1
2
3
4
5
6
7
8
9#include <linux/nfs_fs.h>
10#include <linux/nfs_mount.h>
11#include <linux/nfs_page.h>
12#include <linux/module.h>
13#include <linux/sched/mm.h>
14
15#include <linux/sunrpc/metrics.h>
16
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"
24#include "../nfs.h"
25#include "../nfs42.h"
26
27#define NFSDBG_FACILITY NFSDBG_PNFS_LD
28
29#define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
30#define FF_LAYOUTRETURN_MAXERR 20
31
32static unsigned short io_maxretrans;
33
34static const struct pnfs_commit_ops ff_layout_commit_ops;
35static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
36 struct nfs_pgio_header *hdr);
37static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
38 struct nfs42_layoutstat_devinfo *devinfo,
39 int dev_limit);
40static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
41 const struct nfs42_layoutstat_devinfo *devinfo,
42 struct nfs4_ff_layout_mirror *mirror);
43
44static struct pnfs_layout_hdr *
45ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
46{
47 struct nfs4_flexfile_layout *ffl;
48
49 ffl = kzalloc(sizeof(*ffl), gfp_flags);
50 if (ffl) {
51 pnfs_init_ds_commit_info(&ffl->commit_info);
52 INIT_LIST_HEAD(&ffl->error_list);
53 INIT_LIST_HEAD(&ffl->mirrors);
54 ffl->last_report_time = ktime_get();
55 ffl->commit_info.ops = &ff_layout_commit_ops;
56 return &ffl->generic_hdr;
57 } else
58 return NULL;
59}
60
61static void
62ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
63{
64 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
65 struct nfs4_ff_layout_ds_err *err, *n;
66
67 list_for_each_entry_safe(err, n, &ffl->error_list, list) {
68 list_del(&err->list);
69 kfree(err);
70 }
71 kfree_rcu(ffl, generic_hdr.plh_rcu);
72}
73
74static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
75{
76 __be32 *p;
77
78 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
79 if (unlikely(p == NULL))
80 return -ENOBUFS;
81 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
82 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
83 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
84 p[0], p[1], p[2], p[3]);
85 return 0;
86}
87
88static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
89{
90 __be32 *p;
91
92 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
93 if (unlikely(!p))
94 return -ENOBUFS;
95 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
96 nfs4_print_deviceid(devid);
97 return 0;
98}
99
100static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
101{
102 __be32 *p;
103
104 p = xdr_inline_decode(xdr, 4);
105 if (unlikely(!p))
106 return -ENOBUFS;
107 fh->size = be32_to_cpup(p++);
108 if (fh->size > NFS_MAXFHSIZE) {
109 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
110 fh->size);
111 return -EOVERFLOW;
112 }
113
114 p = xdr_inline_decode(xdr, fh->size);
115 if (unlikely(!p))
116 return -ENOBUFS;
117 memcpy(&fh->data, p, fh->size);
118 dprintk("%s: fh len %d\n", __func__, fh->size);
119
120 return 0;
121}
122
123
124
125
126
127
128
129
130
131static int
132decode_name(struct xdr_stream *xdr, u32 *id)
133{
134 __be32 *p;
135 int len;
136
137
138 p = xdr_inline_decode(xdr, 4);
139 if (unlikely(!p))
140 return -ENOBUFS;
141 len = be32_to_cpup(p++);
142 if (len < 0)
143 return -EINVAL;
144
145 dprintk("%s: len %u\n", __func__, len);
146
147
148 p = xdr_inline_decode(xdr, len);
149 if (unlikely(!p))
150 return -ENOBUFS;
151
152 if (!nfs_map_string_to_numeric((char *)p, len, id))
153 return -EINVAL;
154
155 return 0;
156}
157
158static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
159 const struct nfs4_ff_layout_mirror *m2)
160{
161 int i, j;
162
163 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
164 return false;
165 for (i = 0; i < m1->fh_versions_cnt; i++) {
166 bool found_fh = false;
167 for (j = 0; j < m2->fh_versions_cnt; j++) {
168 if (nfs_compare_fh(&m1->fh_versions[i],
169 &m2->fh_versions[j]) == 0) {
170 found_fh = true;
171 break;
172 }
173 }
174 if (!found_fh)
175 return false;
176 }
177 return true;
178}
179
180static struct nfs4_ff_layout_mirror *
181ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
182 struct nfs4_ff_layout_mirror *mirror)
183{
184 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
185 struct nfs4_ff_layout_mirror *pos;
186 struct inode *inode = lo->plh_inode;
187
188 spin_lock(&inode->i_lock);
189 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
190 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
191 continue;
192 if (!ff_mirror_match_fh(mirror, pos))
193 continue;
194 if (refcount_inc_not_zero(&pos->ref)) {
195 spin_unlock(&inode->i_lock);
196 return pos;
197 }
198 }
199 list_add(&mirror->mirrors, &ff_layout->mirrors);
200 mirror->layout = lo;
201 spin_unlock(&inode->i_lock);
202 return mirror;
203}
204
205static void
206ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
207{
208 struct inode *inode;
209 if (mirror->layout == NULL)
210 return;
211 inode = mirror->layout->plh_inode;
212 spin_lock(&inode->i_lock);
213 list_del(&mirror->mirrors);
214 spin_unlock(&inode->i_lock);
215 mirror->layout = NULL;
216}
217
218static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
219{
220 struct nfs4_ff_layout_mirror *mirror;
221
222 mirror = kzalloc(sizeof(*mirror), gfp_flags);
223 if (mirror != NULL) {
224 spin_lock_init(&mirror->lock);
225 refcount_set(&mirror->ref, 1);
226 INIT_LIST_HEAD(&mirror->mirrors);
227 }
228 return mirror;
229}
230
231static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
232{
233 const struct cred *cred;
234
235 ff_layout_remove_mirror(mirror);
236 kfree(mirror->fh_versions);
237 cred = rcu_access_pointer(mirror->ro_cred);
238 put_cred(cred);
239 cred = rcu_access_pointer(mirror->rw_cred);
240 put_cred(cred);
241 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
242 kfree(mirror);
243}
244
245static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
246{
247 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
248 ff_layout_free_mirror(mirror);
249}
250
251static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
252{
253 u32 i;
254
255 for (i = 0; i < fls->mirror_array_cnt; i++)
256 ff_layout_put_mirror(fls->mirror_array[i]);
257}
258
259static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
260{
261 if (fls) {
262 ff_layout_free_mirror_array(fls);
263 kfree(fls);
264 }
265}
266
267static bool
268ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
269 struct pnfs_layout_segment *l2)
270{
271 const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
272 const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
273 u32 i;
274
275 if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
276 return false;
277 for (i = 0; i < fl1->mirror_array_cnt; i++) {
278 if (fl1->mirror_array[i] != fl2->mirror_array[i])
279 return false;
280 }
281 return true;
282}
283
284static bool
285ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
286 const struct pnfs_layout_range *l2)
287{
288 u64 end1, end2;
289
290 if (l1->iomode != l2->iomode)
291 return l1->iomode != IOMODE_READ;
292 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
293 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
294 if (end1 < l2->offset)
295 return false;
296 if (end2 < l1->offset)
297 return true;
298 return l2->offset <= l1->offset;
299}
300
301static bool
302ff_lseg_merge(struct pnfs_layout_segment *new,
303 struct pnfs_layout_segment *old)
304{
305 u64 new_end, old_end;
306
307 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
308 return false;
309 if (new->pls_range.iomode != old->pls_range.iomode)
310 return false;
311 old_end = pnfs_calc_offset_end(old->pls_range.offset,
312 old->pls_range.length);
313 if (old_end < new->pls_range.offset)
314 return false;
315 new_end = pnfs_calc_offset_end(new->pls_range.offset,
316 new->pls_range.length);
317 if (new_end < old->pls_range.offset)
318 return false;
319 if (!ff_lseg_match_mirrors(new, old))
320 return false;
321
322
323 if (new_end < old_end)
324 new_end = old_end;
325 if (new->pls_range.offset < old->pls_range.offset)
326 new->pls_range.offset = old->pls_range.offset;
327 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
328 new_end);
329 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
330 set_bit(NFS_LSEG_ROC, &new->pls_flags);
331 return true;
332}
333
334static void
335ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
336 struct pnfs_layout_segment *lseg,
337 struct list_head *free_me)
338{
339 pnfs_generic_layout_insert_lseg(lo, lseg,
340 ff_lseg_range_is_after,
341 ff_lseg_merge,
342 free_me);
343}
344
345static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
346{
347 int i, j;
348
