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_mount.h>
12 #include <linux/nfs_page.h>
13 #include <linux/module.h>
14 #include <linux/sched/mm.h>
16 #include <linux/sunrpc/metrics.h>
18 #include "flexfilelayout.h"
19 #include "../nfs4session.h"
20 #include "../nfs4idmap.h"
21 #include "../internal.h"
22 #include "../delegation.h"
23 #include "../nfs4trace.h"
24 #include "../iostat.h"
28 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
30 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
31 #define FF_LAYOUTRETURN_MAXERR 20
33 static unsigned short io_maxretrans;
35 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
36 struct nfs_pgio_header *hdr);
37 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
38 struct nfs42_layoutstat_devinfo *devinfo,
40 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
41 const struct nfs42_layoutstat_devinfo *devinfo,
42 struct nfs4_ff_layout_mirror *mirror);
44 static struct pnfs_layout_hdr *
45 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
47 struct nfs4_flexfile_layout *ffl;
49 ffl = kzalloc(sizeof(*ffl), gfp_flags);
51 INIT_LIST_HEAD(&ffl->error_list);
52 INIT_LIST_HEAD(&ffl->mirrors);
53 ffl->last_report_time = ktime_get();
54 return &ffl->generic_hdr;
60 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
62 struct nfs4_ff_layout_ds_err *err, *n;
64 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
69 kfree(FF_LAYOUT_FROM_HDR(lo));
72 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
76 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
77 if (unlikely(p == NULL))
79 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
80 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
81 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
82 p[0], p[1], p[2], p[3]);
86 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
90 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
93 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
94 nfs4_print_deviceid(devid);
98 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
102 p = xdr_inline_decode(xdr, 4);
105 fh->size = be32_to_cpup(p++);
106 if (fh->size > sizeof(struct nfs_fh)) {
107 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
112 p = xdr_inline_decode(xdr, fh->size);
115 memcpy(&fh->data, p, fh->size);
116 dprintk("%s: fh len %d\n", __func__, fh->size);
122 * Currently only stringified uids and gids are accepted.
123 * I.e., kerberos is not supported to the DSes, so no pricipals.
125 * That means that one common function will suffice, but when
126 * principals are added, this should be split to accomodate
127 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
130 decode_name(struct xdr_stream *xdr, u32 *id)
135 /* opaque_length(4)*/
136 p = xdr_inline_decode(xdr, 4);
139 len = be32_to_cpup(p++);
143 dprintk("%s: len %u\n", __func__, len);
146 p = xdr_inline_decode(xdr, len);
150 if (!nfs_map_string_to_numeric((char *)p, len, id))
156 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
157 const struct nfs4_ff_layout_mirror *m2)
161 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
163 for (i = 0; i < m1->fh_versions_cnt; i++) {
164 bool found_fh = false;
165 for (j = 0; j < m2->fh_versions_cnt; j++) {
166 if (nfs_compare_fh(&m1->fh_versions[i],
167 &m2->fh_versions[j]) == 0) {
178 static struct nfs4_ff_layout_mirror *
179 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
180 struct nfs4_ff_layout_mirror *mirror)
182 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
183 struct nfs4_ff_layout_mirror *pos;
184 struct inode *inode = lo->plh_inode;
186 spin_lock(&inode->i_lock);
187 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
188 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
190 if (!ff_mirror_match_fh(mirror, pos))
192 if (refcount_inc_not_zero(&pos->ref)) {
193 spin_unlock(&inode->i_lock);
197 list_add(&mirror->mirrors, &ff_layout->mirrors);
199 spin_unlock(&inode->i_lock);
204 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
207 if (mirror->layout == NULL)
209 inode = mirror->layout->plh_inode;
210 spin_lock(&inode->i_lock);
211 list_del(&mirror->mirrors);
212 spin_unlock(&inode->i_lock);
213 mirror->layout = NULL;
216 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
218 struct nfs4_ff_layout_mirror *mirror;
220 mirror = kzalloc(sizeof(*mirror), gfp_flags);
221 if (mirror != NULL) {
222 spin_lock_init(&mirror->lock);
223 refcount_set(&mirror->ref, 1);
224 INIT_LIST_HEAD(&mirror->mirrors);
229 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
231 const struct cred *cred;
233 ff_layout_remove_mirror(mirror);
234 kfree(mirror->fh_versions);
235 cred = rcu_access_pointer(mirror->ro_cred);
237 cred = rcu_access_pointer(mirror->rw_cred);
239 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
243 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
245 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
246 ff_layout_free_mirror(mirror);
249 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
253 if (fls->mirror_array) {
254 for (i = 0; i < fls->mirror_array_cnt; i++) {
255 /* normally mirror_ds is freed in
256 * .free_deviceid_node but we still do it here
257 * for .alloc_lseg error path */
258 ff_layout_put_mirror(fls->mirror_array[i]);
260 kfree(fls->mirror_array);
261 fls->mirror_array = NULL;
265 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
269 dprintk("--> %s\n", __func__);
271 /* FIXME: remove this check when layout segment support is added */
272 if (lgr->range.offset != 0 ||
273 lgr->range.length != NFS4_MAX_UINT64) {
274 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
279 dprintk("--> %s returns %d\n", __func__, ret);
283 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
286 ff_layout_free_mirror_array(fls);
292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
293 const struct pnfs_layout_range *l2)
297 if (l1->iomode != l2->iomode)
298 return l1->iomode != IOMODE_READ;
299 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
300 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
301 if (end1 < l2->offset)
303 if (end2 < l1->offset)
305 return l2->offset <= l1->offset;
309 ff_lseg_merge(struct pnfs_layout_segment *new,
310 struct pnfs_layout_segment *old)
312 u64 new_end, old_end;
314 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
316 if (new->pls_range.iomode != old->pls_range.iomode)
318 old_end = pnfs_calc_offset_end(old->pls_range.offset,
319 old->pls_range.length);
320 if (old_end < new->pls_range.offset)
322 new_end = pnfs_calc_offset_end(new->pls_range.offset,
323 new->pls_range.length);
324 if (new_end < old->pls_range.offset)
327 /* Mergeable: copy info from 'old' to 'new' */
328 if (new_end < old_end)
330 if (new->pls_range.offset < old->pls_range.offset)
331 new->pls_range.offset = old->pls_range.offset;
332 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
334 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
335 set_bit(NFS_LSEG_ROC, &new->pls_flags);
340 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
341 struct pnfs_layout_segment *lseg,
342 struct list_head *free_me)
344 pnfs_generic_layout_insert_lseg(lo, lseg,
345 ff_lseg_range_is_after,
350 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
354 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
355 for (j = i + 1; j < fls->mirror_array_cnt; j++)
356 if (fls->mirror_array[i]->efficiency <
357 fls->mirror_array[j]->efficiency)
358 swap(fls->mirror_array[i],
359 fls->mirror_array[j]);
363 static struct pnfs_layout_segment *
364 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
365 struct nfs4_layoutget_res *lgr,
368 struct pnfs_layout_segment *ret;
369 struct nfs4_ff_layout_segment *fls = NULL;
370 struct xdr_stream stream;
372 struct page *scratch;
374 u32 mirror_array_cnt;
378 dprintk("--> %s\n", __func__);
379 scratch = alloc_page(gfp_flags);
381 return ERR_PTR(-ENOMEM);
383 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
385 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
