]> git.ipfire.org Git - people/ms/linux.git/blame - fs/nfs/flexfilelayout/flexfilelayout.c
projects / people / ms / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFS: Fix fix of show_nfs_errors
[people/ms/linux.git] / fs / nfs / flexfilelayout / flexfilelayout.c
CommitLineData
09c434b8 1// SPDX-License-Identifier: GPL-2.0-only
d67ae825
TH		 2/*
3 * Module for pnfs flexfile layout driver.
4 *
5 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6 *
7 * Tao Peng <bergwolf@primarydata.com>
8 */
9
10#include <linux/nfs_fs.h>
7af46292 11#include <linux/nfs_mount.h>
d67ae825
TH		 12#include <linux/nfs_page.h>
13#include <linux/module.h>
97f68c6b 14#include <linux/sched/mm.h>
d67ae825
TH		 15
16#include <linux/sunrpc/metrics.h>
d67ae825
TH		 17
18#include "flexfilelayout.h"
19#include "../nfs4session.h"
40c64c26 20#include "../nfs4idmap.h"
d67ae825
TH		 21#include "../internal.h"
22#include "../delegation.h"
23#include "../nfs4trace.h"
24#include "../iostat.h"
25#include "../nfs.h"
ad4dc53e 26#include "../nfs42.h"
d67ae825
TH		 27
28#define NFSDBG_FACILITY NFSDBG_PNFS_LD
29
30#define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
5b9b3c85
TM		 31#define FF_LAYOUTRETURN_MAXERR 20
32
33344e0f
TM		 33static unsigned short io_maxretrans;
34
54e4a0df
TM		 35static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
36 struct nfs_pgio_header *hdr);
230bc962
TM		 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);
54e4a0df 43
d67ae825
TH		 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 INIT_LIST_HEAD(&ffl->error_list);
266d12d4 52 INIT_LIST_HEAD(&ffl->mirrors);
1c8d477a 53 ffl->last_report_time = ktime_get();
d67ae825
TH		 54 return &ffl->generic_hdr;
55 } else
56 return NULL;
57}
58
59static void
60ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
61{
62 struct nfs4_ff_layout_ds_err *err, *n;
63
64 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
65 list) {
66 list_del(&err->list);
67 kfree(err);
68 }
69 kfree(FF_LAYOUT_FROM_HDR(lo));
70}
71
93b717fd 72static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
d67ae825
TH		 73{
74 __be32 *p;
75
76 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
77 if (unlikely(p == NULL))
78 return -ENOBUFS;
93b717fd
TM		 79 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
80 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
d67ae825
TH		 81 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
82 p[0], p[1], p[2], p[3]);
83 return 0;
84}
85
86static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
87{
88 __be32 *p;
89
90 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
91 if (unlikely(!p))
92 return -ENOBUFS;
93 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
94 nfs4_print_deviceid(devid);
95 return 0;
96}
97
98static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
99{
100 __be32 *p;
101
102 p = xdr_inline_decode(xdr, 4);
103 if (unlikely(!p))
104 return -ENOBUFS;
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",
108 fh->size);
109 return -EOVERFLOW;
110 }
111 /* fh.data */
112 p = xdr_inline_decode(xdr, fh->size);
113 if (unlikely(!p))
114 return -ENOBUFS;
115 memcpy(&fh->data, p, fh->size);
116 dprintk("%s: fh len %d\n", __func__, fh->size);
117
118 return 0;
119}
120
121/*
122 * Currently only stringified uids and gids are accepted.
123 * I.e., kerberos is not supported to the DSes, so no pricipals.
124 *
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().
128 */
129static int
130decode_name(struct xdr_stream *xdr, u32 *id)
131{
132 __be32 *p;
133 int len;
134
135 /* opaque_length(4)*/
136 p = xdr_inline_decode(xdr, 4);
137 if (unlikely(!p))
138 return -ENOBUFS;
139 len = be32_to_cpup(p++);
140 if (len < 0)
141 return -EINVAL;
142
143 dprintk("%s: len %u\n", __func__, len);
144
145 /* opaque body */
146 p = xdr_inline_decode(xdr, len);
147 if (unlikely(!p))
148 return -ENOBUFS;
149
150 if (!nfs_map_string_to_numeric((char *)p, len, id))
151 return -EINVAL;
152
153 return 0;
154}
155
266d12d4
TM		 156static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
157 const struct nfs4_ff_layout_mirror *m2)
158{
159 int i, j;
160
161 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
162 return false;
163 for (i = 0; i < m1->fh_versions_cnt; i++) {
164 bool found_fh = false;
86fb449b 165 for (j = 0; j < m2->fh_versions_cnt; j++) {
266d12d4
TM		 166 if (nfs_compare_fh(&m1->fh_versions[i],
167 &m2->fh_versions[j]) == 0) {
168 found_fh = true;
169 break;
170 }
171 }
172 if (!found_fh)
173 return false;
174 }
175 return true;
176}
177
178static struct nfs4_ff_layout_mirror *
179ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
180 struct nfs4_ff_layout_mirror *mirror)
181{
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;
185
186 spin_lock(&inode->i_lock);
187 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
65990d1a 188 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
266d12d4
TM		 189 continue;
190 if (!ff_mirror_match_fh(mirror, pos))
191 continue;
81a090b9 192 if (refcount_inc_not_zero(&pos->ref)) {
266d12d4
TM		 193 spin_unlock(&inode->i_lock);
194 return pos;
195 }
196 }