349 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
350 for (j = i + 1; j < fls->mirror_array_cnt; j++)
351 if (fls->mirror_array[i]->efficiency <
352 fls->mirror_array[j]->efficiency)
353 swap(fls->mirror_array[i],
354 fls->mirror_array[j]);
355 }
356}
357
358static struct pnfs_layout_segment *
359ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
360 struct nfs4_layoutget_res *lgr,
361 gfp_t gfp_flags)
362{
363 struct pnfs_layout_segment *ret;
364 struct nfs4_ff_layout_segment *fls = NULL;
365 struct xdr_stream stream;
366 struct xdr_buf buf;
367 struct page *scratch;
368 u64 stripe_unit;
369 u32 mirror_array_cnt;
370 __be32 *p;
371 int i, rc;
372
373 dprintk("--> %s\n", __func__);
374 scratch = alloc_page(gfp_flags);
375 if (!scratch)
376 return ERR_PTR(-ENOMEM);
377
378 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
379 lgr->layoutp->len);
380 xdr_set_scratch_page(&stream, scratch);
381
382
383 rc = -EIO;
384 p = xdr_inline_decode(&stream, 8 + 4);
385 if (!p)
386 goto out_err_free;
387
388 p = xdr_decode_hyper(p, &stripe_unit);
389 mirror_array_cnt = be32_to_cpup(p++);
390 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
391 stripe_unit, mirror_array_cnt);
392
393 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
394 mirror_array_cnt == 0)
395 goto out_err_free;
396
397 rc = -ENOMEM;
398 fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
399 gfp_flags);
400 if (!fls)
401 goto out_err_free;
402
403 fls->mirror_array_cnt = mirror_array_cnt;
404 fls->stripe_unit = stripe_unit;
405
406 for (i = 0; i < fls->mirror_array_cnt; i++) {
407 struct nfs4_ff_layout_mirror *mirror;
408 struct cred *kcred;
409 const struct cred __rcu *cred;
410 kuid_t uid;
411 kgid_t gid;
412 u32 ds_count, fh_count, id;
413 int j;
414
415 rc = -EIO;
416 p = xdr_inline_decode(&stream, 4);
417 if (!p)
418 goto out_err_free;
419 ds_count = be32_to_cpup(p);
420
421
422 if (ds_count != 1)
423 goto out_err_free;
424
425 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
426 if (fls->mirror_array[i] == NULL) {
427 rc = -ENOMEM;
428 goto out_err_free;
429 }
430
431 fls->mirror_array[i]->ds_count = ds_count;
432
433
434 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
435 if (rc)
436 goto out_err_free;
437
438
439 rc = -EIO;
440 p = xdr_inline_decode(&stream, 4);
441 if (!p)
442 goto out_err_free;
443 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
444
445
446 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
447 if (rc)
448 goto out_err_free;
449
450
451 rc = -EIO;
452 p = xdr_inline_decode(&stream, 4);
453 if (!p)
454 goto out_err_free;
455 fh_count = be32_to_cpup(p);
456
457 fls->mirror_array[i]->fh_versions =
458 kcalloc(fh_count, sizeof(struct nfs_fh),
459 gfp_flags);
460 if (fls->mirror_array[i]->fh_versions == NULL) {
461 rc = -ENOMEM;
462 goto out_err_free;
463 }
464
465 for (j = 0; j < fh_count; j++) {
466 rc = decode_nfs_fh(&stream,
467 &fls->mirror_array[i]->fh_versions[j]);
468 if (rc)
469 goto out_err_free;
470 }
471
472 fls->mirror_array[i]->fh_versions_cnt = fh_count;
473
474
475 rc = decode_name(&stream, &id);
476 if (rc)
477 goto out_err_free;
478
479 uid = make_kuid(&init_user_ns, id);
480
481
482 rc = decode_name(&stream, &id);
483 if (rc)
484 goto out_err_free;
485
486 gid = make_kgid(&init_user_ns, id);
487
488 if (gfp_flags & __GFP_FS)
489 kcred = prepare_kernel_cred(NULL);
490 else {
491 unsigned int nofs_flags = memalloc_nofs_save();
492 kcred = prepare_kernel_cred(NULL);
493 memalloc_nofs_restore(nofs_flags);
494 }
495 rc = -ENOMEM;
496 if (!kcred)
497 goto out_err_free;
498 kcred->fsuid = uid;
499 kcred->fsgid = gid;
500 cred = RCU_INITIALIZER(kcred);
501
502 if (lgr->range.iomode == IOMODE_READ)
503 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
504 else
505 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
506
507 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
508 if (mirror != fls->mirror_array[i]) {
509
510 if (lgr->range.iomode == IOMODE_READ) {
511 cred = xchg(&mirror->ro_cred, cred);
512 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
513 } else {
514 cred = xchg(&mirror->rw_cred, cred);
515 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
516 }
517 ff_layout_free_mirror(fls->mirror_array[i]);
518 fls->mirror_array[i] = mirror;
519 }
520
521 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
522 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
523 from_kuid(&init_user_ns, uid),
524 from_kgid(&init_user_ns, gid));
525 }
526
527 p = xdr_inline_decode(&stream, 4);
528 if (!p)
529 goto out_sort_mirrors;
530 fls->flags = be32_to_cpup(p);
531
532 p = xdr_inline_decode(&stream, 4);
533 if (!p)
534 goto out_sort_mirrors;
535 for (i=0; i < fls->mirror_array_cnt; i++)
536 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
537
538out_sort_mirrors:
539 ff_layout_sort_mirrors(fls);
540 ret = &fls->generic_hdr;
541 dprintk("<-- %s (success)\n", __func__);
542out_free_page:
543 __free_page(scratch);
544 return ret;
545out_err_free:
546 _ff_layout_free_lseg(fls);
547 ret = ERR_PTR(rc);
548 dprintk("<-- %s (%d)\n", __func__, rc);
549 goto out_free_page;
550}
551
552static void
553ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
554{
555 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
556
557 dprintk("--> %s\n", __func__);
558
559 if (lseg->pls_range.iomode == IOMODE_RW) {
560 struct nfs4_flexfile_layout *ffl;
561 struct inode *inode;
562
563 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
564 inode = ffl->generic_hdr.plh_inode;
565 spin_lock(&inode->i_lock);
566 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
567 spin_unlock(&inode->i_lock);
568 }
569 _ff_layout_free_lseg(fls);
570}
571
572static void
573nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
574{
575
576 if (atomic_inc_return(&timer->n_ops) == 1) {
577 timer->start_time = now;
578 }
579}
580
581static ktime_t
582nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
583{
584 ktime_t start;
585
586 if (atomic_dec_return(&timer->n_ops) < 0)
587 WARN_ON_ONCE(1);
588
589 start = timer->start_time;
590 timer->start_time = now;
591 return ktime_sub(now, start);
592}
593
594static bool
595nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
596 struct nfs4_ff_layoutstat *layoutstat,
597 ktime_t now)
598{
599 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
600 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
601
602 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
603 if (!mirror->start_time)
604 mirror->start_time = now;
605 if (mirror->report_interval != 0)
606 report_interval = (s64)mirror->report_interval * 1000LL;
607 else if (layoutstats_timer != 0)
608 report_interval = (s64)layoutstats_timer * 1000LL;
609 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
610 report_interval) {
611 ffl->last_report_time = now;
612 return true;
613 }
614
615 return false;
616}
617
618static void
619nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
620 __u64 requested)
621{
622 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
623
624 iostat->ops_requested++;
625 iostat->bytes_requested += requested;
626}
627
628static void
629nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
630 __u64 requested,
631 __u64 completed,
632 ktime_t time_completed,
633 ktime_t time_started)
634{
635 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
636 ktime_t completion_time = ktime_sub(time_completed, time_started);
637 ktime_t timer;
638
639 iostat->ops_completed++;
640 iostat->bytes_completed += completed;
641 iostat->bytes_not_delivered += requested - completed;
642
643 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
644 iostat->total_busy_time =
645 ktime_add(iostat->total_busy_time, timer);
646 iostat->aggregate_completion_time =
647 ktime_add(iostat->aggregate_completion_time,
648 completion_time);
649}
650
651static void
652nfs4_ff_layout_stat_io_start_read(struct inode *inode,
653 struct nfs4_ff_layout_mirror *mirror,
654 __u64 requested, ktime_t now)
655{
656 bool report;
657
658 spin_lock(&mirror->lock);
659 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
660 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
661 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
662 spin_unlock(&mirror->lock);
663
664 if (report)
665 pnfs_report_layoutstat(inode, GFP_KERNEL);
666}
667
668static void
669nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
670 struct nfs4_ff_layout_mirror *mirror,
671 __u64 requested,
672 __u64 completed)
673{
674 spin_lock(&mirror->lock);
675 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
676 requested, completed,
677 ktime_get(), task->tk_start);
678 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