387 /* stripe unit and mirror_array_cnt */
389 p = xdr_inline_decode(&stream, 8 + 4);
393 p = xdr_decode_hyper(p, &stripe_unit);
394 mirror_array_cnt = be32_to_cpup(p++);
395 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
396 stripe_unit, mirror_array_cnt);
398 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
399 mirror_array_cnt == 0)
403 fls = kzalloc(sizeof(*fls), gfp_flags);
407 fls->mirror_array_cnt = mirror_array_cnt;
408 fls->stripe_unit = stripe_unit;
409 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
410 sizeof(fls->mirror_array[0]), gfp_flags);
411 if (fls->mirror_array == NULL)
414 for (i = 0; i < fls->mirror_array_cnt; i++) {
415 struct nfs4_ff_layout_mirror *mirror;
417 const struct cred __rcu *cred;
420 u32 ds_count, fh_count, id;
424 p = xdr_inline_decode(&stream, 4);
427 ds_count = be32_to_cpup(p);
429 /* FIXME: allow for striping? */
433 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
434 if (fls->mirror_array[i] == NULL) {
439 fls->mirror_array[i]->ds_count = ds_count;
442 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
448 p = xdr_inline_decode(&stream, 4);
451 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
454 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
460 p = xdr_inline_decode(&stream, 4);
463 fh_count = be32_to_cpup(p);
465 fls->mirror_array[i]->fh_versions =
466 kcalloc(fh_count, sizeof(struct nfs_fh),
468 if (fls->mirror_array[i]->fh_versions == NULL) {
473 for (j = 0; j < fh_count; j++) {
474 rc = decode_nfs_fh(&stream,
475 &fls->mirror_array[i]->fh_versions[j]);
480 fls->mirror_array[i]->fh_versions_cnt = fh_count;
483 rc = decode_name(&stream, &id);
487 uid = make_kuid(&init_user_ns, id);
490 rc = decode_name(&stream, &id);
494 gid = make_kgid(&init_user_ns, id);
496 if (gfp_flags & __GFP_FS)
497 kcred = prepare_kernel_cred(NULL);
499 unsigned int nofs_flags = memalloc_nofs_save();
500 kcred = prepare_kernel_cred(NULL);
501 memalloc_nofs_restore(nofs_flags);
508 cred = RCU_INITIALIZER(kcred);
510 if (lgr->range.iomode == IOMODE_READ)
511 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
513 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
515 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
516 if (mirror != fls->mirror_array[i]) {
517 /* swap cred ptrs so free_mirror will clean up old */
518 if (lgr->range.iomode == IOMODE_READ) {
519 cred = xchg(&mirror->ro_cred, cred);
520 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
522 cred = xchg(&mirror->rw_cred, cred);
523 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
525 ff_layout_free_mirror(fls->mirror_array[i]);
526 fls->mirror_array[i] = mirror;
529 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
530 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
531 from_kuid(&init_user_ns, uid),
532 from_kgid(&init_user_ns, gid));
535 p = xdr_inline_decode(&stream, 4);
537 goto out_sort_mirrors;
538 fls->flags = be32_to_cpup(p);
540 p = xdr_inline_decode(&stream, 4);
542 goto out_sort_mirrors;
543 for (i=0; i < fls->mirror_array_cnt; i++)
544 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
547 ff_layout_sort_mirrors(fls);
548 rc = ff_layout_check_layout(lgr);
551 ret = &fls->generic_hdr;
552 dprintk("<-- %s (success)\n", __func__);
554 __free_page(scratch);
557 _ff_layout_free_lseg(fls);
559 dprintk("<-- %s (%d)\n", __func__, rc);
563 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
565 struct pnfs_layout_segment *lseg;
567 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
568 if (lseg->pls_range.iomode == IOMODE_RW)
575 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
577 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
579 dprintk("--> %s\n", __func__);
581 if (lseg->pls_range.iomode == IOMODE_RW) {
582 struct nfs4_flexfile_layout *ffl;
585 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
586 inode = ffl->generic_hdr.plh_inode;
587 spin_lock(&inode->i_lock);
588 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
589 ffl->commit_info.nbuckets = 0;
590 kfree(ffl->commit_info.buckets);
591 ffl->commit_info.buckets = NULL;
593 spin_unlock(&inode->i_lock);
595 _ff_layout_free_lseg(fls);
598 /* Return 1 until we have multiple lsegs support */
600 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
606 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
608 /* first IO request? */
609 if (atomic_inc_return(&timer->n_ops) == 1) {
610 timer->start_time = now;
615 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
619 if (atomic_dec_return(&timer->n_ops) < 0)
622 start = timer->start_time;
623 timer->start_time = now;
624 return ktime_sub(now, start);
628 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
629 struct nfs4_ff_layoutstat *layoutstat,
632 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
633 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
635 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
636 if (!mirror->start_time)
637 mirror->start_time = now;
638 if (mirror->report_interval != 0)
639 report_interval = (s64)mirror->report_interval * 1000LL;
640 else if (layoutstats_timer != 0)
641 report_interval = (s64)layoutstats_timer * 1000LL;
642 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
644 ffl->last_report_time = now;
652 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
655 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
657 iostat->ops_requested++;
658 iostat->bytes_requested += requested;
662 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
665 ktime_t time_completed,
666 ktime_t time_started)
668 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
669 ktime_t completion_time = ktime_sub(time_completed, time_started);
672 iostat->ops_completed++;
673 iostat->bytes_completed += completed;
674 iostat->bytes_not_delivered += requested - completed;
676 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
677 iostat->total_busy_time =
678 ktime_add(iostat->total_busy_time, timer);
679 iostat->aggregate_completion_time =
680 ktime_add(iostat->aggregate_completion_time,
685 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
686 struct nfs4_ff_layout_mirror *mirror,
687 __u64 requested, ktime_t now)
691 spin_lock(&mirror->lock);
692 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
693 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
694 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
695 spin_unlock(&mirror->lock);
698 pnfs_report_layoutstat(inode, GFP_KERNEL);
702 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
703 struct nfs4_ff_layout_mirror *mirror,
707 spin_lock(&mirror->lock);
708 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
709 requested, completed,
710 ktime_get(), task->tk_start);
711 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
712 spin_unlock(&mirror->lock);
716 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
717 struct nfs4_ff_layout_mirror *mirror,
718 __u64 requested, ktime_t now)
722 spin_lock(&mirror->lock);
723 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
724 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
725 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
726 spin_unlock(&mirror->lock);
729 pnfs_report_layoutstat(inode, GFP_NOIO);
733 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
734 struct nfs4_ff_layout_mirror *mirror,
737 enum nfs3_stable_how committed)
739 if (committed == NFS_UNSTABLE)
740 requested = completed = 0;
742 spin_lock(&mirror->lock);
743 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
744 requested, completed, ktime_get(), task->tk_start);
745 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
746 spin_unlock(&mirror->lock);
750 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
751 struct nfs_commit_info *cinfo,
754 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
755 struct pnfs_commit_bucket *buckets;
758 if (cinfo->ds->nbuckets != 0) {
759 /* This assumes there is only one RW lseg per file.