197 list_add(&mirror->mirrors, &ff_layout->mirrors);
198 mirror->layout = lo;
199 spin_unlock(&inode->i_lock);
200 return mirror;
201}
202
e3b1df2d 203static void
266d12d4
TM		 204ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
205{
206 struct inode *inode;
207 if (mirror->layout == NULL)
208 return;
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;
214}
215
28a0d72c
TM		 216static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
217{
218 struct nfs4_ff_layout_mirror *mirror;
219
220 mirror = kzalloc(sizeof(*mirror), gfp_flags);
221 if (mirror != NULL) {
222 spin_lock_init(&mirror->lock);
81a090b9 223 refcount_set(&mirror->ref, 1);
266d12d4 224 INIT_LIST_HEAD(&mirror->mirrors);
28a0d72c
TM		 225 }
226 return mirror;
227}
228
229static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
230{
a52458b4 231 const struct cred *cred;
3064b686 232
266d12d4 233 ff_layout_remove_mirror(mirror);
28a0d72c 234 kfree(mirror->fh_versions);
3064b686 235 cred = rcu_access_pointer(mirror->ro_cred);
a52458b4 236 put_cred(cred);
3064b686 237 cred = rcu_access_pointer(mirror->rw_cred);
a52458b4 238 put_cred(cred);
28a0d72c
TM		 239 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
240 kfree(mirror);
241}
242
243static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
244{
81a090b9 245 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
28a0d72c
TM		 246 ff_layout_free_mirror(mirror);
247}
248
d67ae825
TH		 249static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
250{
251 int i;
252
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 */
28a0d72c 258 ff_layout_put_mirror(fls->mirror_array[i]);
d67ae825
TH		 259 }
260 kfree(fls->mirror_array);
261 fls->mirror_array = NULL;
262 }
263}
264
265static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
266{
267 int ret = 0;
268
269 dprintk("--> %s\n", __func__);
270
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",
275 __func__);
276 ret = -EINVAL;
277 }
278
279 dprintk("--> %s returns %d\n", __func__, ret);
280 return ret;
281}
282
283static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
284{
285 if (fls) {
286 ff_layout_free_mirror_array(fls);
287 kfree(fls);
288 }
289}
290
0762ed2c
TM		 291static bool
292ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
293 const struct pnfs_layout_range *l2)
294{
295 u64 end1, end2;
296
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)
302 return false;
303 if (end2 < l1->offset)
304 return true;
305 return l2->offset <= l1->offset;
306}
307
308static bool
309ff_lseg_merge(struct pnfs_layout_segment *new,
310 struct pnfs_layout_segment *old)
311{
312 u64 new_end, old_end;
313
ee26bdd6
JL		 314 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
315 return false;
0762ed2c
TM		 316 if (new->pls_range.iomode != old->pls_range.iomode)
317 return false;
318 old_end = pnfs_calc_offset_end(old->pls_range.offset,
319 old->pls_range.length);
320 if (old_end < new->pls_range.offset)
321 return false;
322 new_end = pnfs_calc_offset_end(new->pls_range.offset,
323 new->pls_range.length);
324 if (new_end < old->pls_range.offset)
325 return false;
326
327 /* Mergeable: copy info from 'old' to 'new' */
328 if (new_end < old_end)
329 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,
333 new_end);
334 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
335 set_bit(NFS_LSEG_ROC, &new->pls_flags);
0762ed2c
TM		 336 return true;
337}
338
339static void
340ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
341 struct pnfs_layout_segment *lseg,
342 struct list_head *free_me)
343{
344 pnfs_generic_layout_insert_lseg(lo, lseg,
345 ff_lseg_range_is_after,
346 ff_lseg_merge,
347 free_me);
348}
349
d67ae825
TH		 350static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
351{
d67ae825
TH		 352 int i, j;
353
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 <
455b6ee6
FF		 357 fls->mirror_array[j]->efficiency)
358 swap(fls->mirror_array[i],
359 fls->mirror_array[j]);
d67ae825
TH		 360 }
361}
362
363static struct pnfs_layout_segment *
364ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
365 struct nfs4_layoutget_res *lgr,
366 gfp_t gfp_flags)
367{
368 struct pnfs_layout_segment *ret;
369 struct nfs4_ff_layout_segment *fls = NULL;
370 struct xdr_stream stream;
371 struct xdr_buf buf;
372 struct page *scratch;
373 u64 stripe_unit;
374 u32 mirror_array_cnt;
375 __be32 *p;
376 int i, rc;
377
378 dprintk("--> %s\n", __func__);
379 scratch = alloc_page(gfp_flags);
380 if (!scratch)
381 return ERR_PTR(-ENOMEM);
382
383 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
384 lgr->layoutp->len);
385 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
386
387 /* stripe unit and mirror_array_cnt */
388 rc = -EIO;
389 p = xdr_inline_decode(&stream, 8 + 4);
390 if (!p)
391 goto out_err_free;
392
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);
397
398 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
399 mirror_array_cnt == 0)
400 goto out_err_free;
401
402 rc = -ENOMEM;
403 fls = kzalloc(sizeof(*fls), gfp_flags);
404 if (!fls)
405 goto out_err_free;
406