679 spin_unlock(&mirror->lock);
680}
681
682static void
683nfs4_ff_layout_stat_io_start_write(struct inode *inode,
684 struct nfs4_ff_layout_mirror *mirror,
685 __u64 requested, ktime_t now)
686{
687 bool report;
688
689 spin_lock(&mirror->lock);
690 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
691 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
692 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
693 spin_unlock(&mirror->lock);
694
695 if (report)
696 pnfs_report_layoutstat(inode, GFP_NOIO);
697}
698
699static void
700nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
701 struct nfs4_ff_layout_mirror *mirror,
702 __u64 requested,
703 __u64 completed,
704 enum nfs3_stable_how committed)
705{
706 if (committed == NFS_UNSTABLE)
707 requested = completed = 0;
708
709 spin_lock(&mirror->lock);
710 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
711 requested, completed, ktime_get(), task->tk_start);
712 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
713 spin_unlock(&mirror->lock);
714}
715
716static void
717ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
718{
719 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
720
721 if (devid)
722 nfs4_mark_deviceid_unavailable(devid);
723}
724
725static void
726ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
727{
728 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
729
730 if (devid)
731 nfs4_mark_deviceid_available(devid);
732}
733
734static struct nfs4_pnfs_ds *
735ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
736 u32 start_idx, u32 *best_idx,
737 bool check_device)
738{
739 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
740 struct nfs4_ff_layout_mirror *mirror;
741 struct nfs4_pnfs_ds *ds;
742 u32 idx;
743
744
745 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
746 mirror = FF_LAYOUT_COMP(lseg, idx);
747 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
748 if (!ds)
749 continue;
750
751 if (check_device &&
752 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
753 continue;
754
755 *best_idx = idx;
756 return ds;
757 }
758
759 return NULL;
760}
761
762static struct nfs4_pnfs_ds *
763ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
764 u32 start_idx, u32 *best_idx)
765{
766 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
767}
768
769static struct nfs4_pnfs_ds *
770ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
771 u32 start_idx, u32 *best_idx)
772{
773 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
774}
775
776static struct nfs4_pnfs_ds *
777ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
778 u32 start_idx, u32 *best_idx)
779{
780 struct nfs4_pnfs_ds *ds;
781
782 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
783 if (ds)
784 return ds;
785 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
786}
787
788static struct nfs4_pnfs_ds *
789ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
790 u32 *best_idx)
791{
792 struct pnfs_layout_segment *lseg = pgio->pg_lseg;
793 struct nfs4_pnfs_ds *ds;
794
795 ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
796 best_idx);
797 if (ds || !pgio->pg_mirror_idx)
798 return ds;
799 return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
800}
801
802static void
803ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
804 struct nfs_page *req,
805 bool strict_iomode)
806{
807 pnfs_put_lseg(pgio->pg_lseg);
808 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
809 nfs_req_openctx(req),
810 req_offset(req),
811 req->wb_bytes,
812 IOMODE_READ,
813 strict_iomode,
814 GFP_KERNEL);
815 if (IS_ERR(pgio->pg_lseg)) {
816 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
817 pgio->pg_lseg = NULL;
818 }
819}
820
821static void
822ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
823 struct nfs_page *req)
824{
825 pnfs_generic_pg_check_layout(pgio);
826 pnfs_generic_pg_check_range(pgio, req);
827}
828
829static void
830ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
831 struct nfs_page *req)
832{
833 struct nfs_pgio_mirror *pgm;
834 struct nfs4_ff_layout_mirror *mirror;
835 struct nfs4_pnfs_ds *ds;
836 u32 ds_idx;
837
838retry:
839 ff_layout_pg_check_layout(pgio, req);
840
841 if (!pgio->pg_lseg) {
842 ff_layout_pg_get_read(pgio, req, false);
843 if (!pgio->pg_lseg)
844 goto out_nolseg;
845 }
846 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
847 ff_layout_pg_get_read(pgio, req, true);
848 if (!pgio->pg_lseg)
849 goto out_nolseg;
850 }
851
852 ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
853 if (!ds) {
854 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
855 goto out_mds;
856 pnfs_generic_pg_cleanup(pgio);
857
858 ssleep(1);
859 goto retry;
860 }
861
862 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
863 pgm = &pgio->pg_mirrors[0];
864 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
865
866 pgio->pg_mirror_idx = ds_idx;
867
868 if (NFS_SERVER(pgio->pg_inode)->flags &
869 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
870 pgio->pg_maxretrans = io_maxretrans;
871 return;
872out_nolseg:
873 if (pgio->pg_error < 0)
874 return;
875out_mds:
876 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
877 0, NFS4_MAX_UINT64, IOMODE_READ,
878 NFS_I(pgio->pg_inode)->layout,
879 pgio->pg_lseg);
880 pgio->pg_maxretrans = 0;
881 nfs_pageio_reset_read_mds(pgio);
882}
883
884static void
885ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
886 struct nfs_page *req)
887{
888 struct nfs4_ff_layout_mirror *mirror;
889 struct nfs_pgio_mirror *pgm;
890 struct nfs4_pnfs_ds *ds;
891 u32 i;
892
893retry:
894 ff_layout_pg_check_layout(pgio, req);
895 if (!pgio->pg_lseg) {
896 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
897 nfs_req_openctx(req),
898 req_offset(req),
899 req->wb_bytes,
900 IOMODE_RW,
901 false,
902 GFP_NOFS);
903 if (IS_ERR(pgio->pg_lseg)) {
904 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
905 pgio->pg_lseg = NULL;
906 return;
907 }
908 }
909
910 if (pgio->pg_lseg == NULL)
911 goto out_mds;
912
913
914 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
915 goto out_eagain;
916
917 for (i = 0; i < pgio->pg_mirror_count; i++) {
918 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
919 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
920 if (!ds) {
921 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
922 goto out_mds;
923 pnfs_generic_pg_cleanup(pgio);
924
925 ssleep(1);
926 goto retry;
927 }
928 pgm = &pgio->pg_mirrors[i];
929 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
930 }
931
932 if (NFS_SERVER(pgio->pg_inode)->flags &
933 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
934 pgio->pg_maxretrans = io_maxretrans;
935 return;
936out_eagain:
937 pnfs_generic_pg_cleanup(pgio);
938 pgio->pg_error = -EAGAIN;
939 return;
940out_mds:
941 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
942 0, NFS4_MAX_UINT64, IOMODE_RW,
943 NFS_I(pgio->pg_inode)->layout,
944 pgio->pg_lseg);
945 pgio->pg_maxretrans = 0;
946 nfs_pageio_reset_write_mds(pgio);
947 pgio->pg_error = -EAGAIN;
948}
949
950static unsigned int
951ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
952 struct nfs_page *req)
953{
954 if (!pgio->pg_lseg) {
955 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
956 nfs_req_openctx(req),
957 req_offset(req),
958 req->wb_bytes,
959 IOMODE_RW,
960 false,
961 GFP_NOFS);
962 if (IS_ERR(pgio->pg_lseg)) {
963 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
964 pgio->pg_lseg = NULL;
965 goto out;
966 }
967 }
968 if (pgio->pg_lseg)
969 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
970
971 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
972 0, NFS4_MAX_UINT64, IOMODE_RW,
973 NFS_I(pgio->pg_inode)->layout,
974 pgio->pg_lseg);
975
976 nfs_pageio_reset_write_mds(pgio);
977out:
978 return 1;
979}
980
981static u32
982ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
983{
984 u32 old = desc->pg_mirror_idx;
985
986 desc->pg_mirror_idx = idx;
987 return old;
988}
989
990static struct nfs_pgio_mirror *
991ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
992{
993 return &desc->pg_mirrors[idx];
994}
995
996static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
997 .pg_init = ff_layout_pg_init_read,
998 .pg_test = pnfs_generic_pg_test,
999 .pg_doio = pnfs_generic_pg_readpages,
1000 .pg_cleanup = pnfs_generic_pg_cleanup,
1001};
1002
1003static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1004 .pg_init = ff_layout_pg_init_write,
1005 .pg_test = pnfs_generic_pg_test,
1006 .pg_doio = pnfs_generic_pg_writepages,
1007 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1008 .pg_cleanup = pnfs_generic_pg_cleanup,
