const struct cred *cred;
232 ff_layout_remove_mirror(mirror);
233 kfree(mirror->fh_versions);
234 cred = rcu_access_pointer(mirror->ro_cred);
236 cred = rcu_access_pointer(mirror->rw_cred);
238 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
242 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
244 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
245 ff_layout_free_mirror(mirror);
248 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
252 if (fls->mirror_array) {
253 for (i = 0; i < fls->mirror_array_cnt; i++) {
254 /* normally mirror_ds is freed in
255 * .free_deviceid_node but we still do it here
256 * for .alloc_lseg error path */
257 ff_layout_put_mirror(fls->mirror_array[i]);
259 kfree(fls->mirror_array);
260 fls->mirror_array = NULL;
264 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
268 dprintk("--> %s\n", __func__);
270 /* FIXME: remove this check when layout segment support is added */
271 if (lgr->range.offset != 0 ||
272 lgr->range.length != NFS4_MAX_UINT64) {
273 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
278 dprintk("--> %s returns %d\n", __func__, ret);
282 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
285 ff_layout_free_mirror_array(fls);
291 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
292 const struct pnfs_layout_range *l2)
296 if (l1->iomode != l2->iomode)
297 return l1->iomode != IOMODE_READ;
298 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
299 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
300 if (end1 < l2->offset)
302 if (end2 < l1->offset)
304 return l2->offset <= l1->offset;
308 ff_lseg_merge(struct pnfs_layout_segment *new,
309 struct pnfs_layout_segment *old)
311 u64 new_end, old_end;
313 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
315 if (new->pls_range.iomode != old->pls_range.iomode)
317 old_end = pnfs_calc_offset_end(old->pls_range.offset,
318 old->pls_range.length);
319 if (old_end < new->pls_range.offset)
321 new_end = pnfs_calc_offset_end(new->pls_range.offset,
322 new->pls_range.length);
323 if (new_end < old->pls_range.offset)
326 /* Mergeable: copy info from 'old' to 'new' */
327 if (new_end < old_end)
329 if (new->pls_range.offset < old->pls_range.offset)
330 new->pls_range.offset = old->pls_range.offset;
331 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
333 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
334 set_bit(NFS_LSEG_ROC, &new->pls_flags);
339 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
340 struct pnfs_layout_segment *lseg,
341 struct list_head *free_me)
343 pnfs_generic_layout_insert_lseg(lo, lseg,
344 ff_lseg_range_is_after,
349 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
353 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
354 for (j = i + 1; j < fls->mirror_array_cnt; j++)
355 if (fls->mirror_array[i]->efficiency <
356 fls->mirror_array[j]->efficiency)
357 swap(fls->mirror_array[i],
358 fls->mirror_array[j]);
362 static struct pnfs_layout_segment *
363 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
364 struct nfs4_layoutget_res *lgr,
367 struct pnfs_layout_segment *ret;
368 struct nfs4_ff_layout_segment *fls = NULL;
369 struct xdr_stream stream;
371 struct page *scratch;
373 u32 mirror_array_cnt;
377 dprintk("--> %s\n", __func__);
378 scratch = alloc_page(gfp_flags);
380 return ERR_PTR(-ENOMEM);
382 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
384 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
386 /* stripe unit and mirror_array_cnt */
388 p = xdr_inline_decode(&stream, 8 + 4);
392 p = xdr_decode_hyper(p, &stripe_unit);
393 mirror_array_cnt = be32_to_cpup(p++);
394 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
395 stripe_unit, mirror_array_cnt);
397 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
398 mirror_array_cnt == 0)
402 fls = kzalloc(sizeof(*fls), gfp_flags);
406 fls->mirror_array_cnt = mirror_array_cnt;
407 fls->stripe_unit = stripe_unit;
408 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
409 sizeof(fls->mirror_array[0]), gfp_flags);
410 if (fls->mirror_array == NULL)
413 for (i = 0; i < fls->mirror_array_cnt; i++) {
414 struct nfs4_ff_layout_mirror *mirror;
416 const struct cred __rcu *cred;
419 u32 ds_count, fh_count, id;
423 p = xdr_inline_decode(&stream, 4);
426 ds_count = be32_to_cpup(p);
428 /* FIXME: allow for striping? */
432 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
433 if (fls->mirror_array[i] == NULL) {
438 fls->mirror_array[i]->ds_count = ds_count;
441 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
447 p = xdr_inline_decode(&stream, 4);
450 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
453 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
459 p = xdr_inline_decode(&stream, 4);
462 fh_count = be32_to_cpup(p);
464 fls->mirror_array[i]->fh_versions =
465 kcalloc(fh_count, sizeof(struct nfs_fh),
467 if (fls->mirror_array[i]->fh_versions == NULL) {
472 for (j = 0; j < fh_count; j++) {
473 rc = decode_nfs_fh(&stream,
474 &fls->mirror_array[i]->fh_versions[j]);
479 fls->mirror_array[i]->fh_versions_cnt = fh_count;
482 rc = decode_name(&stream, &id);
486 uid = make_kuid(&init_user_ns, id);
489 rc = decode_name(&stream, &id);
493 gid = make_kgid(&init_user_ns, id);
495 if (gfp_flags & __GFP_FS)
496 kcred = prepare_kernel_cred(NULL);
498 unsigned int nofs_flags = memalloc_nofs_save();
499 kcred = prepare_kernel_cred(NULL);
500 memalloc_nofs_restore(nofs_flags);
507 cred = RCU_INITIALIZER(kcred);
509 if (lgr->range.iomode == IOMODE_READ)
510 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
512 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
514 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
515 if (mirror != fls->mirror_array[i]) {
516 /* swap cred ptrs so free_mirror will clean up old */