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)
412 goto out_err_free;
413
414 for (i = 0; i < fls->mirror_array_cnt; i++) {
266d12d4 415 struct nfs4_ff_layout_mirror *mirror;
97f68c6b 416 struct cred *kcred;
762bb7e9 417 const struct cred __rcu *cred;
8276c902
N		 418 kuid_t uid;
419 kgid_t gid;
90a0be00 420 u32 ds_count, fh_count, id;
d67ae825
TH		 421 int j;
422
423 rc = -EIO;
424 p = xdr_inline_decode(&stream, 4);
425 if (!p)
426 goto out_err_free;
427 ds_count = be32_to_cpup(p);
428
429 /* FIXME: allow for striping? */
430 if (ds_count != 1)
431 goto out_err_free;
432
28a0d72c 433 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
d67ae825
TH		 434 if (fls->mirror_array[i] == NULL) {
435 rc = -ENOMEM;
436 goto out_err_free;
437 }
438
d67ae825
TH		 439 fls->mirror_array[i]->ds_count = ds_count;
440
441 /* deviceid */
65990d1a 442 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
d67ae825
TH		 443 if (rc)
444 goto out_err_free;
445
d67ae825
TH		 446 /* efficiency */
447 rc = -EIO;
448 p = xdr_inline_decode(&stream, 4);
449 if (!p)
450 goto out_err_free;
451 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
452
453 /* stateid */
93b717fd 454 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
d67ae825
TH		 455 if (rc)
456 goto out_err_free;
457
458 /* fh */
662f9a10 459 rc = -EIO;
d67ae825
TH		 460 p = xdr_inline_decode(&stream, 4);
461 if (!p)
462 goto out_err_free;
463 fh_count = be32_to_cpup(p);
464
465 fls->mirror_array[i]->fh_versions =
6396bb22 466 kcalloc(fh_count, sizeof(struct nfs_fh),
d67ae825
TH		 467 gfp_flags);
468 if (fls->mirror_array[i]->fh_versions == NULL) {
469 rc = -ENOMEM;
470 goto out_err_free;
471 }
472
473 for (j = 0; j < fh_count; j++) {
474 rc = decode_nfs_fh(&stream,
475 &fls->mirror_array[i]->fh_versions[j]);
476 if (rc)
477 goto out_err_free;
478 }
479
480 fls->mirror_array[i]->fh_versions_cnt = fh_count;
481
482 /* user */
90a0be00 483 rc = decode_name(&stream, &id);
d67ae825
TH		 484 if (rc)
485 goto out_err_free;
486
8276c902 487 uid = make_kuid(&init_user_ns, id);
90a0be00 488
d67ae825 489 /* group */
90a0be00 490 rc = decode_name(&stream, &id);
d67ae825
TH		 491 if (rc)
492 goto out_err_free;
493
8276c902 494 gid = make_kgid(&init_user_ns, id);
90a0be00 495
97f68c6b
N		 496 if (gfp_flags & __GFP_FS)
497 kcred = prepare_kernel_cred(NULL);
498 else {
499 unsigned int nofs_flags = memalloc_nofs_save();
500 kcred = prepare_kernel_cred(NULL);
501 memalloc_nofs_restore(nofs_flags);
502 }
503 rc = -ENOMEM;
504 if (!kcred)
505 goto out_err_free;
8276c902
N		 506 kcred->fsuid = uid;
507 kcred->fsgid = gid;
762bb7e9 508 cred = RCU_INITIALIZER(kcred);
90a0be00 509
3064b686
JL		 510 if (lgr->range.iomode == IOMODE_READ)
511 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
512 else
513 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
90a0be00 514
266d12d4
TM		 515 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
516 if (mirror != fls->mirror_array[i]) {
90a0be00 517 /* swap cred ptrs so free_mirror will clean up old */
3064b686
JL		 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);
521 } else {
522 cred = xchg(&mirror->rw_cred, cred);
523 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
524 }
266d12d4
TM		 525 ff_layout_free_mirror(fls->mirror_array[i]);
526 fls->mirror_array[i] = mirror;
527 }
528
3064b686
JL		 529 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
530 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
8276c902
N		 531 from_kuid(&init_user_ns, uid),
532 from_kgid(&init_user_ns, gid));
d67ae825
TH		 533 }
534
c0f5f505 535 p = xdr_inline_decode(&stream, 4);
d0379a5d
TM		 536 if (!p)
537 goto out_sort_mirrors;
538 fls->flags = be32_to_cpup(p);
539
540 p = xdr_inline_decode(&stream, 4);
541 if (!p)
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);
c0f5f505 545
d0379a5d 546out_sort_mirrors:
d67ae825
TH		 547 ff_layout_sort_mirrors(fls);
548 rc = ff_layout_check_layout(lgr);
549 if (rc)
550 goto out_err_free;
d67ae825
TH		 551 ret = &fls->generic_hdr;
552 dprintk("<-- %s (success)\n", __func__);
553out_free_page:
554 __free_page(scratch);
555 return ret;
556out_err_free:
557 _ff_layout_free_lseg(fls);
558 ret = ERR_PTR(rc);
559 dprintk("<-- %s (%d)\n", __func__, rc);
560 goto out_free_page;
561}
562
563static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
564{
565 struct pnfs_layout_segment *lseg;
566
567 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
568 if (lseg->pls_range.iomode == IOMODE_RW)
569 return true;
570
571 return false;
572}
573
574static void
575ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
576{
577 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825
TH		 578
579 dprintk("--> %s\n", __func__);
580
d67ae825
TH		 581 if (lseg->pls_range.iomode == IOMODE_RW) {
582 struct nfs4_flexfile_layout *ffl;
583 struct inode *inode;
584
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;
592 }
593 spin_unlock(&inode->i_lock);
594 }
595 _ff_layout_free_lseg(fls);
596}
597
598/* Return 1 until we have multiple lsegs support */
599static int
600ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
601{