760 * To support multiple lseg per file, we need to
761 * change struct pnfs_commit_bucket to allow dynamic
762 * increasing nbuckets.
767 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
769 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
776 spin_lock(&cinfo->inode->i_lock);
777 if (cinfo->ds->nbuckets != 0)
780 cinfo->ds->buckets = buckets;
781 cinfo->ds->nbuckets = size;
782 for (i = 0; i < size; i++) {
783 INIT_LIST_HEAD(&buckets[i].written);
784 INIT_LIST_HEAD(&buckets[i].committing);
785 /* mark direct verifier as unset */
786 buckets[i].direct_verf.committed =
787 NFS_INVALID_STABLE_HOW;
790 spin_unlock(&cinfo->inode->i_lock);
796 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
798 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
801 nfs4_mark_deviceid_unavailable(devid);
805 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
807 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
810 nfs4_mark_deviceid_available(devid);
813 static struct nfs4_pnfs_ds *
814 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
815 int start_idx, int *best_idx,
818 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
819 struct nfs4_ff_layout_mirror *mirror;
820 struct nfs4_pnfs_ds *ds;
821 bool fail_return = false;
824 /* mirrors are initially sorted by efficiency */
825 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
826 if (idx+1 == fls->mirror_array_cnt)
827 fail_return = !check_device;
829 mirror = FF_LAYOUT_COMP(lseg, idx);
830 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
835 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
845 static struct nfs4_pnfs_ds *
846 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
847 int start_idx, int *best_idx)
849 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
852 static struct nfs4_pnfs_ds *
853 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
854 int start_idx, int *best_idx)
856 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
859 static struct nfs4_pnfs_ds *
860 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
861 int start_idx, int *best_idx)
863 struct nfs4_pnfs_ds *ds;
865 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
868 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
872 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
873 struct nfs_page *req,
876 pnfs_put_lseg(pgio->pg_lseg);
877 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
878 nfs_req_openctx(req),
884 if (IS_ERR(pgio->pg_lseg)) {
885 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
886 pgio->pg_lseg = NULL;
891 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
892 struct nfs_page *req)
894 struct nfs_pgio_mirror *pgm;
895 struct nfs4_ff_layout_mirror *mirror;
896 struct nfs4_pnfs_ds *ds;
900 pnfs_generic_pg_check_layout(pgio);
901 /* Use full layout for now */
902 if (!pgio->pg_lseg) {
903 ff_layout_pg_get_read(pgio, req, false);
907 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
908 ff_layout_pg_get_read(pgio, req, true);
913 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
915 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
917 pnfs_put_lseg(pgio->pg_lseg);
918 pgio->pg_lseg = NULL;
919 /* Sleep for 1 second before retrying */
924 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
926 pgio->pg_mirror_idx = ds_idx;
928 /* read always uses only one mirror - idx 0 for pgio layer */
929 pgm = &pgio->pg_mirrors[0];
930 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
932 if (NFS_SERVER(pgio->pg_inode)->flags &
933 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
934 pgio->pg_maxretrans = io_maxretrans;
937 if (pgio->pg_error < 0)
940 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
941 0, NFS4_MAX_UINT64, IOMODE_READ,
942 NFS_I(pgio->pg_inode)->layout,
944 pnfs_put_lseg(pgio->pg_lseg);
945 pgio->pg_lseg = NULL;
946 pgio->pg_maxretrans = 0;
947 nfs_pageio_reset_read_mds(pgio);
951 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
952 struct nfs_page *req)
954 struct nfs4_ff_layout_mirror *mirror;
955 struct nfs_pgio_mirror *pgm;
956 struct nfs_commit_info cinfo;
957 struct nfs4_pnfs_ds *ds;
962 pnfs_generic_pg_check_layout(pgio);
963 if (!pgio->pg_lseg) {
964 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
965 nfs_req_openctx(req),
971 if (IS_ERR(pgio->pg_lseg)) {
972 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
973 pgio->pg_lseg = NULL;
977 /* If no lseg, fall back to write through mds */
978 if (pgio->pg_lseg == NULL)
981 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
982 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
986 /* Use a direct mapping of ds_idx to pgio mirror_idx */
987 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
988 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
991 for (i = 0; i < pgio->pg_mirror_count; i++) {
992 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
993 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
995 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
997 pnfs_put_lseg(pgio->pg_lseg);
998 pgio->pg_lseg = NULL;
999 /* Sleep for 1 second before retrying */
1003 pgm = &pgio->pg_mirrors[i];
1004 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
1007 if (NFS_SERVER(pgio->pg_inode)->flags &
1008 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
1009 pgio->pg_maxretrans = io_maxretrans;
1013 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
1014 0, NFS4_MAX_UINT64, IOMODE_RW,
1015 NFS_I(pgio->pg_inode)->layout,
1017 pnfs_put_lseg(pgio->pg_lseg);
1018 pgio->pg_lseg = NULL;
1019 pgio->pg_maxretrans = 0;
1020 nfs_pageio_reset_write_mds(pgio);
1024 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1025 struct nfs_page *req)
1027 if (!pgio->pg_lseg) {
1028 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1029 nfs_req_openctx(req),
1035 if (IS_ERR(pgio->pg_lseg)) {
1036 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1037 pgio->pg_lseg = NULL;
1042 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1044 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
1045 0, NFS4_MAX_UINT64, IOMODE_RW,
1046 NFS_I(pgio->pg_inode)->layout,
1048 /* no lseg means that pnfs is not in use, so no mirroring here */
1049 nfs_pageio_reset_write_mds(pgio);
1054 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1055 .pg_init = ff_layout_pg_init_read,
1056 .pg_test = pnfs_generic_pg_test,
1057 .pg_doio = pnfs_generic_pg_readpages,
1058 .pg_cleanup = pnfs_generic_pg_cleanup,
1061 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1062 .pg_init = ff_layout_pg_init_write,
1063 .pg_test = pnfs_generic_pg_test,
1064 .pg_doio = pnfs_generic_pg_writepages,
1065 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1066 .pg_cleanup = pnfs_generic_pg_cleanup,
1069 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1071 struct rpc_task *task = &hdr->task;
1073 pnfs_layoutcommit_inode(hdr->inode, false);