517 if (lgr->range.iomode == IOMODE_READ) {
518 cred = xchg(&mirror->ro_cred, cred);
519 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
521 cred = xchg(&mirror->rw_cred, cred);
522 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
524 ff_layout_free_mirror(fls->mirror_array[i]);
525 fls->mirror_array[i] = mirror;
528 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
529 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
530 from_kuid(&init_user_ns, uid),
531 from_kgid(&init_user_ns, gid));
534 p = xdr_inline_decode(&stream, 4);
536 goto out_sort_mirrors;
537 fls->flags = be32_to_cpup(p);
539 p = xdr_inline_decode(&stream, 4);
541 goto out_sort_mirrors;
542 for (i=0; i < fls->mirror_array_cnt; i++)
543 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
546 ff_layout_sort_mirrors(fls);
547 rc = ff_layout_check_layout(lgr);
550 ret = &fls->generic_hdr;
551 dprintk("<-- %s (success)\n", __func__);
553 __free_page(scratch);
556 _ff_layout_free_lseg(fls);
558 dprintk("<-- %s (%d)\n", __func__, rc);
562 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
564 struct pnfs_layout_segment *lseg;
566 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
567 if (lseg->pls_range.iomode == IOMODE_RW)
574 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
576 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
578 dprintk("--> %s\n", __func__);
580 if (lseg->pls_range.iomode == IOMODE_RW) {
581 struct nfs4_flexfile_layout *ffl;
584 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
585 inode = ffl->generic_hdr.plh_inode;
586 spin_lock(&inode->i_lock);
587 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
588 ffl->commit_info.nbuckets = 0;
589 kfree(ffl->commit_info.buckets);
590 ffl->commit_info.buckets = NULL;
592 spin_unlock(&inode->i_lock);
594 _ff_layout_free_lseg(fls);
597 /* Return 1 until we have multiple lsegs support */
599 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
605 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
607 /* first IO request? */
608 if (atomic_inc_return(&timer->n_ops) == 1) {
609 timer->start_time = now;
614 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
618 if (atomic_dec_return(&timer->n_ops) < 0)
621 start = timer->start_time;
622 timer->start_time = now;
623 return ktime_sub(now, start);
627 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
628 struct nfs4_ff_layoutstat *layoutstat,
631 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
632 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
634 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
635 if (!mirror->start_time)
636 mirror->start_time = now;
637 if (mirror->report_interval != 0)
638 report_interval = (s64)mirror->report_interval * 1000LL;
639 else if (layoutstats_timer != 0)
640 report_interval = (s64)layoutstats_timer * 1000LL;
641 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
643 ffl->last_report_time = now;
651 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
654 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
656 iostat->ops_requested++;
657 iostat->bytes_requested += requested;
661 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
664 ktime_t time_completed,
665 ktime_t time_started)
667 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
668 ktime_t completion_time = ktime_sub(time_completed, time_started);
671 iostat->ops_completed++;
672 iostat->bytes_completed += completed;
673 iostat->bytes_not_delivered += requested - completed;
675 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
676 iostat->total_busy_time =
677 ktime_add(iostat->total_busy_time, timer);
678 iostat->aggregate_completion_time =
679 ktime_add(iostat->aggregate_completion_time,
684 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
685 struct nfs4_ff_layout_mirror *mirror,
686 __u64 requested, ktime_t now)
690 spin_lock(&mirror->lock);
691 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
692 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
693 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
694 spin_unlock(&mirror->lock);
697 pnfs_report_layoutstat(inode, GFP_KERNEL);
701 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
702 struct nfs4_ff_layout_mirror *mirror,
706 spin_lock(&mirror->lock);
707 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
708 requested, completed,
709 ktime_get(), task->tk_start);
710 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
711 spin_unlock(&mirror->lock);
715 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
716 struct nfs4_ff_layout_mirror *mirror,
717 __u64 requested, ktime_t now)
721 spin_lock(&mirror->lock);
722 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
723 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
724 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
725 spin_unlock(&mirror->lock);
728 pnfs_report_layoutstat(inode, GFP_NOIO);
732 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
733 struct nfs4_ff_layout_mirror *mirror,
736 enum nfs3_stable_how committed)
738 if (committed == NFS_UNSTABLE)
739 requested = completed = 0;
741 spin_lock(&mirror->lock);
742 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
743 requested, completed, ktime_get(), task->tk_start);
744 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
745 spin_unlock(&mirror->lock);
749 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
750 struct nfs_commit_info *cinfo,
753 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
754 struct pnfs_commit_bucket *buckets;
757 if (cinfo->ds->nbuckets != 0) {
758 /* This assumes there is only one RW lseg per file.
759 * To support multiple lseg per file, we need to
760 * change struct pnfs_commit_bucket to allow dynamic
761 * increasing nbuckets.