602 return 1;
603}
604
abcb7bfc 605static void
e76d28dd 606nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 607{
9bbd9bb4
PT		 608 /* first IO request? */
609 if (atomic_inc_return(&timer->n_ops) == 1) {
e76d28dd 610 timer->start_time = now;
abcb7bfc
TM		 611 }
612}
613
614static ktime_t
e76d28dd 615nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 616{
e76d28dd 617 ktime_t start;
abcb7bfc 618
9bbd9bb4
PT		 619 if (atomic_dec_return(&timer->n_ops) < 0)
620 WARN_ON_ONCE(1);
621
9bbd9bb4
PT		 622 start = timer->start_time;
623 timer->start_time = now;
abcb7bfc
TM		 624 return ktime_sub(now, start);
625}
626
97ba375b 627static bool
d983803d 628nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
e76d28dd
TM		 629 struct nfs4_ff_layoutstat *layoutstat,
630 ktime_t now)
abcb7bfc 631{
bbf58bf3 632 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
1c8d477a 633 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
d983803d 634
e76d28dd 635 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
1f3a8e49 636 if (!mirror->start_time)
9bbd9bb4 637 mirror->start_time = now;
d0379a5d
TM		 638 if (mirror->report_interval != 0)
639 report_interval = (s64)mirror->report_interval * 1000LL;
640 else if (layoutstats_timer != 0)
bbf58bf3 641 report_interval = (s64)layoutstats_timer * 1000LL;
1c8d477a 642 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
bbf58bf3 643 report_interval) {
1c8d477a 644 ffl->last_report_time = now;
97ba375b
PT		 645 return true;
646 }
647
648 return false;
abcb7bfc
TM		 649}
650
651static void
652nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
653 __u64 requested)
654{
655 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
656
657 iostat->ops_requested++;
658 iostat->bytes_requested += requested;
659}
660
661static void
662nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
663 __u64 requested,
664 __u64 completed,
e76d28dd
TM		 665 ktime_t time_completed,
666 ktime_t time_started)
abcb7bfc
TM		 667{
668 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
e76d28dd 669 ktime_t completion_time = ktime_sub(time_completed, time_started);
abcb7bfc
TM		 670 ktime_t timer;
671
672 iostat->ops_completed++;
673 iostat->bytes_completed += completed;
674 iostat->bytes_not_delivered += requested - completed;
675
e76d28dd 676 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
abcb7bfc
TM		 677 iostat->total_busy_time =
678 ktime_add(iostat->total_busy_time, timer);
679 iostat->aggregate_completion_time =
e76d28dd
TM		 680 ktime_add(iostat->aggregate_completion_time,
681 completion_time);
abcb7bfc
TM		 682}
683
684static void
0b7baf94
TM		 685nfs4_ff_layout_stat_io_start_read(struct inode *inode,
686 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 687 __u64 requested, ktime_t now)
abcb7bfc 688{
97ba375b
PT		 689 bool report;
690
abcb7bfc 691 spin_lock(&mirror->lock);
e76d28dd 692 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
abcb7bfc 693 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
06946c6a 694 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 695 spin_unlock(&mirror->lock);
97ba375b
PT		 696
697 if (report)
0b7baf94 698 pnfs_report_layoutstat(inode, GFP_KERNEL);
abcb7bfc
TM		 699}
700
701static void
702nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
703 struct nfs4_ff_layout_mirror *mirror,
704 __u64 requested,
705 __u64 completed)
706{
707 spin_lock(&mirror->lock);
708 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
709 requested, completed,
e76d28dd 710 ktime_get(), task->tk_start);
06946c6a 711 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 712 spin_unlock(&mirror->lock);
713}
714
715static void
0b7baf94
TM		 716nfs4_ff_layout_stat_io_start_write(struct inode *inode,
717 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 718 __u64 requested, ktime_t now)
abcb7bfc 719{
97ba375b
PT		 720 bool report;
721
abcb7bfc 722 spin_lock(&mirror->lock);
e76d28dd 723 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
abcb7bfc 724 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
06946c6a 725 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 726 spin_unlock(&mirror->lock);
97ba375b
PT		 727
728 if (report)
0b7baf94 729 pnfs_report_layoutstat(inode, GFP_NOIO);
abcb7bfc
TM		 730}
731
732static void
733nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
734 struct nfs4_ff_layout_mirror *mirror,
735 __u64 requested,
736 __u64 completed,
737 enum nfs3_stable_how committed)
738{
739 if (committed == NFS_UNSTABLE)
740 requested = completed = 0;
741
742 spin_lock(&mirror->lock);
743 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
e76d28dd 744 requested, completed, ktime_get(), task->tk_start);
06946c6a 745 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 746 spin_unlock(&mirror->lock);
747}
748
d67ae825
TH		 749static int
750ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
751 struct nfs_commit_info *cinfo,
752 gfp_t gfp_flags)
753{
754 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
755 struct pnfs_commit_bucket *buckets;
756 int size;
757
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.