1076 dprintk("%s Reset task %5u for i/o through pNFS "
1077 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1079 hdr->inode->i_sb->s_id,
1080 (unsigned long long)NFS_FILEID(hdr->inode),
1082 (unsigned long long)hdr->args.offset);
1084 hdr->completion_ops->reschedule_io(hdr);
1088 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1089 dprintk("%s Reset task %5u for i/o through MDS "
1090 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1092 hdr->inode->i_sb->s_id,
1093 (unsigned long long)NFS_FILEID(hdr->inode),
1095 (unsigned long long)hdr->args.offset);
1097 trace_pnfs_mds_fallback_write_done(hdr->inode,
1098 hdr->args.offset, hdr->args.count,
1099 IOMODE_RW, NFS_I(hdr->inode)->layout,
1101 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1105 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1107 struct rpc_task *task = &hdr->task;
1109 pnfs_layoutcommit_inode(hdr->inode, false);
1111 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1112 dprintk("%s Reset task %5u for i/o through MDS "
1113 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1115 hdr->inode->i_sb->s_id,
1116 (unsigned long long)NFS_FILEID(hdr->inode),
1118 (unsigned long long)hdr->args.offset);
1120 trace_pnfs_mds_fallback_read_done(hdr->inode,
1121 hdr->args.offset, hdr->args.count,
1122 IOMODE_READ, NFS_I(hdr->inode)->layout,
1124 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1128 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1129 struct nfs4_state *state,
1130 struct nfs_client *clp,
1131 struct pnfs_layout_segment *lseg,
1134 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1135 struct inode *inode = lo->plh_inode;
1136 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1137 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1139 switch (task->tk_status) {
1140 case -NFS4ERR_BADSESSION:
1141 case -NFS4ERR_BADSLOT:
1142 case -NFS4ERR_BAD_HIGH_SLOT:
1143 case -NFS4ERR_DEADSESSION:
1144 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1145 case -NFS4ERR_SEQ_FALSE_RETRY:
1146 case -NFS4ERR_SEQ_MISORDERED:
1147 dprintk("%s ERROR %d, Reset session. Exchangeid "
1148 "flags 0x%x\n", __func__, task->tk_status,
1149 clp->cl_exchange_flags);
1150 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1152 case -NFS4ERR_DELAY:
1153 case -NFS4ERR_GRACE:
1154 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1156 case -NFS4ERR_RETRY_UNCACHED_REP:
1158 /* Invalidate Layout errors */
1159 case -NFS4ERR_PNFS_NO_LAYOUT:
1160 case -ESTALE: /* mapped NFS4ERR_STALE */
1161 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1162 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1163 case -NFS4ERR_FHEXPIRED:
1164 case -NFS4ERR_WRONG_TYPE:
1165 dprintk("%s Invalid layout error %d\n", __func__,
1168 * Destroy layout so new i/o will get a new layout.
1169 * Layout will not be destroyed until all current lseg
1170 * references are put. Mark layout as invalid to resend failed
1171 * i/o and all i/o waiting on the slot table to the MDS until
1172 * layout is destroyed and a new valid layout is obtained.
1174 pnfs_destroy_layout(NFS_I(inode));
1175 rpc_wake_up(&tbl->slot_tbl_waitq);
1177 /* RPC connection errors */
1185 dprintk("%s DS connection error %d\n", __func__,
1187 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1189 rpc_wake_up(&tbl->slot_tbl_waitq);
1192 if (ff_layout_avoid_mds_available_ds(lseg))
1193 return -NFS4ERR_RESET_TO_PNFS;
1195 dprintk("%s Retry through MDS. Error %d\n", __func__,
1197 return -NFS4ERR_RESET_TO_MDS;
1199 task->tk_status = 0;
1203 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1204 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1205 struct pnfs_layout_segment *lseg,
1208 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1210 switch (task->tk_status) {
1211 /* File access problems. Don't mark the device as unavailable */
1220 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1223 dprintk("%s DS connection error %d\n", __func__,
1225 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1228 /* FIXME: Need to prevent infinite looping here. */
1229 return -NFS4ERR_RESET_TO_PNFS;
1231 task->tk_status = 0;
1232 rpc_restart_call_prepare(task);
1233 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1237 static int ff_layout_async_handle_error(struct rpc_task *task,
1238 struct nfs4_state *state,
1239 struct nfs_client *clp,
1240 struct pnfs_layout_segment *lseg,
1243 int vers = clp->cl_nfs_mod->rpc_vers->number;
1245 if (task->tk_status >= 0) {
1246 ff_layout_mark_ds_reachable(lseg, idx);
1250 /* Handle the case of an invalid layout segment */
1251 if (!pnfs_is_valid_lseg(lseg))
1252 return -NFS4ERR_RESET_TO_PNFS;
1256 return ff_layout_async_handle_error_v3(task, lseg, idx);
1258 return ff_layout_async_handle_error_v4(task, state, clp,
1261 /* should never happen */
1267 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1268 int idx, u64 offset, u64 length,
1269 u32 status, int opnum, int error)
1271 struct nfs4_ff_layout_mirror *mirror;
1278 case -EPROTONOSUPPORT:
1289 status = NFS4ERR_NXIO;
1292 status = NFS4ERR_ACCESS;
1307 mirror = FF_LAYOUT_COMP(lseg, idx);
1308 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1309 mirror, offset, length, status, opnum,
1311 if (status == NFS4ERR_NXIO)
1312 ff_layout_mark_ds_unreachable(lseg, idx);
1313 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1314 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1317 /* NFS_PROTO call done callback routines */
1318 static int ff_layout_read_done_cb(struct rpc_task *task,
1319 struct nfs_pgio_header *hdr)
1321 int new_idx = hdr->pgio_mirror_idx;
1324 trace_nfs4_pnfs_read(hdr, task->tk_status);
1325 if (task->tk_status < 0)
1326 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1327 hdr->args.offset, hdr->args.count,
1328 hdr->res.op_status, OP_READ,
1330 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1331 hdr->ds_clp, hdr->lseg,
1332 hdr->pgio_mirror_idx);
1334 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1335 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1337 case -NFS4ERR_RESET_TO_PNFS:
1338 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1339 hdr->pgio_mirror_idx + 1,
1341 goto out_layouterror;
1342 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1343 return task->tk_status;
1344 case -NFS4ERR_RESET_TO_MDS:
1345 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1346 return task->tk_status;
1353 ff_layout_read_record_layoutstats_done(task, hdr);
1354 ff_layout_send_layouterror(hdr->lseg);
1355 hdr->pgio_mirror_idx = new_idx;
1357 rpc_restart_call_prepare(task);
1362 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1364 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1368 * We reference the rpc_cred of the first WRITE that triggers the need for
1369 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1370 * rfc5661 is not clear about which credential should be used.