766 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
768 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
775 spin_lock(&cinfo->inode->i_lock);
776 if (cinfo->ds->nbuckets != 0)
779 cinfo->ds->buckets = buckets;
780 cinfo->ds->nbuckets = size;
781 for (i = 0; i < size; i++) {
782 INIT_LIST_HEAD(&buckets[i].written);
783 INIT_LIST_HEAD(&buckets[i].committing);
784 /* mark direct verifier as unset */
785 buckets[i].direct_verf.committed =
786 NFS_INVALID_STABLE_HOW;
789 spin_unlock(&cinfo->inode->i_lock);
795 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
797 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
800 nfs4_mark_deviceid_unavailable(devid);
804 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
806 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
809 nfs4_mark_deviceid_available(devid);
812 static struct nfs4_pnfs_ds *
813 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
814 int start_idx, int *best_idx,
817 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
818 struct nfs4_ff_layout_mirror *mirror;
819 struct nfs4_pnfs_ds *ds;
820 bool fail_return = false;
823 /* mirrors are initially sorted by efficiency */
824 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
825 if (idx+1 == fls->mirror_array_cnt)
826 fail_return = !check_device;
828 mirror = FF_LAYOUT_COMP(lseg, idx);
829 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
834 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
844 static struct nfs4_pnfs_ds *
845 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
846 int start_idx, int *best_idx)
848 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
851 static struct nfs4_pnfs_ds *
852 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
853 int start_idx, int *best_idx)
855 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
858 static struct nfs4_pnfs_ds *
859 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
860 int start_idx, int *best_idx)
862 struct nfs4_pnfs_ds *ds;
864 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
867 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
871 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
872 struct nfs_page *req,
875 pnfs_put_lseg(pgio->pg_lseg);
876 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
877 nfs_req_openctx(req),
883 if (IS_ERR(pgio->pg_lseg)) {
884 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
885 pgio->pg_lseg = NULL;
890 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
891 struct nfs_page *req)
893 struct nfs_pgio_mirror *pgm;
894 struct nfs4_ff_layout_mirror *mirror;
895 struct nfs4_pnfs_ds *ds;
899 pnfs_generic_pg_check_layout(pgio);
900 /* Use full layout for now */
901 if (!pgio->pg_lseg) {
902 ff_layout_pg_get_read(pgio, req, false);
906 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
907 ff_layout_pg_get_read(pgio, req, true);
912 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
914 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
916 pnfs_put_lseg(pgio->pg_lseg);
917 pgio->pg_lseg = NULL;
918 /* Sleep for 1 second before retrying */
923 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
925 pgio->pg_mirror_idx = ds_idx;
927 /* read always uses only one mirror - idx 0 for pgio layer */
928 pgm = &pgio->pg_mirrors[0];
929 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
931 pgio->pg_maxretrans = io_maxretrans;
934 if (pgio->pg_error < 0)
937 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
938 0, NFS4_MAX_UINT64, IOMODE_READ,
939 NFS_I(pgio->pg_inode)->layout,
941 pnfs_put_lseg(pgio->pg_lseg);
942 pgio->pg_lseg = NULL;
943 nfs_pageio_reset_read_mds(pgio);
947 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
948 struct nfs_page *req)
950 struct nfs4_ff_layout_mirror *mirror;
951 struct nfs_pgio_mirror *pgm;
952 struct nfs_commit_info cinfo;
953 struct nfs4_pnfs_ds *ds;
958 pnfs_generic_pg_check_layout(pgio);
959 if (!pgio->pg_lseg) {
960 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
961 nfs_req_openctx(req),
967 if (IS_ERR(pgio->pg_lseg)) {
968 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
969 pgio->pg_lseg = NULL;
973 /* If no lseg, fall back to write through mds */
974 if (pgio->pg_lseg == NULL)
977 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
978 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
982 /* Use a direct mapping of ds_idx to pgio mirror_idx */
983 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
984 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
987 for (i = 0; i < pgio->pg_mirror_count; i++) {
988 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
989 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
991 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
993 pnfs_put_lseg(pgio->pg_lseg);
994 pgio->pg_lseg = NULL;
995 /* Sleep for 1 second before retrying */
999 pgm = &pgio->pg_mirrors[i];
1000 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
1003 pgio->pg_maxretrans = io_maxretrans;
1007 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
1008 0, NFS4_MAX_UINT64, IOMODE_RW,
1009 NFS_I(pgio->pg_inode)->layout,
1011 pnfs_put_lseg(pgio->pg_lseg);
1012 pgio->pg_lseg = NULL;
1013 nfs_pageio_reset_write_mds(pgio);
1017 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1018 struct nfs_page *req)
1020 if (!pgio->pg_lseg) {
1021 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1022 nfs_req_openctx(req),
1028 if (IS_ERR(pgio->pg_lseg)) {
1029 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1030 pgio->pg_lseg = NULL;
1035 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1037 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
1038 0, NFS4_MAX_UINT64, IOMODE_RW,
1039 NFS_I(pgio->pg_inode)->layout,
1041 /* no lseg means that pnfs is not in use, so no mirroring here */
1042 nfs_pageio_reset_write_mds(pgio);
1047 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1048 .pg_init = ff_layout_pg_init_read,
1049 .pg_test = pnfs_generic_pg_test,
1050 .pg_doio = pnfs_generic_pg_readpages,
1051 .pg_cleanup = pnfs_generic_pg_cleanup,
1054 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1055 .pg_init = ff_layout_pg_init_write,
1056 .pg_test = pnfs_generic_pg_test,
1057 .pg_doio = pnfs_generic_pg_writepages,
1058 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1059 .pg_cleanup = pnfs_generic_pg_cleanup,
1062 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1064 struct rpc_task *task = &hdr->task;
1066 pnfs_layoutcommit_inode(hdr->inode, false);
1069 dprintk("%s Reset task %5u for i/o through pNFS "
1070 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1072 hdr->inode->i_sb->s_id,
1073 (unsigned long long)NFS_FILEID(hdr->inode),
1075 (unsigned long long)hdr->args.offset);
1077 hdr->completion_ops->reschedule_io(hdr);
1081 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1082 dprintk("%s Reset task %5u for i/o through MDS "
1083 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1085 hdr->inode->i_sb->s_id,
1086 (unsigned long long)NFS_FILEID(hdr->inode),
1088 (unsigned long long)hdr->args.offset);
1090 trace_pnfs_mds_fallback_write_done(hdr->inode,
1091 hdr->args.offset, hdr->args.count,
1092 IOMODE_RW, NFS_I(hdr->inode)->layout,
1094 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1098 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1100 struct rpc_task *task = &hdr->task;
1102 pnfs_layoutcommit_inode(hdr->inode, false);
1104 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1105 dprintk("%s Reset task %5u for i/o through MDS "
1106 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1108 hdr->inode->i_sb->s_id,
1109 (unsigned long long)NFS_FILEID(hdr->inode),
1111 (unsigned long long)hdr->args.offset);
1113 trace_pnfs_mds_fallback_read_done(hdr->inode,
1114 hdr->args.offset, hdr->args.count,
1115 IOMODE_READ, NFS_I(hdr->inode)->layout,
1117 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1121 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1122 struct nfs4_state *state,
1123 struct nfs_client *clp,
1124 struct pnfs_layout_segment *lseg,
1127 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1128 struct inode *inode = lo->plh_inode;
1129 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1130 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1132 switch (task->tk_status) {
1133 case -NFS4ERR_BADSESSION:
1134 case -NFS4ERR_BADSLOT:
1135 case -NFS4ERR_BAD_HIGH_SLOT:
1136 case -NFS4ERR_DEADSESSION:
1137 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1138 case -NFS4ERR_SEQ_FALSE_RETRY:
1139 case -NFS4ERR_SEQ_MISORDERED:
1140 dprintk("%s ERROR %d, Reset session. Exchangeid "
1141 "flags 0x%x\n", __func__, task->tk_status,
1142 clp->cl_exchange_flags);
1143 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1145 case -NFS4ERR_DELAY:
1146 case -NFS4ERR_GRACE:
1147 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1149 case -NFS4ERR_RETRY_UNCACHED_REP:
1152 return -NFS4ERR_RESET_TO_PNFS;
1153 /* Invalidate Layout errors */
1154 case -NFS4ERR_PNFS_NO_LAYOUT:
1155 case -ESTALE: /* mapped NFS4ERR_STALE */
1156 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1157 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1158 case -NFS4ERR_FHEXPIRED:
1159 case -NFS4ERR_WRONG_TYPE:
1160 dprintk("%s Invalid layout error %d\n", __func__,
1163 * Destroy layout so new i/o will get a new layout.