1009 .pg_get_mirror = ff_layout_pg_get_mirror_write,
1010 .pg_set_mirror = ff_layout_pg_set_mirror_write,
1011};
1012
1013static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1014{
1015 struct rpc_task *task = &hdr->task;
1016
1017 pnfs_layoutcommit_inode(hdr->inode, false);
1018
1019 if (retry_pnfs) {
1020 dprintk("%s Reset task %5u for i/o through pNFS "
1021 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1022 hdr->task.tk_pid,
1023 hdr->inode->i_sb->s_id,
1024 (unsigned long long)NFS_FILEID(hdr->inode),
1025 hdr->args.count,
1026 (unsigned long long)hdr->args.offset);
1027
1028 hdr->completion_ops->reschedule_io(hdr);
1029 return;
1030 }
1031
1032 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1033 dprintk("%s Reset task %5u for i/o through MDS "
1034 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1035 hdr->task.tk_pid,
1036 hdr->inode->i_sb->s_id,
1037 (unsigned long long)NFS_FILEID(hdr->inode),
1038 hdr->args.count,
1039 (unsigned long long)hdr->args.offset);
1040
1041 trace_pnfs_mds_fallback_write_done(hdr->inode,
1042 hdr->args.offset, hdr->args.count,
1043 IOMODE_RW, NFS_I(hdr->inode)->layout,
1044 hdr->lseg);
1045 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1046 }
1047}
1048
1049static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1050{
1051 u32 idx = hdr->pgio_mirror_idx + 1;
1052 u32 new_idx = 0;
1053
1054 if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx))
1055 ff_layout_send_layouterror(hdr->lseg);
1056 else
1057 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1058 pnfs_read_resend_pnfs(hdr, new_idx);
1059}
1060
1061static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1062{
1063 struct rpc_task *task = &hdr->task;
1064
1065 pnfs_layoutcommit_inode(hdr->inode, false);
1066 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1067
1068 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1069 dprintk("%s Reset task %5u for i/o through MDS "
1070 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1071 hdr->task.tk_pid,
1072 hdr->inode->i_sb->s_id,
1073 (unsigned long long)NFS_FILEID(hdr->inode),
1074 hdr->args.count,
1075 (unsigned long long)hdr->args.offset);
1076
1077 trace_pnfs_mds_fallback_read_done(hdr->inode,
1078 hdr->args.offset, hdr->args.count,
1079 IOMODE_READ, NFS_I(hdr->inode)->layout,
1080 hdr->lseg);
1081 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1082 }
1083}
1084
1085static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1086 struct nfs4_state *state,
1087 struct nfs_client *clp,
1088 struct pnfs_layout_segment *lseg,
1089 u32 idx)
1090{
1091 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1092 struct inode *inode = lo->plh_inode;
1093 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1094 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1095
1096 switch (task->tk_status) {
1097 case -NFS4ERR_BADSESSION:
1098 case -NFS4ERR_BADSLOT:
1099 case -NFS4ERR_BAD_HIGH_SLOT:
1100 case -NFS4ERR_DEADSESSION:
1101 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1102 case -NFS4ERR_SEQ_FALSE_RETRY:
1103 case -NFS4ERR_SEQ_MISORDERED:
1104 dprintk("%s ERROR %d, Reset session. Exchangeid "
1105 "flags 0x%x\n", __func__, task->tk_status,
1106 clp->cl_exchange_flags);
1107 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1108 break;
1109 case -NFS4ERR_DELAY:
1110 case -NFS4ERR_GRACE:
1111 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1112 break;
1113 case -NFS4ERR_RETRY_UNCACHED_REP:
1114 break;
1115
1116 case -NFS4ERR_PNFS_NO_LAYOUT:
1117 case -ESTALE:
1118 case -EBADHANDLE:
1119 case -EISDIR:
1120 case -NFS4ERR_FHEXPIRED:
1121 case -NFS4ERR_WRONG_TYPE:
1122 dprintk("%s Invalid layout error %d\n", __func__,
1123 task->tk_status);
1124
1125
1126
1127
1128
1129
1130
1131 pnfs_destroy_layout(NFS_I(inode));
1132 rpc_wake_up(&tbl->slot_tbl_waitq);
1133 goto reset;
1134
1135 case -ECONNREFUSED:
1136 case -EHOSTDOWN:
1137 case -EHOSTUNREACH:
1138 case -ENETUNREACH:
1139 case -EIO:
1140 case -ETIMEDOUT:
1141 case -EPIPE:
1142 dprintk("%s DS connection error %d\n", __func__,
1143 task->tk_status);
1144 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1145 &devid->deviceid);
1146 rpc_wake_up(&tbl->slot_tbl_waitq);
1147
1148 default:
1149 if (ff_layout_avoid_mds_available_ds(lseg))
1150 return -NFS4ERR_RESET_TO_PNFS;
1151reset:
1152 dprintk("%s Retry through MDS. Error %d\n", __func__,
1153 task->tk_status);
1154 return -NFS4ERR_RESET_TO_MDS;
1155 }
1156 task->tk_status = 0;
1157 return -EAGAIN;
1158}
1159
1160
1161static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1162 struct pnfs_layout_segment *lseg,
1163 u32 idx)
1164{
1165 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1166
1167 switch (task->tk_status) {
1168
1169 case -EACCES:
1170 case -ESTALE:
1171 case -EISDIR:
1172 case -EBADHANDLE:
1173 case -ELOOP:
1174 case -ENOSPC:
1175 break;
1176 case -EJUKEBOX:
1177 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1178 goto out_retry;
1179 default:
1180 dprintk("%s DS connection error %d\n", __func__,
1181 task->tk_status);
1182 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1183 &devid->deviceid);
1184 }
1185
1186 return -NFS4ERR_RESET_TO_PNFS;
1187out_retry:
1188 task->tk_status = 0;
1189 rpc_restart_call_prepare(task);
1190 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1191 return -EAGAIN;
1192}
1193
1194static int ff_layout_async_handle_error(struct rpc_task *task,
1195 struct nfs4_state *state,
1196 struct nfs_client *clp,
1197 struct pnfs_layout_segment *lseg,
1198 u32 idx)
1199{
1200 int vers = clp->cl_nfs_mod->rpc_vers->number;
1201
1202 if (task->tk_status >= 0) {
1203 ff_layout_mark_ds_reachable(lseg, idx);
1204 return 0;
1205 }
1206
1207
1208 if (!pnfs_is_valid_lseg(lseg))
1209 return -NFS4ERR_RESET_TO_PNFS;
1210
1211 switch (vers) {
1212 case 3:
1213 return ff_layout_async_handle_error_v3(task, lseg, idx);
1214 case 4:
1215 return ff_layout_async_handle_error_v4(task, state, clp,
1216 lseg, idx);
1217 default:
1218
1219 WARN_ON_ONCE(1);
1220 return 0;
1221 }
1222}
1223
1224static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1225 u32 idx, u64 offset, u64 length,
1226 u32 *op_status, int opnum, int error)
1227{
1228 struct nfs4_ff_layout_mirror *mirror;
1229 u32 status = *op_status;
1230 int err;
1231
1232 if (status == 0) {
1233 switch (error) {
1234 case -ETIMEDOUT:
1235 case -EPFNOSUPPORT:
1236 case -EPROTONOSUPPORT:
1237 case -EOPNOTSUPP:
1238 case -ECONNREFUSED:
1239 case -ECONNRESET:
1240 case -EHOSTDOWN:
1241 case -EHOSTUNREACH:
1242 case -ENETUNREACH:
1243 case -EADDRINUSE:
1244 case -ENOBUFS:
1245 case -EPIPE:
1246 case -EPERM:
1247 *op_status = status = NFS4ERR_NXIO;
1248 break;
1249 case -EACCES:
1250 *op_status = status = NFS4ERR_ACCESS;
1251 break;
1252 default:
1253 return;
1254 }
1255 }
1256
1257 mirror = FF_LAYOUT_COMP(lseg, idx);
1258 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1259 mirror, offset, length, status, opnum,
1260 GFP_NOIO);
1261
1262 switch (status) {
1263 case NFS4ERR_DELAY:
1264 case NFS4ERR_GRACE:
1265 break;
1266 case NFS4ERR_NXIO:
1267 ff_layout_mark_ds_unreachable(lseg, idx);
1268
1269
1270
1271
1272 if (opnum == OP_READ)
1273 break;
1274
1275 default:
1276 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1277 lseg);
1278 }
1279
1280 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1281}
1282
1283
1284static int ff_layout_read_done_cb(struct rpc_task *task,
1285 struct nfs_pgio_header *hdr)
1286{
1287 int err;
1288
1289 if (task->tk_status < 0) {
1290 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1291 hdr->args.offset, hdr->args.count,
1292 &hdr->res.op_status, OP_READ,
1293 task->tk_status);
1294 trace_ff_layout_read_error(hdr);
1295 }
1296
1297 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1298 hdr->ds_clp, hdr->lseg,
1299 hdr->pgio_mirror_idx);
1300
1301 trace_nfs4_pnfs_read(hdr, err);
1302 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1303 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1304 switch (err) {
1305 case -NFS4ERR_RESET_TO_PNFS:
1306 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1307 return task->tk_status;
1308 case -NFS4ERR_RESET_TO_MDS:
1309 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1310 return task->tk_status;
1311 case -EAGAIN:
1312 goto out_eagain;
1313 }
1314
1315 return 0;
1316out_eagain:
1317 rpc_restart_call_prepare(task);
1318 return -EAGAIN;
1319}
1320
1321static bool
1322ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1323{
1324 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1325}
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336static void
1337ff_layout_set_layoutcommit(struct inode *inode,