1372 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1373 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1374 * we always send layoutcommit after DS writes.
1377 ff_layout_set_layoutcommit(struct inode *inode,
1378 struct pnfs_layout_segment *lseg,
1381 if (!ff_layout_need_layoutcommit(lseg))
1384 pnfs_set_layoutcommit(inode, lseg, end_offset);
1385 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1386 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1389 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1390 struct nfs_pgio_header *hdr)
1392 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1394 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1395 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1400 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1401 struct nfs_pgio_header *hdr)
1403 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1405 nfs4_ff_layout_stat_io_end_read(task,
1406 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1409 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1412 static int ff_layout_read_prepare_common(struct rpc_task *task,
1413 struct nfs_pgio_header *hdr)
1415 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1416 rpc_exit(task, -EIO);
1420 ff_layout_read_record_layoutstats_start(task, hdr);
1425 * Call ops for the async read/write cases
1426 * In the case of dense layouts, the offset needs to be reset to its
1429 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1431 struct nfs_pgio_header *hdr = data;
1433 if (ff_layout_read_prepare_common(task, hdr))
1436 rpc_call_start(task);
1439 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1441 struct nfs_pgio_header *hdr = data;
1443 if (nfs4_setup_sequence(hdr->ds_clp,
1444 &hdr->args.seq_args,
1449 ff_layout_read_prepare_common(task, hdr);
1452 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1454 struct nfs_pgio_header *hdr = data;
1456 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1458 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1459 task->tk_status == 0) {
1460 nfs4_sequence_done(task, &hdr->res.seq_res);
1464 /* Note this may cause RPC to be resent */
1465 hdr->mds_ops->rpc_call_done(task, hdr);
1468 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1470 struct nfs_pgio_header *hdr = data;
1472 ff_layout_read_record_layoutstats_done(task, hdr);
1473 rpc_count_iostats_metrics(task,
1474 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1477 static void ff_layout_read_release(void *data)
1479 struct nfs_pgio_header *hdr = data;
1481 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1482 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1483 ff_layout_send_layouterror(hdr->lseg);
1484 pnfs_read_resend_pnfs(hdr);
1485 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1486 ff_layout_reset_read(hdr);
1487 pnfs_generic_rw_release(data);
1491 static int ff_layout_write_done_cb(struct rpc_task *task,
1492 struct nfs_pgio_header *hdr)
1494 loff_t end_offs = 0;
1497 trace_nfs4_pnfs_write(hdr, task->tk_status);
1498 if (task->tk_status < 0)
1499 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1500 hdr->args.offset, hdr->args.count,
1501 hdr->res.op_status, OP_WRITE,
1503 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1504 hdr->ds_clp, hdr->lseg,
1505 hdr->pgio_mirror_idx);
1507 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1508 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1510 case -NFS4ERR_RESET_TO_PNFS:
1511 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1512 return task->tk_status;
1513 case -NFS4ERR_RESET_TO_MDS:
1514 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1515 return task->tk_status;
1520 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1521 hdr->res.verf->committed == NFS_DATA_SYNC)
1522 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1524 /* Note: if the write is unstable, don't set end_offs until commit */
1525 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1527 /* zero out fattr since we don't care DS attr at all */
1528 hdr->fattr.valid = 0;
1529 if (task->tk_status >= 0)
1530 nfs_writeback_update_inode(hdr);
1535 static int ff_layout_commit_done_cb(struct rpc_task *task,
1536 struct nfs_commit_data *data)
1540 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1541 if (task->tk_status < 0)
1542 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1543 data->args.offset, data->args.count,
1544 data->res.op_status, OP_COMMIT,
1546 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1547 data->lseg, data->ds_commit_index);
1550 case -NFS4ERR_RESET_TO_PNFS:
1551 pnfs_generic_prepare_to_resend_writes(data);
1553 case -NFS4ERR_RESET_TO_MDS:
1554 pnfs_generic_prepare_to_resend_writes(data);
1557 rpc_restart_call_prepare(task);
1561 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1566 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1567 struct nfs_pgio_header *hdr)
1569 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1571 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1572 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1577 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1578 struct nfs_pgio_header *hdr)
1580 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1582 nfs4_ff_layout_stat_io_end_write(task,
1583 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1584 hdr->args.count, hdr->res.count,
1585 hdr->res.verf->committed);
1586 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1589 static int ff_layout_write_prepare_common(struct rpc_task *task,
1590 struct nfs_pgio_header *hdr)
1592 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1593 rpc_exit(task, -EIO);
1597 ff_layout_write_record_layoutstats_start(task, hdr);
1601 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1603 struct nfs_pgio_header *hdr = data;
1605 if (ff_layout_write_prepare_common(task, hdr))
1608 rpc_call_start(task);
1611 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1613 struct nfs_pgio_header *hdr = data;
1615 if (nfs4_setup_sequence(hdr->ds_clp,
1616 &hdr->args.seq_args,
1621 ff_layout_write_prepare_common(task, hdr);
1624 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1626 struct nfs_pgio_header *hdr = data;
1628 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1629 task->tk_status == 0) {
1630 nfs4_sequence_done(task, &hdr->res.seq_res);
1634 /* Note this may cause RPC to be resent */
1635 hdr->mds_ops->rpc_call_done(task, hdr);
1638 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1640 struct nfs_pgio_header *hdr = data;
1642 ff_layout_write_record_layoutstats_done(task, hdr);
1643 rpc_count_iostats_metrics(task,
1644 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1647 static void ff_layout_write_release(void *data)