1164 * Layout will not be destroyed until all current lseg
1165 * references are put. Mark layout as invalid to resend failed
1166 * i/o and all i/o waiting on the slot table to the MDS until
1167 * layout is destroyed and a new valid layout is obtained.
1169 pnfs_destroy_layout(NFS_I(inode));
1170 rpc_wake_up(&tbl->slot_tbl_waitq);
1172 /* RPC connection errors */
1180 dprintk("%s DS connection error %d\n", __func__,
1182 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1184 rpc_wake_up(&tbl->slot_tbl_waitq);
1187 if (ff_layout_avoid_mds_available_ds(lseg))
1188 return -NFS4ERR_RESET_TO_PNFS;
1190 dprintk("%s Retry through MDS. Error %d\n", __func__,
1192 return -NFS4ERR_RESET_TO_MDS;
1194 task->tk_status = 0;
1198 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1199 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1200 struct pnfs_layout_segment *lseg,
1203 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1205 switch (task->tk_status) {
1206 /* File access problems. Don't mark the device as unavailable */
1216 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1219 dprintk("%s DS connection error %d\n", __func__,
1221 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1224 /* FIXME: Need to prevent infinite looping here. */
1225 return -NFS4ERR_RESET_TO_PNFS;
1227 task->tk_status = 0;
1228 rpc_restart_call_prepare(task);
1229 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1233 static int ff_layout_async_handle_error(struct rpc_task *task,
1234 struct nfs4_state *state,
1235 struct nfs_client *clp,
1236 struct pnfs_layout_segment *lseg,
1239 int vers = clp->cl_nfs_mod->rpc_vers->number;
1241 if (task->tk_status >= 0) {
1242 ff_layout_mark_ds_reachable(lseg, idx);
1246 /* Handle the case of an invalid layout segment */
1247 if (!pnfs_is_valid_lseg(lseg))
1248 return -NFS4ERR_RESET_TO_PNFS;
1252 return ff_layout_async_handle_error_v3(task, lseg, idx);
1254 return ff_layout_async_handle_error_v4(task, state, clp,
1257 /* should never happen */
1263 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1264 int idx, u64 offset, u64 length,
1265 u32 status, int opnum, int error)
1267 struct nfs4_ff_layout_mirror *mirror;
1274 case -EPROTONOSUPPORT:
1285 status = NFS4ERR_NXIO;
1288 status = NFS4ERR_ACCESS;
1303 mirror = FF_LAYOUT_COMP(lseg, idx);
1304 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1305 mirror, offset, length, status, opnum,
1307 if (status == NFS4ERR_NXIO)
1308 ff_layout_mark_ds_unreachable(lseg, idx);
1309 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1310 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1313 /* NFS_PROTO call done callback routines */
1314 static int ff_layout_read_done_cb(struct rpc_task *task,
1315 struct nfs_pgio_header *hdr)
1317 int new_idx = hdr->pgio_mirror_idx;
1320 trace_nfs4_pnfs_read(hdr, task->tk_status);
1321 if (task->tk_status < 0)
1322 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1323 hdr->args.offset, hdr->args.count,
1324 hdr->res.op_status, OP_READ,
1326 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1327 hdr->ds_clp, hdr->lseg,
1328 hdr->pgio_mirror_idx);
1330 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1331 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1333 case -NFS4ERR_RESET_TO_PNFS:
1334 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1335 hdr->pgio_mirror_idx + 1,
1337 goto out_layouterror;
1338 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1339 return task->tk_status;
1340 case -NFS4ERR_RESET_TO_MDS:
1341 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1342 return task->tk_status;
1349 ff_layout_read_record_layoutstats_done(task, hdr);
1350 ff_layout_send_layouterror(hdr->lseg);
1351 hdr->pgio_mirror_idx = new_idx;
1353 rpc_restart_call_prepare(task);
1358 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1360 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1364 * We reference the rpc_cred of the first WRITE that triggers the need for
1365 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1366 * rfc5661 is not clear about which credential should be used.
1368 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1369 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1370 * we always send layoutcommit after DS writes.