1338 struct pnfs_layout_segment *lseg,
1339 loff_t end_offset)
1340{
1341 if (!ff_layout_need_layoutcommit(lseg))
1342 return;
1343
1344 pnfs_set_layoutcommit(inode, lseg, end_offset);
1345 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1346 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1347}
1348
1349static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1350 struct nfs_pgio_header *hdr)
1351{
1352 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1353 return;
1354 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1355 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1356 hdr->args.count,
1357 task->tk_start);
1358}
1359
1360static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1361 struct nfs_pgio_header *hdr)
1362{
1363 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1364 return;
1365 nfs4_ff_layout_stat_io_end_read(task,
1366 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1367 hdr->args.count,
1368 hdr->res.count);
1369 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1370}
1371
1372static int ff_layout_read_prepare_common(struct rpc_task *task,
1373 struct nfs_pgio_header *hdr)
1374{
1375 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1376 rpc_exit(task, -EIO);
1377 return -EIO;
1378 }
1379
1380 ff_layout_read_record_layoutstats_start(task, hdr);
1381 return 0;
1382}
1383
1384
1385
1386
1387
1388
1389static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1390{
1391 struct nfs_pgio_header *hdr = data;
1392
1393 if (ff_layout_read_prepare_common(task, hdr))
1394 return;
1395
1396 rpc_call_start(task);
1397}
1398
1399static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1400{
1401 struct nfs_pgio_header *hdr = data;
1402
1403 if (nfs4_setup_sequence(hdr->ds_clp,
1404 &hdr->args.seq_args,
1405 &hdr->res.seq_res,
1406 task))
1407 return;
1408
1409 ff_layout_read_prepare_common(task, hdr);
1410}
1411
1412static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1413{
1414 struct nfs_pgio_header *hdr = data;
1415
1416 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1417
1418 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1419 task->tk_status == 0) {
1420 nfs4_sequence_done(task, &hdr->res.seq_res);
1421 return;
1422 }
1423
1424
1425 hdr->mds_ops->rpc_call_done(task, hdr);
1426}
1427
1428static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1429{
1430 struct nfs_pgio_header *hdr = data;
1431
1432 ff_layout_read_record_layoutstats_done(task, hdr);
1433 rpc_count_iostats_metrics(task,
1434 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1435}
1436
1437static void ff_layout_read_release(void *data)
1438{
1439 struct nfs_pgio_header *hdr = data;
1440
1441 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1442 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1443 ff_layout_resend_pnfs_read(hdr);
1444 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1445 ff_layout_reset_read(hdr);
1446 pnfs_generic_rw_release(data);
1447}
1448
1449
1450static int ff_layout_write_done_cb(struct rpc_task *task,
1451 struct nfs_pgio_header *hdr)
1452{
1453 loff_t end_offs = 0;
1454 int err;
1455
1456 if (task->tk_status < 0) {
1457 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1458 hdr->args.offset, hdr->args.count,
1459 &hdr->res.op_status, OP_WRITE,
1460 task->tk_status);
1461 trace_ff_layout_write_error(hdr);
1462 }
1463
1464 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1465 hdr->ds_clp, hdr->lseg,
1466 hdr->pgio_mirror_idx);
1467
1468 trace_nfs4_pnfs_write(hdr, err);
1469 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1470 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1471 switch (err) {
1472 case -NFS4ERR_RESET_TO_PNFS:
1473 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1474 return task->tk_status;
1475 case -NFS4ERR_RESET_TO_MDS:
1476 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1477 return task->tk_status;
1478 case -EAGAIN:
1479 return -EAGAIN;
1480 }
1481
1482 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1483 hdr->res.verf->committed == NFS_DATA_SYNC)
1484 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1485
1486
1487 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1488
1489
1490 hdr->fattr.valid = 0;
1491 if (task->tk_status >= 0)
1492 nfs_writeback_update_inode(hdr);
1493
1494 return 0;
1495}
1496
1497static int ff_layout_commit_done_cb(struct rpc_task *task,
1498 struct nfs_commit_data *data)
1499{
1500 int err;
1501
1502 if (task->tk_status < 0) {
1503 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1504 data->args.offset, data->args.count,
1505 &data->res.op_status, OP_COMMIT,
1506 task->tk_status);
1507 trace_ff_layout_commit_error(data);
1508 }
1509
1510 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1511 data->lseg, data->ds_commit_index);
1512
1513 trace_nfs4_pnfs_commit_ds(data, err);
1514 switch (err) {
1515 case -NFS4ERR_RESET_TO_PNFS:
1516 pnfs_generic_prepare_to_resend_writes(data);
1517 return -EAGAIN;
1518 case -NFS4ERR_RESET_TO_MDS:
1519 pnfs_generic_prepare_to_resend_writes(data);
1520 return -EAGAIN;
1521 case -EAGAIN:
1522 rpc_restart_call_prepare(task);
1523 return -EAGAIN;
1524 }
1525
1526 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1527
1528 return 0;
1529}
1530
1531static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1532 struct nfs_pgio_header *hdr)
1533{
1534 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1535 return;
1536 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1537 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1538 hdr->args.count,
1539 task->tk_start);
1540}
1541
1542static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1543 struct nfs_pgio_header *hdr)
1544{
1545 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1546 return;
1547 nfs4_ff_layout_stat_io_end_write(task,
1548 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1549 hdr->args.count, hdr->res.count,
1550 hdr->res.verf->committed);
1551 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1552}
1553
1554static int ff_layout_write_prepare_common(struct rpc_task *task,
1555 struct nfs_pgio_header *hdr)
1556{
1557 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1558 rpc_exit(task, -EIO);
1559 return -EIO;
1560 }
1561
1562 ff_layout_write_record_layoutstats_start(task, hdr);
1563 return 0;
1564}
1565
1566static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1567{
1568 struct nfs_pgio_header *hdr = data;
1569
1570 if (ff_layout_write_prepare_common(task, hdr))
1571 return;
1572
1573 rpc_call_start(task);
1574}
1575
1576static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1577{
1578 struct nfs_pgio_header *hdr = data;
1579
1580 if (nfs4_setup_sequence(hdr->ds_clp,
1581 &hdr->args.seq_args,
1582 &hdr->res.seq_res,
1583 task))
1584 return;
1585
1586 ff_layout_write_prepare_common(task, hdr);
1587}
1588
1589static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1590{
1591 struct nfs_pgio_header *hdr = data;
1592
1593 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1594 task->tk_status == 0) {
1595 nfs4_sequence_done(task, &hdr->res.seq_res);
1596 return;
1597 }
1598
1599
1600 hdr->mds_ops->rpc_call_done(task, hdr);
1601}
1602
1603static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1604{
1605 struct nfs_pgio_header *hdr = data;
1606
1607 ff_layout_write_record_layoutstats_done(task, hdr);
1608 rpc_count_iostats_metrics(task,
1609 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1610}
1611
1612static void ff_layout_write_release(void *data)
1613{
1614 struct nfs_pgio_header *hdr = data;
1615
1616 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1617 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1618 ff_layout_send_layouterror(hdr->lseg);
1619 ff_layout_reset_write(hdr, true);
1620 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1621 ff_layout_reset_write(hdr, false);
1622 pnfs_generic_rw_release(data);
1623}
1624
1625static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1626 struct nfs_commit_data *cdata)
1627{
1628 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1629 return;
1630 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1631 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1632 0, task->tk_start);
1633}
1634
1635static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1636 struct nfs_commit_data *cdata)
1637{
1638 struct nfs_page *req;
1639 __u64 count = 0;
1640
1641 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1642 return;
1643
1644 if (task->tk_status == 0) {
1645 list_for_each_entry(req, &cdata->pages, wb_list)
1646 count += req->wb_bytes;
1647 }
1648 nfs4_ff_layout_stat_io_end_write(task,