763 */
764 return 0;
765 }
766
767 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
768
769 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
770 gfp_flags);
771 if (!buckets)
772 return -ENOMEM;
773 else {
774 int i;
775
fe238e60 776 spin_lock(&cinfo->inode->i_lock);
d67ae825
TH		 777 if (cinfo->ds->nbuckets != 0)
778 kfree(buckets);
779 else {
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;
788 }
789 }
fe238e60 790 spin_unlock(&cinfo->inode->i_lock);
d67ae825
TH		 791 return 0;
792 }
793}
794
76c66905
TM		 795static void
796ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
797{
798 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
799
800 if (devid)
801 nfs4_mark_deviceid_unavailable(devid);
802}
803
804static void
805ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
806{
807 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
808
809 if (devid)
810 nfs4_mark_deviceid_available(devid);
811}
812
d67ae825 813static struct nfs4_pnfs_ds *
76c66905
TM		 814ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
815 int start_idx, int *best_idx,
816 bool check_device)
d67ae825 817{
13544412 818 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
2444ff27 819 struct nfs4_ff_layout_mirror *mirror;
d67ae825 820 struct nfs4_pnfs_ds *ds;
3dc14735 821 bool fail_return = false;
d67ae825
TH		 822 int idx;
823
76c66905 824 /* mirrors are initially sorted by efficiency */
13544412 825 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
3dc14735 826 if (idx+1 == fls->mirror_array_cnt)
76c66905 827 fail_return = !check_device;
2444ff27
TM		 828
829 mirror = FF_LAYOUT_COMP(lseg, idx);
76c66905
TM		 830 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
831 if (!ds)
2444ff27
TM		 832 continue;
833
76c66905
TM		 834 if (check_device &&
835 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
836 continue;
837
838 *best_idx = idx;
839 return ds;
d67ae825
TH		 840 }
841
842 return NULL;
843}
844
76c66905
TM		 845static struct nfs4_pnfs_ds *
846ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
847 int start_idx, int *best_idx)
848{
849 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
850}
851
852static struct nfs4_pnfs_ds *
853ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
854 int start_idx, int *best_idx)
855{
856 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
857}
858
859static struct nfs4_pnfs_ds *
860ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
861 int start_idx, int *best_idx)
862{
863 struct nfs4_pnfs_ds *ds;
864
865 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
866 if (ds)
867 return ds;
868 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
869}
870
c7d73af2
TH		 871static void
872ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
873 struct nfs_page *req,
874 bool strict_iomode)
875{
c7d73af2
TH		 876 pnfs_put_lseg(pgio->pg_lseg);
877 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 878 nfs_req_openctx(req),
c7d73af2
TH		 879 0,
880 NFS4_MAX_UINT64,
881 IOMODE_READ,
882 strict_iomode,
883 GFP_KERNEL);
884 if (IS_ERR(pgio->pg_lseg)) {
885 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
886 pgio->pg_lseg = NULL;
887 }
c7d73af2
TH		 888}
889
d67ae825
TH		 890static void
891ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
892 struct nfs_page *req)
893{
894 struct nfs_pgio_mirror *pgm;
895 struct nfs4_ff_layout_mirror *mirror;
896 struct nfs4_pnfs_ds *ds;
897 int ds_idx;
898
3dc14735 899retry:
b3230e80 900 pnfs_generic_pg_check_layout(pgio);
d67ae825 901 /* Use full layout for now */
1c1aeaf1 902 if (!pgio->pg_lseg) {
c7d73af2 903 ff_layout_pg_get_read(pgio, req, false);
1c1aeaf1
TM		 904 if (!pgio->pg_lseg)
905 goto out_nolseg;
906 }
907 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
c7d73af2 908 ff_layout_pg_get_read(pgio, req, true);
1c1aeaf1
TM		 909 if (!pgio->pg_lseg)
910 goto out_nolseg;
911 }
d67ae825 912
13544412 913 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
446ca219 914 if (!ds) {
3dc14735 915 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 916 goto out_mds;
3dc14735
TM		 917 pnfs_put_lseg(pgio->pg_lseg);
918 pgio->pg_lseg = NULL;
919 /* Sleep for 1 second before retrying */
920 ssleep(1);
921 goto retry;
446ca219
TH		 922 }
923
d67ae825
TH		 924 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
925
926 pgio->pg_mirror_idx = ds_idx;
927
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;
931
7af46292
TM		 932 if (NFS_SERVER(pgio->pg_inode)->flags &
933 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
934 pgio->pg_maxretrans = io_maxretrans;