1649 struct nfs_pgio_header *hdr = data;
1651 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1652 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1653 ff_layout_send_layouterror(hdr->lseg);
1654 ff_layout_reset_write(hdr, true);
1655 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1656 ff_layout_reset_write(hdr, false);
1657 pnfs_generic_rw_release(data);
1660 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1661 struct nfs_commit_data *cdata)
1663 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1665 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1666 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1670 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1671 struct nfs_commit_data *cdata)
1673 struct nfs_page *req;
1676 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1679 if (task->tk_status == 0) {
1680 list_for_each_entry(req, &cdata->pages, wb_list)
1681 count += req->wb_bytes;
1683 nfs4_ff_layout_stat_io_end_write(task,
1684 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1685 count, count, NFS_FILE_SYNC);
1686 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1689 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1690 struct nfs_commit_data *cdata)
1692 ff_layout_commit_record_layoutstats_start(task, cdata);
1695 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1697 ff_layout_commit_prepare_common(task, data);
1698 rpc_call_start(task);
1701 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1703 struct nfs_commit_data *wdata = data;
1705 if (nfs4_setup_sequence(wdata->ds_clp,
1706 &wdata->args.seq_args,
1707 &wdata->res.seq_res,
1710 ff_layout_commit_prepare_common(task, data);
1713 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1715 pnfs_generic_write_commit_done(task, data);
1718 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1720 struct nfs_commit_data *cdata = data;
1722 ff_layout_commit_record_layoutstats_done(task, cdata);
1723 rpc_count_iostats_metrics(task,
1724 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1727 static void ff_layout_commit_release(void *data)
1729 struct nfs_commit_data *cdata = data;
1731 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1732 pnfs_generic_commit_release(data);
1735 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1736 .rpc_call_prepare = ff_layout_read_prepare_v3,
1737 .rpc_call_done = ff_layout_read_call_done,
1738 .rpc_count_stats = ff_layout_read_count_stats,
1739 .rpc_release = ff_layout_read_release,
1742 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1743 .rpc_call_prepare = ff_layout_read_prepare_v4,
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_write_call_ops_v3 = {
1750 .rpc_call_prepare = ff_layout_write_prepare_v3,
1751 .rpc_call_done = ff_layout_write_call_done,
1752 .rpc_count_stats = ff_layout_write_count_stats,
1753 .rpc_release = ff_layout_write_release,
1756 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1757 .rpc_call_prepare = ff_layout_write_prepare_v4,
1758 .rpc_call_done = ff_layout_write_call_done,
1759 .rpc_count_stats = ff_layout_write_count_stats,
1760 .rpc_release = ff_layout_write_release,
1763 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1764 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1765 .rpc_call_done = ff_layout_commit_done,
1766 .rpc_count_stats = ff_layout_commit_count_stats,
1767 .rpc_release = ff_layout_commit_release,
1770 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1771 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1772 .rpc_call_done = ff_layout_commit_done,
1773 .rpc_count_stats = ff_layout_commit_count_stats,
1774 .rpc_release = ff_layout_commit_release,
1777 static enum pnfs_try_status
1778 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1780 struct pnfs_layout_segment *lseg = hdr->lseg;
1781 struct nfs4_pnfs_ds *ds;
1782 struct rpc_clnt *ds_clnt;
1783 struct nfs4_ff_layout_mirror *mirror;
1784 const struct cred *ds_cred;
1785 loff_t offset = hdr->args.offset;
1786 u32 idx = hdr->pgio_mirror_idx;
1790 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1791 __func__, hdr->inode->i_ino,
1792 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1794 mirror = FF_LAYOUT_COMP(lseg, idx);
1795 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1799 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1801 if (IS_ERR(ds_clnt))
1804 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1808 vers = nfs4_ff_layout_ds_version(mirror);
1810 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1811 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1813 hdr->pgio_done_cb = ff_layout_read_done_cb;
1814 refcount_inc(&ds->ds_clp->cl_count);
1815 hdr->ds_clp = ds->ds_clp;
1816 fh = nfs4_ff_layout_select_ds_fh(mirror);
1820 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1823 * Note that if we ever decide to split across DSes,
1824 * then we may need to handle dense-like offsets.
1826 hdr->args.offset = offset;
1827 hdr->mds_offset = offset;
1829 /* Perform an asynchronous read to ds */
1830 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1831 vers == 3 ? &ff_layout_read_call_ops_v3 :
1832 &ff_layout_read_call_ops_v4,
1833 0, RPC_TASK_SOFTCONN);
1835 return PNFS_ATTEMPTED;
1838 if (ff_layout_avoid_mds_available_ds(lseg))
1839 return PNFS_TRY_AGAIN;
1840 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1841 hdr->args.offset, hdr->args.count,
1842 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1843 return PNFS_NOT_ATTEMPTED;
1846 /* Perform async writes. */
1847 static enum pnfs_try_status
1848 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1850 struct pnfs_layout_segment *lseg = hdr->lseg;
1851 struct nfs4_pnfs_ds *ds;
1852 struct rpc_clnt *ds_clnt;
1853 struct nfs4_ff_layout_mirror *mirror;
1854 const struct cred *ds_cred;
1855 loff_t offset = hdr->args.offset;
1858 int idx = hdr->pgio_mirror_idx;
1860 mirror = FF_LAYOUT_COMP(lseg, idx);
1861 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1865 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1867 if (IS_ERR(ds_clnt))
1870 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1874 vers = nfs4_ff_layout_ds_version(mirror);
1876 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1877 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1878 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1881 hdr->pgio_done_cb = ff_layout_write_done_cb;
1882 refcount_inc(&ds->ds_clp->cl_count);
1883 hdr->ds_clp = ds->ds_clp;
1884 hdr->ds_commit_idx = idx;
1885 fh = nfs4_ff_layout_select_ds_fh(mirror);
1889 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1892 * Note that if we ever decide to split across DSes,
1893 * then we may need to handle dense-like offsets.