1649 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1650 count, count, NFS_FILE_SYNC);
1651 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1652}
1653
1654static void ff_layout_commit_prepare_common(struct rpc_task *task,
1655 struct nfs_commit_data *cdata)
1656{
1657 ff_layout_commit_record_layoutstats_start(task, cdata);
1658}
1659
1660static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1661{
1662 ff_layout_commit_prepare_common(task, data);
1663 rpc_call_start(task);
1664}
1665
1666static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1667{
1668 struct nfs_commit_data *wdata = data;
1669
1670 if (nfs4_setup_sequence(wdata->ds_clp,
1671 &wdata->args.seq_args,
1672 &wdata->res.seq_res,
1673 task))
1674 return;
1675 ff_layout_commit_prepare_common(task, data);
1676}
1677
1678static void ff_layout_commit_done(struct rpc_task *task, void *data)
1679{
1680 pnfs_generic_write_commit_done(task, data);
1681}
1682
1683static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1684{
1685 struct nfs_commit_data *cdata = data;
1686
1687 ff_layout_commit_record_layoutstats_done(task, cdata);
1688 rpc_count_iostats_metrics(task,
1689 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1690}
1691
1692static void ff_layout_commit_release(void *data)
1693{
1694 struct nfs_commit_data *cdata = data;
1695
1696 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1697 pnfs_generic_commit_release(data);
1698}
1699
1700static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1701 .rpc_call_prepare = ff_layout_read_prepare_v3,
1702 .rpc_call_done = ff_layout_read_call_done,
1703 .rpc_count_stats = ff_layout_read_count_stats,
1704 .rpc_release = ff_layout_read_release,
1705};
1706
1707static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1708 .rpc_call_prepare = ff_layout_read_prepare_v4,
1709 .rpc_call_done = ff_layout_read_call_done,
1710 .rpc_count_stats = ff_layout_read_count_stats,
1711 .rpc_release = ff_layout_read_release,
1712};
1713
1714static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1715 .rpc_call_prepare = ff_layout_write_prepare_v3,
1716 .rpc_call_done = ff_layout_write_call_done,
1717 .rpc_count_stats = ff_layout_write_count_stats,
1718 .rpc_release = ff_layout_write_release,
1719};
1720
1721static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1722 .rpc_call_prepare = ff_layout_write_prepare_v4,
1723 .rpc_call_done = ff_layout_write_call_done,
1724 .rpc_count_stats = ff_layout_write_count_stats,
1725 .rpc_release = ff_layout_write_release,
1726};
1727
1728static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1729 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1730 .rpc_call_done = ff_layout_commit_done,
1731 .rpc_count_stats = ff_layout_commit_count_stats,
1732 .rpc_release = ff_layout_commit_release,
1733};
1734
1735static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1736 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1737 .rpc_call_done = ff_layout_commit_done,
1738 .rpc_count_stats = ff_layout_commit_count_stats,
1739 .rpc_release = ff_layout_commit_release,
1740};
1741
1742static enum pnfs_try_status
1743ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1744{
1745 struct pnfs_layout_segment *lseg = hdr->lseg;
1746 struct nfs4_pnfs_ds *ds;
1747 struct rpc_clnt *ds_clnt;
1748 struct nfs4_ff_layout_mirror *mirror;
1749 const struct cred *ds_cred;
1750 loff_t offset = hdr->args.offset;
1751 u32 idx = hdr->pgio_mirror_idx;
1752 int vers;
1753 struct nfs_fh *fh;
1754
1755 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1756 __func__, hdr->inode->i_ino,
1757 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1758
1759 mirror = FF_LAYOUT_COMP(lseg, idx);
1760 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1761 if (!ds)
1762 goto out_failed;
1763
1764 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1765 hdr->inode);
1766 if (IS_ERR(ds_clnt))
1767 goto out_failed;
1768
1769 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1770 if (!ds_cred)
1771 goto out_failed;
1772
1773 vers = nfs4_ff_layout_ds_version(mirror);
1774
1775 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1776 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1777
1778 hdr->pgio_done_cb = ff_layout_read_done_cb;
1779 refcount_inc(&ds->ds_clp->cl_count);
1780 hdr->ds_clp = ds->ds_clp;
1781 fh = nfs4_ff_layout_select_ds_fh(mirror);
1782 if (fh)
1783 hdr->args.fh = fh;
1784
1785 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1786
1787
1788
1789
1790
1791 hdr->args.offset = offset;
1792 hdr->mds_offset = offset;
1793
1794
1795 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1796 vers == 3 ? &ff_layout_read_call_ops_v3 :
1797 &ff_layout_read_call_ops_v4,
1798 0, RPC_TASK_SOFTCONN);
1799 put_cred(ds_cred);
1800 return PNFS_ATTEMPTED;
1801
1802out_failed:
1803 if (ff_layout_avoid_mds_available_ds(lseg))
1804 return PNFS_TRY_AGAIN;
1805 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1806 hdr->args.offset, hdr->args.count,
1807 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1808 return PNFS_NOT_ATTEMPTED;
1809}
1810
1811
1812static enum pnfs_try_status
1813ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1814{
1815 struct pnfs_layout_segment *lseg = hdr->lseg;
1816 struct nfs4_pnfs_ds *ds;
1817 struct rpc_clnt *ds_clnt;
1818 struct nfs4_ff_layout_mirror *mirror;
1819 const struct cred *ds_cred;
1820 loff_t offset = hdr->args.offset;
1821 int vers;
1822 struct nfs_fh *fh;
1823 u32 idx = hdr->pgio_mirror_idx;
1824
1825 mirror = FF_LAYOUT_COMP(lseg, idx);
1826 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1827 if (!ds)
1828 goto out_failed;
1829
1830 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1831 hdr->inode);
1832 if (IS_ERR(ds_clnt))
1833 goto out_failed;
1834
1835 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1836 if (!ds_cred)
1837 goto out_failed;
1838
1839 vers = nfs4_ff_layout_ds_version(mirror);
1840
1841 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1842 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1843 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1844 vers);
1845
1846 hdr->pgio_done_cb = ff_layout_write_done_cb;
1847 refcount_inc(&ds->ds_clp->cl_count);
1848 hdr->ds_clp = ds->ds_clp;
1849 hdr->ds_commit_idx = idx;
1850 fh = nfs4_ff_layout_select_ds_fh(mirror);
1851 if (fh)
1852 hdr->args.fh = fh;
1853
1854 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1855
1856
1857
1858
1859
1860 hdr->args.offset = offset;
1861
1862
1863 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1864 vers == 3 ? &ff_layout_write_call_ops_v3 :
1865 &ff_layout_write_call_ops_v4,
1866 sync, RPC_TASK_SOFTCONN);
1867 put_cred(ds_cred);
1868 return PNFS_ATTEMPTED;
1869
1870out_failed:
1871 if (ff_layout_avoid_mds_available_ds(lseg))
1872 return PNFS_TRY_AGAIN;
1873 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1874 hdr->args.offset, hdr->args.count,
1875 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1876 return PNFS_NOT_ATTEMPTED;
1877}
1878
1879static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1880{
1881 return i;
1882}
1883
1884static struct nfs_fh *
1885select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1886{
1887 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1888
1889
1890
1891
1892 return &flseg->mirror_array[i]->fh_versions[0];
1893}
1894
1895static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1896{
1897 struct pnfs_layout_segment *lseg = data->lseg;
1898 struct nfs4_pnfs_ds *ds;
1899 struct rpc_clnt *ds_clnt;
1900 struct nfs4_ff_layout_mirror *mirror;
1901 const struct cred *ds_cred;
1902 u32 idx;
1903 int vers, ret;
1904 struct nfs_fh *fh;
1905
1906 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1907 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1908 goto out_err;
1909
1910 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1911 mirror = FF_LAYOUT_COMP(lseg, idx);
1912 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1913 if (!ds)
1914 goto out_err;
1915
1916 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1917 data->inode);
1918 if (IS_ERR(ds_clnt))
1919 goto out_err;
1920
1921 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1922 if (!ds_cred)
1923 goto out_err;
1924
1925 vers = nfs4_ff_layout_ds_version(mirror);
1926
1927 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1928 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1929 vers);
1930 data->commit_done_cb = ff_layout_commit_done_cb;
1931 data->cred = ds_cred;
1932 refcount_inc(&ds->ds_clp->cl_count);
1933 data->ds_clp = ds->ds_clp;
1934 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1935 if (fh)
1936 data->args.fh = fh;
1937