1373 ff_layout_set_layoutcommit(struct inode *inode,
1374 struct pnfs_layout_segment *lseg,
1377 if (!ff_layout_need_layoutcommit(lseg))
1380 pnfs_set_layoutcommit(inode, lseg, end_offset);
1381 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1382 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1385 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1386 struct nfs_pgio_header *hdr)
1388 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1390 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1391 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1396 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1397 struct nfs_pgio_header *hdr)
1399 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1401 nfs4_ff_layout_stat_io_end_read(task,
1402 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1405 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1408 static int ff_layout_read_prepare_common(struct rpc_task *task,
1409 struct nfs_pgio_header *hdr)
1411 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1412 rpc_exit(task, -EIO);
1416 ff_layout_read_record_layoutstats_start(task, hdr);
1421 * Call ops for the async read/write cases
1422 * In the case of dense layouts, the offset needs to be reset to its
1425 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1427 struct nfs_pgio_header *hdr = data;
1429 if (ff_layout_read_prepare_common(task, hdr))
1432 rpc_call_start(task);
1435 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1437 struct nfs_pgio_header *hdr = data;
1439 if (nfs4_setup_sequence(hdr->ds_clp,
1440 &hdr->args.seq_args,
1445 ff_layout_read_prepare_common(task, hdr);
1449 ff_layout_io_prepare_transmit(struct rpc_task *task,
1452 struct nfs_pgio_header *hdr = data;
1454 if (!pnfs_is_valid_lseg(hdr->lseg))
1455 rpc_exit(task, -EAGAIN);
1458 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1460 struct nfs_pgio_header *hdr = data;
1462 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1464 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1465 task->tk_status == 0) {
1466 nfs4_sequence_done(task, &hdr->res.seq_res);
1470 /* Note this may cause RPC to be resent */
1471 hdr->mds_ops->rpc_call_done(task, hdr);
1474 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1476 struct nfs_pgio_header *hdr = data;
1478 ff_layout_read_record_layoutstats_done(task, hdr);
1479 rpc_count_iostats_metrics(task,
1480 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1483 static void ff_layout_read_release(void *data)
1485 struct nfs_pgio_header *hdr = data;
1487 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1488 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1489 ff_layout_send_layouterror(hdr->lseg);
1490 pnfs_read_resend_pnfs(hdr);
1491 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1492 ff_layout_reset_read(hdr);
1493 pnfs_generic_rw_release(data);
1497 static int ff_layout_write_done_cb(struct rpc_task *task,
1498 struct nfs_pgio_header *hdr)
1500 loff_t end_offs = 0;
1503 trace_nfs4_pnfs_write(hdr, task->tk_status);
1504 if (task->tk_status < 0)
1505 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1506 hdr->args.offset, hdr->args.count,
1507 hdr->res.op_status, OP_WRITE,
1509 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1510 hdr->ds_clp, hdr->lseg,
1511 hdr->pgio_mirror_idx);
1513 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1514 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1516 case -NFS4ERR_RESET_TO_PNFS:
1517 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1518 return task->tk_status;
1519 case -NFS4ERR_RESET_TO_MDS:
1520 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1521 return task->tk_status;
1526 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1527 hdr->res.verf->committed == NFS_DATA_SYNC)
1528 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1530 /* Note: if the write is unstable, don't set end_offs until commit */
1531 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1533 /* zero out fattr since we don't care DS attr at all */
1534 hdr->fattr.valid = 0;
1535 if (task->tk_status >= 0)
1536 nfs_writeback_update_inode(hdr);
1541 static int ff_layout_commit_done_cb(struct rpc_task *task,
1542 struct nfs_commit_data *data)
1546 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1547 if (task->tk_status < 0)
1548 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1549 data->args.offset, data->args.count,
1550 data->res.op_status, OP_COMMIT,
1552 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1553 data->lseg, data->ds_commit_index);
1556 case -NFS4ERR_RESET_TO_PNFS:
1557 pnfs_generic_prepare_to_resend_writes(data);
1559 case -NFS4ERR_RESET_TO_MDS:
1560 pnfs_generic_prepare_to_resend_writes(data);
1563 rpc_restart_call_prepare(task);
1567 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1572 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1573 struct nfs_pgio_header *hdr)
1575 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1577 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1578 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1583 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1584 struct nfs_pgio_header *hdr)
1586 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1588 nfs4_ff_layout_stat_io_end_write(task,
1589 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1590 hdr->args.count, hdr->res.count,
1591 hdr->res.verf->committed);
1592 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1595 static int ff_layout_write_prepare_common(struct rpc_task *task,
1596 struct nfs_pgio_header *hdr)
1598 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1599 rpc_exit(task, -EIO);
1603 ff_layout_write_record_layoutstats_start(task, hdr);
1607 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1609 struct nfs_pgio_header *hdr = data;
1611 if (ff_layout_write_prepare_common(task, hdr))
1614 rpc_call_start(task);
1617 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1619 struct nfs_pgio_header *hdr = data;
1621 if (nfs4_setup_sequence(hdr->ds_clp,
1622 &hdr->args.seq_args,
1627 ff_layout_write_prepare_common(task, hdr);
1630 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1632 struct nfs_pgio_header *hdr = data;
1634 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1635 task->tk_status == 0) {
1636 nfs4_sequence_done(task, &hdr->res.seq_res);
1640 /* Note this may cause RPC to be resent */
1641 hdr->mds_ops->rpc_call_done(task, hdr);
1644 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1646 struct nfs_pgio_header *hdr = data;
1648 ff_layout_write_record_layoutstats_done(task, hdr);
1649 rpc_count_iostats_metrics(task,
1650 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1653 static void ff_layout_write_release(void *data)
1655 struct nfs_pgio_header *hdr = data;
1657 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1658 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1659 ff_layout_send_layouterror(hdr->lseg);