d67ae825 935 return;
1c1aeaf1
TM		 936out_nolseg:
937 if (pgio->pg_error < 0)
938 return;
d67ae825 939out_mds:
d5b9216f
TM		 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,
943 pgio->pg_lseg);
d67ae825
TH		 944 pnfs_put_lseg(pgio->pg_lseg);
945 pgio->pg_lseg = NULL;
7af46292 946 pgio->pg_maxretrans = 0;
d67ae825
TH		 947 nfs_pageio_reset_read_mds(pgio);
948}
949
950static void
951ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
952 struct nfs_page *req)
953{
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;
958 int i;
959 int status;
960
3dc14735 961retry:
b3230e80 962 pnfs_generic_pg_check_layout(pgio);
d600ad1f 963 if (!pgio->pg_lseg) {
d67ae825 964 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 965 nfs_req_openctx(req),
d67ae825
TH		 966 0,
967 NFS4_MAX_UINT64,
968 IOMODE_RW,
c7d73af2 969 false,
d67ae825 970 GFP_NOFS);
d600ad1f
PT		 971 if (IS_ERR(pgio->pg_lseg)) {
972 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
973 pgio->pg_lseg = NULL;
974 return;
975 }
976 }
d67ae825
TH		 977 /* If no lseg, fall back to write through mds */
978 if (pgio->pg_lseg == NULL)
979 goto out_mds;
980
981 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
982 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
983 if (status < 0)
984 goto out_mds;
985
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)))
989 goto out_mds;
990
991 for (i = 0; i < pgio->pg_mirror_count; i++) {
2444ff27 992 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
2444ff27 993 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
446ca219 994 if (!ds) {
3dc14735 995 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 996 goto out_mds;
3dc14735
TM		 997 pnfs_put_lseg(pgio->pg_lseg);
998 pgio->pg_lseg = NULL;
999 /* Sleep for 1 second before retrying */
1000 ssleep(1);
1001 goto retry;
446ca219 1002 }
d67ae825 1003 pgm = &pgio->pg_mirrors[i];
d67ae825
TH		 1004 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
1005 }
1006
7af46292
TM		 1007 if (NFS_SERVER(pgio->pg_inode)->flags &
1008 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
1009 pgio->pg_maxretrans = io_maxretrans;
d67ae825
TH		 1010 return;
1011
1012out_mds:
d5b9216f
TM		 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,
1016 pgio->pg_lseg);
d67ae825
TH		 1017 pnfs_put_lseg(pgio->pg_lseg);
1018 pgio->pg_lseg = NULL;
7af46292 1019 pgio->pg_maxretrans = 0;
d67ae825
TH		 1020 nfs_pageio_reset_write_mds(pgio);
1021}
1022
1023static unsigned int
1024ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1025 struct nfs_page *req)
1026{
d600ad1f 1027 if (!pgio->pg_lseg) {
d67ae825 1028 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 1029 nfs_req_openctx(req),
d67ae825
TH		 1030 0,
1031 NFS4_MAX_UINT64,
1032 IOMODE_RW,
c7d73af2 1033 false,
d67ae825 1034 GFP_NOFS);
d600ad1f
PT		 1035 if (IS_ERR(pgio->pg_lseg)) {
1036 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1037 pgio->pg_lseg = NULL;
1038 goto out;
1039 }
1040 }
d67ae825
TH		 1041 if (pgio->pg_lseg)
1042 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1043
d5b9216f
TM		 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,
1047 pgio->pg_lseg);
d67ae825 1048 /* no lseg means that pnfs is not in use, so no mirroring here */
d67ae825 1049 nfs_pageio_reset_write_mds(pgio);
d600ad1f 1050out:
d67ae825
TH		 1051 return 1;
1052}
1053
1054static 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,
1059};
1060
1061static 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,
1067};
1068
1069static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1070{
1071 struct rpc_task *task = &hdr->task;
1072
1073 pnfs_layoutcommit_inode(hdr->inode, false);
1074
1075 if (retry_pnfs) {
1076 dprintk("%s Reset task %5u for i/o through pNFS "
1077 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1078 hdr->task.tk_pid,
1079 hdr->inode->i_sb->s_id,
1080 (unsigned long long)NFS_FILEID(hdr->inode),
1081 hdr->args.count,
1082 (unsigned long long)hdr->args.offset);
1083
dc602dd7 1084 hdr->completion_ops->reschedule_io(hdr);
d67ae825
TH		 1085 return;
1086 }
1087
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__,
1091 hdr->task.tk_pid,
1092 hdr->inode->i_sb->s_id,
1093 (unsigned long long)NFS_FILEID(hdr->inode),
1094 hdr->args.count,
1095 (unsigned long long)hdr->args.offset);
1096
d5b9216f
TM		 1097 trace_pnfs_mds_fallback_write_done(hdr->inode,
1098 hdr->args.offset, hdr->args.count,
1099 IOMODE_RW, NFS_I(hdr->inode)->layout,
1100 hdr->lseg);
d67ae825
TH		 1101 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1102 }
1103}
1104