1938 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1939 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1940 &ff_layout_commit_call_ops_v4,
1941 how, RPC_TASK_SOFTCONN);
1942 put_cred(ds_cred);
1943 return ret;
1944out_err:
1945 pnfs_generic_prepare_to_resend_writes(data);
1946 pnfs_generic_commit_release(data);
1947 return -EAGAIN;
1948}
1949
1950static int
1951ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1952 int how, struct nfs_commit_info *cinfo)
1953{
1954 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1955 ff_layout_initiate_commit);
1956}
1957
1958static struct pnfs_ds_commit_info *
1959ff_layout_get_ds_info(struct inode *inode)
1960{
1961 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1962
1963 if (layout == NULL)
1964 return NULL;
1965
1966 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1967}
1968
1969static void
1970ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1971 struct pnfs_layout_segment *lseg)
1972{
1973 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1974 struct inode *inode = lseg->pls_layout->plh_inode;
1975 struct pnfs_commit_array *array, *new;
1976
1977 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1978 if (new) {
1979 spin_lock(&inode->i_lock);
1980 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1981 spin_unlock(&inode->i_lock);
1982 if (array != new)
1983 pnfs_free_commit_array(new);
1984 }
1985}
1986
1987static void
1988ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1989 struct inode *inode)
1990{
1991 spin_lock(&inode->i_lock);
1992 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1993 spin_unlock(&inode->i_lock);
1994}
1995
1996static void
1997ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1998{
1999 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2000 id_node));
2001}
2002
2003static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2004 const struct nfs4_layoutreturn_args *args,
2005 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2006{
2007 __be32 *start;
2008
2009 start = xdr_reserve_space(xdr, 4);
2010 if (unlikely(!start))
2011 return -E2BIG;
2012
2013 *start = cpu_to_be32(ff_args->num_errors);
2014
2015 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2016}
2017
2018static void
2019encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2020{
2021 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2022}
2023
2024static void
2025ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2026 const nfs4_stateid *stateid,
2027 const struct nfs42_layoutstat_devinfo *devinfo)
2028{
2029 __be32 *p;
2030
2031 p = xdr_reserve_space(xdr, 8 + 8);
2032 p = xdr_encode_hyper(p, devinfo->offset);
2033 p = xdr_encode_hyper(p, devinfo->length);
2034 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2035 p = xdr_reserve_space(xdr, 4*8);
2036 p = xdr_encode_hyper(p, devinfo->read_count);
2037 p = xdr_encode_hyper(p, devinfo->read_bytes);
2038 p = xdr_encode_hyper(p, devinfo->write_count);
2039 p = xdr_encode_hyper(p, devinfo->write_bytes);
2040 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2041}
2042
2043static void
2044ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2045 const nfs4_stateid *stateid,
2046 const struct nfs42_layoutstat_devinfo *devinfo)
2047{
2048 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2049 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2050 devinfo->ld_private.data);
2051}
2052
2053
2054static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2055 const struct nfs4_layoutreturn_args *args,
2056 struct nfs4_flexfile_layoutreturn_args *ff_args)
2057{
2058 __be32 *p;
2059 int i;
2060
2061 p = xdr_reserve_space(xdr, 4);
2062 *p = cpu_to_be32(ff_args->num_dev);
2063 for (i = 0; i < ff_args->num_dev; i++)
2064 ff_layout_encode_ff_iostat(xdr,
2065 &args->layout->plh_stateid,
2066 &ff_args->devinfo[i]);
2067}
2068
2069static void
2070ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2071 unsigned int num_entries)
2072{
2073 unsigned int i;
2074
2075 for (i = 0; i < num_entries; i++) {
2076 if (!devinfo[i].ld_private.ops)
2077 continue;
2078 if (!devinfo[i].ld_private.ops->free)
2079 continue;
2080 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2081 }
2082}
2083
2084static struct nfs4_deviceid_node *
2085ff_layout_alloc_deviceid_node(struct nfs_server *server,
2086 struct pnfs_device *pdev, gfp_t gfp_flags)
2087{
2088 struct nfs4_ff_layout_ds *dsaddr;
2089
2090 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2091 if (!dsaddr)
2092 return NULL;
2093 return &dsaddr->id_node;
2094}
2095
2096static void
2097ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2098 const void *voidargs,
2099 const struct nfs4_xdr_opaque_data *ff_opaque)
2100{
2101 const struct nfs4_layoutreturn_args *args = voidargs;
2102 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2103 struct xdr_buf tmp_buf = {
2104 .head = {
2105 [0] = {
2106 .iov_base = page_address(ff_args->pages[0]),
2107 },
2108 },
2109 .buflen = PAGE_SIZE,
2110 };
2111 struct xdr_stream tmp_xdr;
2112 __be32 *start;
2113
2114 dprintk("%s: Begin\n", __func__);
2115
2116 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2117
2118 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2119 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2120
2121 start = xdr_reserve_space(xdr, 4);
2122 *start = cpu_to_be32(tmp_buf.len);
2123 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2124
2125 dprintk("%s: Return\n", __func__);
2126}
2127
2128static void
2129ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2130{
2131 struct nfs4_flexfile_layoutreturn_args *ff_args;
2132
2133 if (!args->data)
2134 return;
2135 ff_args = args->data;
2136 args->data = NULL;
2137
2138 ff_layout_free_ds_ioerr(&ff_args->errors);
2139 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2140
2141 put_page(ff_args->pages[0]);
2142 kfree(ff_args);
2143}
2144
2145static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2146 .encode = ff_layout_encode_layoutreturn,
2147 .free = ff_layout_free_layoutreturn,
2148};
2149
2150static int
2151ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2152{
2153 struct nfs4_flexfile_layoutreturn_args *ff_args;
2154 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2155
2156 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2157 if (!ff_args)
2158 goto out_nomem;
2159 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2160 if (!ff_args->pages[0])
2161 goto out_nomem_free;
2162
2163 INIT_LIST_HEAD(&ff_args->errors);
2164 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2165 &args->range, &ff_args->errors,
2166 FF_LAYOUTRETURN_MAXERR);
2167
2168 spin_lock(&args->inode->i_lock);
2169 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2170 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2171 spin_unlock(&args->inode->i_lock);
2172
2173 args->ld_private->ops = &layoutreturn_ops;
2174 args->ld_private->data = ff_args;
2175 return 0;
2176out_nomem_free:
2177 kfree(ff_args);
2178out_nomem:
2179 return -ENOMEM;
2180}
2181
2182#ifdef CONFIG_NFS_V4_2
2183void
2184ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2185{
2186 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2187 struct nfs42_layout_error *errors;
2188 LIST_HEAD(head);
2189
2190 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2191 return;
2192 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2193 if (list_empty(&head))
2194 return;
2195
2196 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2197 sizeof(*errors), GFP_NOFS);
2198 if (errors != NULL) {
2199 const struct nfs4_ff_layout_ds_err *pos;
2200 size_t n = 0;
2201
2202 list_for_each_entry(pos, &head, list) {
2203 errors[n].offset = pos->offset;
2204 errors[n].length = pos->length;
2205 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2206 errors[n].errors[0].dev_id = pos->deviceid;
2207 errors[n].errors[0].status = pos->status;
2208 errors[n].errors[0].opnum = pos->opnum;
2209 n++;
2210 if (!list_is_last(&pos->list, &head) &&
2211 n < NFS42_LAYOUTERROR_MAX)
2212 continue;
2213 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2214 break;
2215 n = 0;
2216 }
2217 kfree(errors);
2218 }
2219 ff_layout_free_ds_ioerr(&head);
2220}
2221#else
2222void
2223ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2224{
2225}
2226#endif
2227
2228static int
2229ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2230{
2231 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2232
2233 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2234}
2235
2236static size_t
2237ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2238 const int buflen)
2239{
2240 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2241 const struct in6_addr *addr = &sin6->sin6_addr;