1660 ff_layout_reset_write(hdr, true);
1661 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1662 ff_layout_reset_write(hdr, false);
1663 pnfs_generic_rw_release(data);
1666 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1667 struct nfs_commit_data *cdata)
1669 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1671 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1672 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1676 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1677 struct nfs_commit_data *cdata)
1679 struct nfs_page *req;
1682 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1685 if (task->tk_status == 0) {
1686 list_for_each_entry(req, &cdata->pages, wb_list)
1687 count += req->wb_bytes;
1689 nfs4_ff_layout_stat_io_end_write(task,
1690 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1691 count, count, NFS_FILE_SYNC);
1692 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1695 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1696 struct nfs_commit_data *cdata)
1698 ff_layout_commit_record_layoutstats_start(task, cdata);
1701 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1703 ff_layout_commit_prepare_common(task, data);
1704 rpc_call_start(task);
1707 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1709 struct nfs_commit_data *wdata = data;
1711 if (nfs4_setup_sequence(wdata->ds_clp,
1712 &wdata->args.seq_args,
1713 &wdata->res.seq_res,
1716 ff_layout_commit_prepare_common(task, data);
1719 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1721 pnfs_generic_write_commit_done(task, data);
1724 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1726 struct nfs_commit_data *cdata = data;
1728 ff_layout_commit_record_layoutstats_done(task, cdata);
1729 rpc_count_iostats_metrics(task,
1730 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1733 static void ff_layout_commit_release(void *data)
1735 struct nfs_commit_data *cdata = data;
1737 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1738 pnfs_generic_commit_release(data);
1741 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1742 .rpc_call_prepare = ff_layout_read_prepare_v3,
1743 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
1744 .rpc_call_done = ff_layout_read_call_done,
1745 .rpc_count_stats = ff_layout_read_count_stats,
1746 .rpc_release = ff_layout_read_release,
1749 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1750 .rpc_call_prepare = ff_layout_read_prepare_v4,
1751 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
1752 .rpc_call_done = ff_layout_read_call_done,
1753 .rpc_count_stats = ff_layout_read_count_stats,
1754 .rpc_release = ff_layout_read_release,
1757 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1758 .rpc_call_prepare = ff_layout_write_prepare_v3,
1759 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
1760 .rpc_call_done = ff_layout_write_call_done,
1761 .rpc_count_stats = ff_layout_write_count_stats,
1762 .rpc_release = ff_layout_write_release,
1765 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1766 .rpc_call_prepare = ff_layout_write_prepare_v4,
1767 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
1768 .rpc_call_done = ff_layout_write_call_done,
1769 .rpc_count_stats = ff_layout_write_count_stats,
1770 .rpc_release = ff_layout_write_release,
1773 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1774 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1775 .rpc_call_done = ff_layout_commit_done,
1776 .rpc_count_stats = ff_layout_commit_count_stats,
1777 .rpc_release = ff_layout_commit_release,
1780 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1781 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1782 .rpc_call_done = ff_layout_commit_done,
1783 .rpc_count_stats = ff_layout_commit_count_stats,
1784 .rpc_release = ff_layout_commit_release,
1787 static enum pnfs_try_status
1788 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1790 struct pnfs_layout_segment *lseg = hdr->lseg;
1791 struct nfs4_pnfs_ds *ds;
1792 struct rpc_clnt *ds_clnt;
1793 struct nfs4_ff_layout_mirror *mirror;
1794 const struct cred *ds_cred;
1795 loff_t offset = hdr->args.offset;
1796 u32 idx = hdr->pgio_mirror_idx;
1800 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1801 __func__, hdr->inode->i_ino,
1802 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1804 mirror = FF_LAYOUT_COMP(lseg, idx);
1805 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1809 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1811 if (IS_ERR(ds_clnt))
1814 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1818 vers = nfs4_ff_layout_ds_version(mirror);
1820 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1821 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1823 hdr->pgio_done_cb = ff_layout_read_done_cb;
1824 refcount_inc(&ds->ds_clp->cl_count);
1825 hdr->ds_clp = ds->ds_clp;
1826 fh = nfs4_ff_layout_select_ds_fh(mirror);
1830 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1833 * Note that if we ever decide to split across DSes,
1834 * then we may need to handle dense-like offsets.
1836 hdr->args.offset = offset;
1837 hdr->mds_offset = offset;
1839 /* Perform an asynchronous read to ds */
1840 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1841 vers == 3 ? &ff_layout_read_call_ops_v3 :
1842 &ff_layout_read_call_ops_v4,
1843 0, RPC_TASK_SOFTCONN);
1845 return PNFS_ATTEMPTED;
1848 if (ff_layout_avoid_mds_available_ds(lseg))
1849 return PNFS_TRY_AGAIN;
1850 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1851 hdr->args.offset, hdr->args.count,
1852 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1853 return PNFS_NOT_ATTEMPTED;
1856 /* Perform async writes. */
1857 static enum pnfs_try_status
1858 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1860 struct pnfs_layout_segment *lseg = hdr->lseg;
1861 struct nfs4_pnfs_ds *ds;
1862 struct rpc_clnt *ds_clnt;
1863 struct nfs4_ff_layout_mirror *mirror;
1864 const struct cred *ds_cred;
1865 loff_t offset = hdr->args.offset;
1868 int idx = hdr->pgio_mirror_idx;
1870 mirror = FF_LAYOUT_COMP(lseg, idx);
1871 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1875 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1877 if (IS_ERR(ds_clnt))
1880 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1884 vers = nfs4_ff_layout_ds_version(mirror);
1886 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1887 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1888 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1891 hdr->pgio_done_cb = ff_layout_write_done_cb;
1892 refcount_inc(&ds->ds_clp->cl_count);
1893 hdr->ds_clp = ds->ds_clp;
1894 hdr->ds_commit_idx = idx;
1895 fh = nfs4_ff_layout_select_ds_fh(mirror);
1899 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1902 * Note that if we ever decide to split across DSes,
1903 * then we may need to handle dense-like offsets.