1895 hdr->args.offset = offset;
1897 /* Perform an asynchronous write */
1898 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1899 vers == 3 ? &ff_layout_write_call_ops_v3 :
1900 &ff_layout_write_call_ops_v4,
1901 sync, RPC_TASK_SOFTCONN);
1903 return PNFS_ATTEMPTED;
1906 if (ff_layout_avoid_mds_available_ds(lseg))
1907 return PNFS_TRY_AGAIN;
1908 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1909 hdr->args.offset, hdr->args.count,
1910 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1911 return PNFS_NOT_ATTEMPTED;
1914 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1919 static struct nfs_fh *
1920 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1922 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1924 /* FIXME: Assume that there is only one NFS version available
1927 return &flseg->mirror_array[i]->fh_versions[0];
1930 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1932 struct pnfs_layout_segment *lseg = data->lseg;
1933 struct nfs4_pnfs_ds *ds;
1934 struct rpc_clnt *ds_clnt;
1935 struct nfs4_ff_layout_mirror *mirror;
1936 const struct cred *ds_cred;
1941 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1942 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1945 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1946 mirror = FF_LAYOUT_COMP(lseg, idx);
1947 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1951 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1953 if (IS_ERR(ds_clnt))
1956 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1960 vers = nfs4_ff_layout_ds_version(mirror);
1962 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1963 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1965 data->commit_done_cb = ff_layout_commit_done_cb;
1966 data->cred = ds_cred;
1967 refcount_inc(&ds->ds_clp->cl_count);
1968 data->ds_clp = ds->ds_clp;
1969 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1973 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1974 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1975 &ff_layout_commit_call_ops_v4,
1976 how, RPC_TASK_SOFTCONN);
1980 pnfs_generic_prepare_to_resend_writes(data);
1981 pnfs_generic_commit_release(data);
1986 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1987 int how, struct nfs_commit_info *cinfo)
1989 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1990 ff_layout_initiate_commit);
1993 static struct pnfs_ds_commit_info *
1994 ff_layout_get_ds_info(struct inode *inode)
1996 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2001 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2005 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2007 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2011 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2012 const struct nfs4_layoutreturn_args *args,
2013 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2017 start = xdr_reserve_space(xdr, 4);
2018 if (unlikely(!start))
2021 *start = cpu_to_be32(ff_args->num_errors);
2022 /* This assume we always return _ALL_ layouts */
2023 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2027 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2029 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2033 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2034 const nfs4_stateid *stateid,
2035 const struct nfs42_layoutstat_devinfo *devinfo)
2039 p = xdr_reserve_space(xdr, 8 + 8);
2040 p = xdr_encode_hyper(p, devinfo->offset);
2041 p = xdr_encode_hyper(p, devinfo->length);
2042 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2043 p = xdr_reserve_space(xdr, 4*8);
2044 p = xdr_encode_hyper(p, devinfo->read_count);
2045 p = xdr_encode_hyper(p, devinfo->read_bytes);
2046 p = xdr_encode_hyper(p, devinfo->write_count);
2047 p = xdr_encode_hyper(p, devinfo->write_bytes);
2048 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2052 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2053 const nfs4_stateid *stateid,
2054 const struct nfs42_layoutstat_devinfo *devinfo)
2056 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2057 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2058 devinfo->ld_private.data);
2061 /* report nothing for now */
2062 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2063 const struct nfs4_layoutreturn_args *args,
2064 struct nfs4_flexfile_layoutreturn_args *ff_args)
2069 p = xdr_reserve_space(xdr, 4);
2070 *p = cpu_to_be32(ff_args->num_dev);
2071 for (i = 0; i < ff_args->num_dev; i++)
2072 ff_layout_encode_ff_iostat(xdr,
2073 &args->layout->plh_stateid,
2074 &ff_args->devinfo[i]);
2078 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2079 unsigned int num_entries)
2083 for (i = 0; i < num_entries; i++) {
2084 if (!devinfo[i].ld_private.ops)
2086 if (!devinfo[i].ld_private.ops->free)
2088 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2092 static struct nfs4_deviceid_node *
2093 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2094 struct pnfs_device *pdev, gfp_t gfp_flags)
2096 struct nfs4_ff_layout_ds *dsaddr;
2098 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2101 return &dsaddr->id_node;
2105 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2106 const void *voidargs,
2107 const struct nfs4_xdr_opaque_data *ff_opaque)
2109 const struct nfs4_layoutreturn_args *args = voidargs;
2110 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2111 struct xdr_buf tmp_buf = {
2114 .iov_base = page_address(ff_args->pages[0]),
2117 .buflen = PAGE_SIZE,
2119 struct xdr_stream tmp_xdr;
2122 dprintk("%s: Begin\n", __func__);
2124 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2126 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2127 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2129 start = xdr_reserve_space(xdr, 4);
2130 *start = cpu_to_be32(tmp_buf.len);
2131 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2133 dprintk("%s: Return\n", __func__);
2137 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2139 struct nfs4_flexfile_layoutreturn_args *ff_args;
2143 ff_args = args->data;
2146 ff_layout_free_ds_ioerr(&ff_args->errors);
2147 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2149 put_page(ff_args->pages[0]);
2153 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2154 .encode = ff_layout_encode_layoutreturn,
2155 .free = ff_layout_free_layoutreturn,
2159 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2161 struct nfs4_flexfile_layoutreturn_args *ff_args;
2162 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2164 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2167 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2168 if (!ff_args->pages[0])
2169 goto out_nomem_free;
2171 INIT_LIST_HEAD(&ff_args->errors);
2172 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2173 &args->range, &ff_args->errors,
2174 FF_LAYOUTRETURN_MAXERR);
2176 spin_lock(&args->inode->i_lock);
2177 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2178 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2179 spin_unlock(&args->inode->i_lock);
2181 args->ld_private->ops = &layoutreturn_ops;
2182 args->ld_private->data = ff_args;
2190 #ifdef CONFIG_NFS_V4_2
2192 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2194 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2195 struct nfs42_layout_error *errors;
2198 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2200 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2201 if (list_empty(&head))
2204 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2205 sizeof(*errors), GFP_NOFS);
2206 if (errors != NULL) {
2207 const struct nfs4_ff_layout_ds_err *pos;
2210 list_for_each_entry(pos, &head, list) {
2211 errors[n].offset = pos->offset;
2212 errors[n].length = pos->length;
2213 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2214 errors[n].errors[0].dev_id = pos->deviceid;
2215 errors[n].errors[0].status = pos->status;
2216 errors[n].errors[0].opnum = pos->opnum;
2218 if (!list_is_last(&pos->list, &head) &&
2219 n < NFS42_LAYOUTERROR_MAX)
2221 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2227 ff_layout_free_ds_ioerr(&head);