2242
2243
2244
2245
2246
2247
2248 if (ipv6_addr_any(addr))
2249 return snprintf(buf, buflen, "::");
2250
2251
2252
2253
2254
2255
2256 if (ipv6_addr_loopback(addr))
2257 return snprintf(buf, buflen, "::1");
2258
2259
2260
2261
2262
2263
2264
2265 if (ipv6_addr_v4mapped(addr))
2266 return snprintf(buf, buflen, "::ffff:%pI4",
2267 &addr->s6_addr32[3]);
2268
2269
2270
2271
2272 return snprintf(buf, buflen, "%pI6c", addr);
2273}
2274
2275
2276static void
2277ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2278{
2279 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2280 char portbuf[RPCBIND_MAXUADDRPLEN];
2281 char addrbuf[RPCBIND_MAXUADDRLEN];
2282 unsigned short port;
2283 int len, netid_len;
2284 __be32 *p;
2285
2286 switch (sap->sa_family) {
2287 case AF_INET:
2288 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2289 return;
2290 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2291 break;
2292 case AF_INET6:
2293 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2294 return;
2295 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2296 break;
2297 default:
2298 WARN_ON_ONCE(1);
2299 return;
2300 }
2301
2302 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2303 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2304
2305 netid_len = strlen(da->da_netid);
2306 p = xdr_reserve_space(xdr, 4 + netid_len);
2307 xdr_encode_opaque(p, da->da_netid, netid_len);
2308
2309 p = xdr_reserve_space(xdr, 4 + len);
2310 xdr_encode_opaque(p, addrbuf, len);
2311}
2312
2313static void
2314ff_layout_encode_nfstime(struct xdr_stream *xdr,
2315 ktime_t t)
2316{
2317 struct timespec64 ts;
2318 __be32 *p;
2319
2320 p = xdr_reserve_space(xdr, 12);
2321 ts = ktime_to_timespec64(t);
2322 p = xdr_encode_hyper(p, ts.tv_sec);
2323 *p++ = cpu_to_be32(ts.tv_nsec);
2324}
2325
2326static void
2327ff_layout_encode_io_latency(struct xdr_stream *xdr,
2328 struct nfs4_ff_io_stat *stat)
2329{
2330 __be32 *p;
2331
2332 p = xdr_reserve_space(xdr, 5 * 8);
2333 p = xdr_encode_hyper(p, stat->ops_requested);
2334 p = xdr_encode_hyper(p, stat->bytes_requested);
2335 p = xdr_encode_hyper(p, stat->ops_completed);
2336 p = xdr_encode_hyper(p, stat->bytes_completed);
2337 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2338 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2339 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2340}
2341
2342static void
2343ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2344 const struct nfs42_layoutstat_devinfo *devinfo,
2345 struct nfs4_ff_layout_mirror *mirror)
2346{
2347 struct nfs4_pnfs_ds_addr *da;
2348 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2349 struct nfs_fh *fh = &mirror->fh_versions[0];
2350 __be32 *p;
2351
2352 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2353 dprintk("%s: DS %s: encoding address %s\n",
2354 __func__, ds->ds_remotestr, da->da_remotestr);
2355
2356 ff_layout_encode_netaddr(xdr, da);
2357
2358 p = xdr_reserve_space(xdr, 4 + fh->size);
2359 xdr_encode_opaque(p, fh->data, fh->size);
2360
2361 spin_lock(&mirror->lock);
2362 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2363
2364 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2365 spin_unlock(&mirror->lock);
2366
2367 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2368
2369 p = xdr_reserve_space(xdr, 4);
2370 *p = cpu_to_be32(false);
2371}
2372
2373static void
2374ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2375 const struct nfs4_xdr_opaque_data *opaque)
2376{
2377 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2378 struct nfs42_layoutstat_devinfo, ld_private);
2379 __be32 *start;
2380
2381
2382 start = xdr_reserve_space(xdr, 4);
2383 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2384
2385 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2386}
2387
2388static void
2389ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2390{
2391 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2392
2393 ff_layout_put_mirror(mirror);
2394}
2395
2396static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2397 .encode = ff_layout_encode_layoutstats,
2398 .free = ff_layout_free_layoutstats,
2399};
2400
2401static int
2402ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2403 struct nfs42_layoutstat_devinfo *devinfo,
2404 int dev_limit)
2405{
2406 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2407 struct nfs4_ff_layout_mirror *mirror;
2408 struct nfs4_deviceid_node *dev;
2409 int i = 0;
2410
2411 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2412 if (i >= dev_limit)
2413 break;
2414 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2415 continue;
2416 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2417 continue;
2418
2419 if (!refcount_inc_not_zero(&mirror->ref))
2420 continue;
2421 dev = &mirror->mirror_ds->id_node;
2422 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2423 devinfo->offset = 0;
2424 devinfo->length = NFS4_MAX_UINT64;
2425 spin_lock(&mirror->lock);
2426 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2427 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2428 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2429 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2430 spin_unlock(&mirror->lock);
2431 devinfo->layout_type = LAYOUT_FLEX_FILES;
2432 devinfo->ld_private.ops = &layoutstat_ops;
2433 devinfo->ld_private.data = mirror;
2434
2435 devinfo++;
2436 i++;
2437 }
2438 return i;
2439}
2440
2441static int
2442ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2443{
2444 struct nfs4_flexfile_layout *ff_layout;
2445 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2446
2447
2448 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2449 if (!args->devinfo)
2450 return -ENOMEM;
2451
2452 spin_lock(&args->inode->i_lock);
2453 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2454 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2455 &args->devinfo[0], dev_count);
2456 spin_unlock(&args->inode->i_lock);
2457 if (!args->num_dev) {
2458 kfree(args->devinfo);
2459 args->devinfo = NULL;
2460 return -ENOENT;
2461 }
2462
2463 return 0;
2464}
2465
2466static int
2467ff_layout_set_layoutdriver(struct nfs_server *server,
2468 const struct nfs_fh *dummy)
2469{
2470#if IS_ENABLED(CONFIG_NFS_V4_2)
2471 server->caps |= NFS_CAP_LAYOUTSTATS;
2472#endif
2473 return 0;
2474}
2475
2476static const struct pnfs_commit_ops ff_layout_commit_ops = {
2477 .setup_ds_info = ff_layout_setup_ds_info,
2478 .release_ds_info = ff_layout_release_ds_info,
2479 .mark_request_commit = pnfs_layout_mark_request_commit,
2480 .clear_request_commit = pnfs_generic_clear_request_commit,
2481 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2482 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2483 .commit_pagelist = ff_layout_commit_pagelist,
2484};
2485
2486static struct pnfs_layoutdriver_type flexfilelayout_type = {
2487 .id = LAYOUT_FLEX_FILES,
2488 .name = "LAYOUT_FLEX_FILES",
2489 .owner = THIS_MODULE,
2490 .flags = PNFS_LAYOUTGET_ON_OPEN,
2491 .max_layoutget_response = 4096,
2492 .set_layoutdriver = ff_layout_set_layoutdriver,
2493 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2494 .free_layout_hdr = ff_layout_free_layout_hdr,
2495 .alloc_lseg = ff_layout_alloc_lseg,
2496 .free_lseg = ff_layout_free_lseg,
2497 .add_lseg = ff_layout_add_lseg,
2498 .pg_read_ops = &ff_layout_pg_read_ops,
2499 .pg_write_ops = &ff_layout_pg_write_ops,
2500 .get_ds_info = ff_layout_get_ds_info,
2501 .free_deviceid_node = ff_layout_free_deviceid_node,
2502 .read_pagelist = ff_layout_read_pagelist,
2503 .write_pagelist = ff_layout_write_pagelist,
2504 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2505 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2506 .sync = pnfs_nfs_generic_sync,
2507 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2508};
2509
2510static int __init nfs4flexfilelayout_init(void)
2511{
2512 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2513 __func__);
2514 return pnfs_register_layoutdriver(&flexfilelayout_type);
2515}
2516
2517static void __exit nfs4flexfilelayout_exit(void)
2518{
2519 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2520 __func__);
2521 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2522}
2523
2524MODULE_ALIAS("nfs-layouttype4-4");
2525
2526MODULE_LICENSE("GPL");
2527MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2528
2529module_init(nfs4flexfilelayout_init);
2530module_exit(nfs4flexfilelayout_exit);
2531
2532module_param(io_maxretrans, ushort, 0644);
2533MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2534 "retries an I/O request before returning an error. ");
2535