1905 hdr->args.offset = offset;
1907 /* Perform an asynchronous write */
1908 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1909 vers == 3 ? &ff_layout_write_call_ops_v3 :
1910 &ff_layout_write_call_ops_v4,
1911 sync, RPC_TASK_SOFTCONN);
1913 return PNFS_ATTEMPTED;
1916 if (ff_layout_avoid_mds_available_ds(lseg))
1917 return PNFS_TRY_AGAIN;
1918 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1919 hdr->args.offset, hdr->args.count,
1920 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1921 return PNFS_NOT_ATTEMPTED;
1924 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1929 static struct nfs_fh *
1930 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1932 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1934 /* FIXME: Assume that there is only one NFS version available
1937 return &flseg->mirror_array[i]->fh_versions[0];
1940 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1942 struct pnfs_layout_segment *lseg = data->lseg;
1943 struct nfs4_pnfs_ds *ds;
1944 struct rpc_clnt *ds_clnt;
1945 struct nfs4_ff_layout_mirror *mirror;
1946 const struct cred *ds_cred;
1951 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1952 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1955 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1956 mirror = FF_LAYOUT_COMP(lseg, idx);
1957 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1961 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1963 if (IS_ERR(ds_clnt))
1966 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1970 vers = nfs4_ff_layout_ds_version(mirror);
1972 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1973 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1975 data->commit_done_cb = ff_layout_commit_done_cb;
1976 data->cred = ds_cred;
1977 refcount_inc(&ds->ds_clp->cl_count);
1978 data->ds_clp = ds->ds_clp;
1979 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1983 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1984 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1985 &ff_layout_commit_call_ops_v4,
1986 how, RPC_TASK_SOFTCONN);
1990 pnfs_generic_prepare_to_resend_writes(data);
1991 pnfs_generic_commit_release(data);
1996 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1997 int how, struct nfs_commit_info *cinfo)
1999 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
2000 ff_layout_initiate_commit);
2003 static struct pnfs_ds_commit_info *
2004 ff_layout_get_ds_info(struct inode *inode)
2006 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2011 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2015 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2017 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2021 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2022 const struct nfs4_layoutreturn_args *args,
2023 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2027 start = xdr_reserve_space(xdr, 4);
2028 if (unlikely(!start))
2031 *start = cpu_to_be32(ff_args->num_errors);
2032 /* This assume we always return _ALL_ layouts */
2033 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2037 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2039 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2043 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2044 const nfs4_stateid *stateid,
2045 const struct nfs42_layoutstat_devinfo *devinfo)
2049 p = xdr_reserve_space(xdr, 8 + 8);
2050 p = xdr_encode_hyper(p, devinfo->offset);
2051 p = xdr_encode_hyper(p, devinfo->length);
2052 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2053 p = xdr_reserve_space(xdr, 4*8);
2054 p = xdr_encode_hyper(p, devinfo->read_count);
2055 p = xdr_encode_hyper(p, devinfo->read_bytes);
2056 p = xdr_encode_hyper(p, devinfo->write_count);
2057 p = xdr_encode_hyper(p, devinfo->write_bytes);
2058 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2062 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2063 const nfs4_stateid *stateid,
2064 const struct nfs42_layoutstat_devinfo *devinfo)
2066 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2067 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2068 devinfo->ld_private.data);
2071 /* report nothing for now */
2072 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2073 const struct nfs4_layoutreturn_args *args,
2074 struct nfs4_flexfile_layoutreturn_args *ff_args)
2079 p = xdr_reserve_space(xdr, 4);
2080 *p = cpu_to_be32(ff_args->num_dev);
2081 for (i = 0; i < ff_args->num_dev; i++)
2082 ff_layout_encode_ff_iostat(xdr,
2083 &args->layout->plh_stateid,
2084 &ff_args->devinfo[i]);
2088 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2089 unsigned int num_entries)
2093 for (i = 0; i < num_entries; i++) {
2094 if (!devinfo[i].ld_private.ops)
2096 if (!devinfo[i].ld_private.ops->free)
2098 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2102 static struct nfs4_deviceid_node *
2103 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2104 struct pnfs_device *pdev, gfp_t gfp_flags)
2106 struct nfs4_ff_layout_ds *dsaddr;
2108 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2111 return &dsaddr->id_node;
2115 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2116 const void *voidargs,
2117 const struct nfs4_xdr_opaque_data *ff_opaque)
2119 const struct nfs4_layoutreturn_args *args = voidargs;
2120 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2121 struct xdr_buf tmp_buf = {
2124 .iov_base = page_address(ff_args->pages[0]),
2127 .buflen = PAGE_SIZE,
2129 struct xdr_stream tmp_xdr;
2132 dprintk("%s: Begin\n", __func__);
2134 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2136 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2137 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2139 start = xdr_reserve_space(xdr, 4);
2140 *start = cpu_to_be32(tmp_buf.len);
2141 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2143 dprintk("%s: Return\n", __func__);
2147 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2149 struct nfs4_flexfile_layoutreturn_args *ff_args;
2153 ff_args = args->data;
2156 ff_layout_free_ds_ioerr(&ff_args->errors);
2157 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2159 put_page(ff_args->pages[0]);
2163 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2164 .encode = ff_layout_encode_layoutreturn,
2165 .free = ff_layout_free_layoutreturn,
2169 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2171 struct nfs4_flexfile_layoutreturn_args *ff_args;
2172 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2174 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2177 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2178 if (!ff_args->pages[0])
2179 goto out_nomem_free;
2181 INIT_LIST_HEAD(&ff_args->errors);
2182 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2183 &args->range, &ff_args->errors,
2184 FF_LAYOUTRETURN_MAXERR);
2186 spin_lock(&args->inode->i_lock);
2187 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2188 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2189 spin_unlock(&args->inode->i_lock);
2191 args->ld_private->ops = &layoutreturn_ops;
2192 args->ld_private->data = ff_args;
2200 #ifdef CONFIG_NFS_V4_2
2202 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2204 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2205 struct nfs42_layout_error *errors;
2208 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2210 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2211 if (list_empty(&head))
2214 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2215 sizeof(*errors), GFP_NOFS);
2216 if (errors != NULL) {
2217 const struct nfs4_ff_layout_ds_err *pos;
2220 list_for_each_entry(pos, &head, list) {
2221 errors[n].offset = pos->offset;
2222 errors[n].length = pos->length;
2223 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2224 errors[n].errors[0].dev_id = pos->deviceid;
2225 errors[n].errors[0].status = pos->status;
2226 errors[n].errors[0].opnum = pos->opnum;
2228 if (!list_is_last(&pos->list, &head) &&
2229 n < NFS42_LAYOUTERROR_MAX)
2231 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2237 ff_layout_free_ds_ioerr(&head);