2231 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2237 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2239 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2241 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2245 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2248 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2249 const struct in6_addr *addr = &sin6->sin6_addr;
2252 * RFC 4291, Section 2.2.2
2254 * Shorthanded ANY address
2256 if (ipv6_addr_any(addr))
2257 return snprintf(buf, buflen, "::");
2260 * RFC 4291, Section 2.2.2
2262 * Shorthanded loopback address
2264 if (ipv6_addr_loopback(addr))
2265 return snprintf(buf, buflen, "::1");
2268 * RFC 4291, Section 2.2.3
2270 * Special presentation address format for mapped v4
2273 if (ipv6_addr_v4mapped(addr))
2274 return snprintf(buf, buflen, "::ffff:%pI4",
2275 &addr->s6_addr32[3]);
2278 * RFC 4291, Section 2.2.1
2280 return snprintf(buf, buflen, "%pI6c", addr);
2283 /* Derived from rpc_sockaddr2uaddr */
2285 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2287 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2288 char portbuf[RPCBIND_MAXUADDRPLEN];
2289 char addrbuf[RPCBIND_MAXUADDRLEN];
2291 unsigned short port;
2295 switch (sap->sa_family) {
2297 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2299 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2304 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2306 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2311 /* we only support tcp and tcp6 */
2316 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2317 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2319 p = xdr_reserve_space(xdr, 4 + netid_len);
2320 xdr_encode_opaque(p, netid, netid_len);
2322 p = xdr_reserve_space(xdr, 4 + len);
2323 xdr_encode_opaque(p, addrbuf, len);
2327 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2330 struct timespec64 ts;
2333 p = xdr_reserve_space(xdr, 12);
2334 ts = ktime_to_timespec64(t);
2335 p = xdr_encode_hyper(p, ts.tv_sec);
2336 *p++ = cpu_to_be32(ts.tv_nsec);
2340 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2341 struct nfs4_ff_io_stat *stat)
2345 p = xdr_reserve_space(xdr, 5 * 8);
2346 p = xdr_encode_hyper(p, stat->ops_requested);
2347 p = xdr_encode_hyper(p, stat->bytes_requested);
2348 p = xdr_encode_hyper(p, stat->ops_completed);
2349 p = xdr_encode_hyper(p, stat->bytes_completed);
2350 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2351 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2352 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2356 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2357 const struct nfs42_layoutstat_devinfo *devinfo,
2358 struct nfs4_ff_layout_mirror *mirror)
2360 struct nfs4_pnfs_ds_addr *da;
2361 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2362 struct nfs_fh *fh = &mirror->fh_versions[0];
2365 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2366 dprintk("%s: DS %s: encoding address %s\n",
2367 __func__, ds->ds_remotestr, da->da_remotestr);
2369 ff_layout_encode_netaddr(xdr, da);
2371 p = xdr_reserve_space(xdr, 4 + fh->size);
2372 xdr_encode_opaque(p, fh->data, fh->size);
2373 /* ff_io_latency4 read */
2374 spin_lock(&mirror->lock);
2375 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2376 /* ff_io_latency4 write */
2377 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2378 spin_unlock(&mirror->lock);
2380 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2382 p = xdr_reserve_space(xdr, 4);
2383 *p = cpu_to_be32(false);
2387 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2388 const struct nfs4_xdr_opaque_data *opaque)
2390 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2391 struct nfs42_layoutstat_devinfo, ld_private);
2394 /* layoutupdate length */
2395 start = xdr_reserve_space(xdr, 4);
2396 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2398 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2402 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2404 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2406 ff_layout_put_mirror(mirror);
2409 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2410 .encode = ff_layout_encode_layoutstats,
2411 .free = ff_layout_free_layoutstats,
2415 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2416 struct nfs42_layoutstat_devinfo *devinfo,
2419 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2420 struct nfs4_ff_layout_mirror *mirror;
2421 struct nfs4_deviceid_node *dev;
2424 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2427 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2429 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2431 /* mirror refcount put in cleanup_layoutstats */
2432 if (!refcount_inc_not_zero(&mirror->ref))
2434 dev = &mirror->mirror_ds->id_node;
2435 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2436 devinfo->offset = 0;
2437 devinfo->length = NFS4_MAX_UINT64;
2438 spin_lock(&mirror->lock);
2439 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2440 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2441 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2442 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2443 spin_unlock(&mirror->lock);
2444 devinfo->layout_type = LAYOUT_FLEX_FILES;
2445 devinfo->ld_private.ops = &layoutstat_ops;
2446 devinfo->ld_private.data = mirror;
2455 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2457 struct nfs4_flexfile_layout *ff_layout;
2458 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2460 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2461 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2465 spin_lock(&args->inode->i_lock);
2466 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2467 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2468 &args->devinfo[0], dev_count);
2469 spin_unlock(&args->inode->i_lock);
2470 if (!args->num_dev) {
2471 kfree(args->devinfo);
2472 args->devinfo = NULL;
2480 ff_layout_set_layoutdriver(struct nfs_server *server,
2481 const struct nfs_fh *dummy)
2483 #if IS_ENABLED(CONFIG_NFS_V4_2)
2484 server->caps |= NFS_CAP_LAYOUTSTATS;
2489 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2490 .id = LAYOUT_FLEX_FILES,
2491 .name = "LAYOUT_FLEX_FILES",
2492 .owner = THIS_MODULE,
2493 .flags = PNFS_LAYOUTGET_ON_OPEN,
2494 .max_layoutget_response = 4096, /* 1 page or so... */
2495 .set_layoutdriver = ff_layout_set_layoutdriver,
2496 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2497 .free_layout_hdr = ff_layout_free_layout_hdr,
2498 .alloc_lseg = ff_layout_alloc_lseg,
2499 .free_lseg = ff_layout_free_lseg,
2500 .add_lseg = ff_layout_add_lseg,
2501 .pg_read_ops = &ff_layout_pg_read_ops,
2502 .pg_write_ops = &ff_layout_pg_write_ops,
2503 .get_ds_info = ff_layout_get_ds_info,
2504 .free_deviceid_node = ff_layout_free_deviceid_node,
2505 .mark_request_commit = pnfs_layout_mark_request_commit,
2506 .clear_request_commit = pnfs_generic_clear_request_commit,
2507 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2508 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2509 .commit_pagelist = ff_layout_commit_pagelist,
2510 .read_pagelist = ff_layout_read_pagelist,
2511 .write_pagelist = ff_layout_write_pagelist,
2512 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2513 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2514 .sync = pnfs_nfs_generic_sync,
2515 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2518 static int __init nfs4flexfilelayout_init(void)
2520 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2522 return pnfs_register_layoutdriver(&flexfilelayout_type);
2525 static void __exit nfs4flexfilelayout_exit(void)
2527 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2529 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2532 MODULE_ALIAS("nfs-layouttype4-4");
2534 MODULE_LICENSE("GPL");
2535 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2537 module_init(nfs4flexfilelayout_init);
2538 module_exit(nfs4flexfilelayout_exit);
2540 module_param(io_maxretrans, ushort, 0644);
2541 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2542 "retries an I/O request before returning an error. ");