2241 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2247 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2249 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2251 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2255 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2258 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2259 const struct in6_addr *addr = &sin6->sin6_addr;
2262 * RFC 4291, Section 2.2.2
2264 * Shorthanded ANY address
2266 if (ipv6_addr_any(addr))
2267 return snprintf(buf, buflen, "::");
2270 * RFC 4291, Section 2.2.2
2272 * Shorthanded loopback address
2274 if (ipv6_addr_loopback(addr))
2275 return snprintf(buf, buflen, "::1");
2278 * RFC 4291, Section 2.2.3
2280 * Special presentation address format for mapped v4
2283 if (ipv6_addr_v4mapped(addr))
2284 return snprintf(buf, buflen, "::ffff:%pI4",
2285 &addr->s6_addr32[3]);
2288 * RFC 4291, Section 2.2.1
2290 return snprintf(buf, buflen, "%pI6c", addr);
2293 /* Derived from rpc_sockaddr2uaddr */
2295 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2297 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2298 char portbuf[RPCBIND_MAXUADDRPLEN];
2299 char addrbuf[RPCBIND_MAXUADDRLEN];
2301 unsigned short port;
2305 switch (sap->sa_family) {
2307 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2309 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2314 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2316 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2321 /* we only support tcp and tcp6 */
2326 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2327 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2329 p = xdr_reserve_space(xdr, 4 + netid_len);
2330 xdr_encode_opaque(p, netid, netid_len);
2332 p = xdr_reserve_space(xdr, 4 + len);
2333 xdr_encode_opaque(p, addrbuf, len);
2337 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2340 struct timespec64 ts;
2343 p = xdr_reserve_space(xdr, 12);
2344 ts = ktime_to_timespec64(t);
2345 p = xdr_encode_hyper(p, ts.tv_sec);
2346 *p++ = cpu_to_be32(ts.tv_nsec);
2350 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2351 struct nfs4_ff_io_stat *stat)
2355 p = xdr_reserve_space(xdr, 5 * 8);
2356 p = xdr_encode_hyper(p, stat->ops_requested);
2357 p = xdr_encode_hyper(p, stat->bytes_requested);
2358 p = xdr_encode_hyper(p, stat->ops_completed);
2359 p = xdr_encode_hyper(p, stat->bytes_completed);
2360 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2361 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2362 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2366 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2367 const struct nfs42_layoutstat_devinfo *devinfo,
2368 struct nfs4_ff_layout_mirror *mirror)
2370 struct nfs4_pnfs_ds_addr *da;
2371 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2372 struct nfs_fh *fh = &mirror->fh_versions[0];
2375 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2376 dprintk("%s: DS %s: encoding address %s\n",
2377 __func__, ds->ds_remotestr, da->da_remotestr);
2379 ff_layout_encode_netaddr(xdr, da);
2381 p = xdr_reserve_space(xdr, 4 + fh->size);
2382 xdr_encode_opaque(p, fh->data, fh->size);
2383 /* ff_io_latency4 read */
2384 spin_lock(&mirror->lock);
2385 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2386 /* ff_io_latency4 write */
2387 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2388 spin_unlock(&mirror->lock);
2390 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2392 p = xdr_reserve_space(xdr, 4);
2393 *p = cpu_to_be32(false);
2397 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2398 const struct nfs4_xdr_opaque_data *opaque)
2400 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2401 struct nfs42_layoutstat_devinfo, ld_private);
2404 /* layoutupdate length */
2405 start = xdr_reserve_space(xdr, 4);
2406 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2408 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2412 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2414 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2416 ff_layout_put_mirror(mirror);
2419 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2420 .encode = ff_layout_encode_layoutstats,
2421 .free = ff_layout_free_layoutstats,
2425 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2426 struct nfs42_layoutstat_devinfo *devinfo,
2429 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2430 struct nfs4_ff_layout_mirror *mirror;
2431 struct nfs4_deviceid_node *dev;
2434 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2437 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2439 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2441 /* mirror refcount put in cleanup_layoutstats */
2442 if (!refcount_inc_not_zero(&mirror->ref))
2444 dev = &mirror->mirror_ds->id_node;
2445 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2446 devinfo->offset = 0;
2447 devinfo->length = NFS4_MAX_UINT64;
2448 spin_lock(&mirror->lock);
2449 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2450 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2451 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2452 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2453 spin_unlock(&mirror->lock);
2454 devinfo->layout_type = LAYOUT_FLEX_FILES;
2455 devinfo->ld_private.ops = &layoutstat_ops;
2456 devinfo->ld_private.data = mirror;
2465 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2467 struct nfs4_flexfile_layout *ff_layout;
2468 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2470 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2471 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2475 spin_lock(&args->inode->i_lock);
2476 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2477 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2478 &args->devinfo[0], dev_count);
2479 spin_unlock(&args->inode->i_lock);
2480 if (!args->num_dev) {
2481 kfree(args->devinfo);
2482 args->devinfo = NULL;
2490 ff_layout_set_layoutdriver(struct nfs_server *server,
2491 const struct nfs_fh *dummy)
2493 #if IS_ENABLED(CONFIG_NFS_V4_2)
2494 server->caps |= NFS_CAP_LAYOUTSTATS;
2499 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2500 .id = LAYOUT_FLEX_FILES,
2501 .name = "LAYOUT_FLEX_FILES",
2502 .owner = THIS_MODULE,
2503 .flags = PNFS_LAYOUTGET_ON_OPEN,
2504 .max_layoutget_response = 4096, /* 1 page or so... */
2505 .set_layoutdriver = ff_layout_set_layoutdriver,
2506 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2507 .free_layout_hdr = ff_layout_free_layout_hdr,
2508 .alloc_lseg = ff_layout_alloc_lseg,
2509 .free_lseg = ff_layout_free_lseg,
2510 .add_lseg = ff_layout_add_lseg,
2511 .pg_read_ops = &ff_layout_pg_read_ops,
2512 .pg_write_ops = &ff_layout_pg_write_ops,
2513 .get_ds_info = ff_layout_get_ds_info,
2514 .free_deviceid_node = ff_layout_free_deviceid_node,
2515 .mark_request_commit = pnfs_layout_mark_request_commit,
2516 .clear_request_commit = pnfs_generic_clear_request_commit,
2517 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2518 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2519 .commit_pagelist = ff_layout_commit_pagelist,
2520 .read_pagelist = ff_layout_read_pagelist,
2521 .write_pagelist = ff_layout_write_pagelist,
2522 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2523 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2524 .sync = pnfs_nfs_generic_sync,
2525 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2528 static int __init nfs4flexfilelayout_init(void)
2530 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2532 return pnfs_register_layoutdriver(&flexfilelayout_type);
2535 static void __exit nfs4flexfilelayout_exit(void)
2537 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2539 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2542 MODULE_ALIAS("nfs-layouttype4-4");
2544 MODULE_LICENSE("GPL");
2545 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2547 module_init(nfs4flexfilelayout_init);
2548 module_exit(nfs4flexfilelayout_exit);
2550 module_param(io_maxretrans, ushort, 0644);
2551 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2552 "retries an I/O request before returning an error. ");