1105static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1106{
1107 struct rpc_task *task = &hdr->task;
1108
1109 pnfs_layoutcommit_inode(hdr->inode, false);
1110
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__,
1114 hdr->task.tk_pid,
1115 hdr->inode->i_sb->s_id,
1116 (unsigned long long)NFS_FILEID(hdr->inode),
1117 hdr->args.count,
1118 (unsigned long long)hdr->args.offset);
1119
d5b9216f
TM		 1120 trace_pnfs_mds_fallback_read_done(hdr->inode,
1121 hdr->args.offset, hdr->args.count,
1122 IOMODE_READ, NFS_I(hdr->inode)->layout,
1123 hdr->lseg);
d67ae825
TH		 1124 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1125 }
1126}
1127
1128static 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,
1132 int idx)
1133{
1134 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1135 struct inode *inode = lo->plh_inode;
d67ae825 1136 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
d67ae825
TH		 1137 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1138
d67ae825 1139 switch (task->tk_status) {
d67ae825
TH		 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);
1151 break;
1152 case -NFS4ERR_DELAY:
1153 case -NFS4ERR_GRACE:
1154 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1155 break;
1156 case -NFS4ERR_RETRY_UNCACHED_REP:
1157 break;
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__,
1166 task->tk_status);
1167 /*
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.
1173 */
1174 pnfs_destroy_layout(NFS_I(inode));
1175 rpc_wake_up(&tbl->slot_tbl_waitq);
1176 goto reset;
1177 /* RPC connection errors */
1178 case -ECONNREFUSED:
1179 case -EHOSTDOWN:
1180 case -EHOSTUNREACH:
1181 case -ENETUNREACH:
1182 case -EIO:
1183 case -ETIMEDOUT:
1184 case -EPIPE:
1185 dprintk("%s DS connection error %d\n", __func__,
1186 task->tk_status);
1c48cee8
WAA		 1187 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1188 &devid->deviceid);
d67ae825
TH		 1189 rpc_wake_up(&tbl->slot_tbl_waitq);
1190 /* fall through */
1191 default:
3b13b4b3 1192 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1193 return -NFS4ERR_RESET_TO_PNFS;
1194reset:
1195 dprintk("%s Retry through MDS. Error %d\n", __func__,
1196 task->tk_status);
1197 return -NFS4ERR_RESET_TO_MDS;
1198 }
d67ae825
TH		 1199 task->tk_status = 0;
1200 return -EAGAIN;
d67ae825
TH		 1201}
1202
1203/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1204static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1205 struct pnfs_layout_segment *lseg,
1206 int idx)
1207{
1208 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1209
90816d1d
TM		 1210 switch (task->tk_status) {
1211 /* File access problems. Don't mark the device as unavailable */
1212 case -EACCES:
1213 case -ESTALE:
1214 case -EISDIR:
1215 case -EBADHANDLE:
1216 case -ELOOP:
1217 case -ENOSPC:
1218 break;
1219 case -EJUKEBOX:
1220 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1221 goto out_retry;
1222 default:
d67ae825
TH		 1223 dprintk("%s DS connection error %d\n", __func__,
1224 task->tk_status);
1c48cee8
WAA		 1225 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1226 &devid->deviceid);
d67ae825 1227 }
90816d1d
TM		 1228 /* FIXME: Need to prevent infinite looping here. */
1229 return -NFS4ERR_RESET_TO_PNFS;
1230out_retry:
d67ae825 1231 task->tk_status = 0;
7eeea167 1232 rpc_restart_call_prepare(task);
d67ae825
TH		 1233 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1234 return -EAGAIN;
1235}
1236
1237static 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,
1241 int idx)
1242{
1243 int vers = clp->cl_nfs_mod->rpc_vers->number;
1244
76c66905
TM		 1245 if (task->tk_status >= 0) {
1246 ff_layout_mark_ds_reachable(lseg, idx);
df3ab232 1247 return 0;
76c66905 1248 }
df3ab232
TM		 1249
1250 /* Handle the case of an invalid layout segment */
1251 if (!pnfs_is_valid_lseg(lseg))
1252 return -NFS4ERR_RESET_TO_PNFS;
1253
d67ae825
TH		 1254 switch (vers) {
1255 case 3:
1256 return ff_layout_async_handle_error_v3(task, lseg, idx);
1257 case 4:
1258 return ff_layout_async_handle_error_v4(task, state, clp,
1259 lseg, idx);
1260 default:
1261 /* should never happen */
1262 WARN_ON_ONCE(1);
1263 return 0;
1264 }
1265}
1266
1267static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1268 int idx, u64 offset, u64 length,
dd52128a 1269 u32 status, int opnum, int error)
d67ae825
TH		 1270{
1271 struct nfs4_ff_layout_mirror *mirror;
1272 int err;
1273
dd52128a
TM		 1274 if (status == 0) {
1275 switch (error) {
1276 case -ETIMEDOUT:
1277 case -EPFNOSUPPORT:
1278 case -EPROTONOSUPPORT:
1279 case -EOPNOTSUPP:
1280 case -ECONNREFUSED:
1281 case -ECONNRESET:
1282 case -EHOSTDOWN:
1283 case -EHOSTUNREACH:
1284 case -ENETUNREACH:
1285 case -EADDRINUSE:
1286 case -ENOBUFS:
1287 case -EPIPE:
1288 case -EPERM:
1289 status = NFS4ERR_NXIO;
1290 break;
1291 case -EACCES:
1292 status = NFS4ERR_ACCESS;
1293 break;
1294 default:
1295 return;
1296 }
1297 }
1298
c22eeb86
PT		 1299 switch (status) {
1300 case NFS4ERR_DELAY:
1301 case NFS4ERR_GRACE:
1302 return;
1303 default:
1304 break;
1305 }
1306
d67ae825
TH		 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,
1310 GFP_NOIO);
76c66905
TM		 1311 if (status == NFS4ERR_NXIO)
1312 ff_layout_mark_ds_unreachable(lseg, idx);
f95c03b2 1313 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
d67ae825
TH		 1314 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1315}
1316
1317/* NFS_PROTO call done callback routines */
d67ae825
TH		 1318static int ff_layout_read_done_cb(struct rpc_task *task,
1319 struct nfs_pgio_header *hdr)
1320{
166bd5b8 1321 int new_idx = hdr->pgio_mirror_idx;
d67ae825
TH		 1322 int err;
1323
1324 trace_nfs4_pnfs_read(hdr, task->tk_status);
dd52128a 1325 if (task->tk_status < 0)
d67ae825
TH		 1326 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1327 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1328 hdr->res.op_status, OP_READ,
1329 task->tk_status);
d67ae825
TH		 1330 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1331 hdr->ds_clp, hdr->lseg,
1332 hdr->pgio_mirror_idx);
1333
42f86b44
TM		 1334 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1335 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1336 switch (err) {
1337 case -NFS4ERR_RESET_TO_PNFS:
13544412
TM		 1338 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1339 hdr->pgio_mirror_idx + 1,
166bd5b8 1340 &new_idx))
f0922a6c 1341 goto out_layouterror;
42f86b44 1342 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
d67ae825
TH		 1343 return task->tk_status;
1344 case -NFS4ERR_RESET_TO_MDS:
42f86b44 1345 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1346 return task->tk_status;
1347 case -EAGAIN:
13544412 1348 goto out_eagain;
d67ae825
TH		 1349 }
1350
1351 return 0;
f0922a6c 1352out_layouterror:
166bd5b8 1353 ff_layout_read_record_layoutstats_done(task, hdr);
f0922a6c 1354 ff_layout_send_layouterror(hdr->lseg);
166bd5b8 1355 hdr->pgio_mirror_idx = new_idx;
13544412
TM		 1356out_eagain:
1357 rpc_restart_call_prepare(task);
1358 return -EAGAIN;
d67ae825
TH		 1359}
1360
c0f5f505
TM		 1361static bool
1362ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1363{
1364 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1365}
1366
d67ae825
TH		 1367/*
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.
1371 *
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.
1375 */
1376static void
bc28e1c2
TM		 1377ff_layout_set_layoutcommit(struct inode *inode,
1378 struct pnfs_layout_segment *lseg,
1379 loff_t end_offset)
d67ae825 1380{
bc28e1c2 1381 if (!ff_layout_need_layoutcommit(lseg))
c0f5f505
TM		 1382 return;
1383
bc28e1c2
TM		 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);
d67ae825
TH		 1387}
1388
37e9ed22
TM		 1389static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1390 struct nfs_pgio_header *hdr)
d67ae825 1391{
37e9ed22
TM		 1392 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1393 return;
0b7baf94 1394 nfs4_ff_layout_stat_io_start_read(hdr->inode,
abcb7bfc 1395 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1396 hdr->args.count,
1397 task->tk_start);
37e9ed22
TM		 1398}
1399
1400static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1401 struct nfs_pgio_header *hdr)
1402{
1403 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1404 return;
1405 nfs4_ff_layout_stat_io_end_read(task,
1406 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1407 hdr->args.count,
1408 hdr->res.count);
056f9ad6 1409 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1410}
abcb7bfc 1411
37e9ed22
TM		 1412static int ff_layout_read_prepare_common(struct rpc_task *task,
1413 struct nfs_pgio_header *hdr)
1414{
d67ae825
TH		 1415 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1416 rpc_exit(task, -EIO);
1417 return -EIO;
1418 }
d67ae825 1419
37e9ed22 1420 ff_layout_read_record_layoutstats_start(task, hdr);
d67ae825
TH		 1421 return 0;
1422}
1423
1424/*
1425 * Call ops for the async read/write cases
1426 * In the case of dense layouts, the offset needs to be reset to its
1427 * original value.
1428 */
1429static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1430{
1431 struct nfs_pgio_header *hdr = data;
1432
1433 if (ff_layout_read_prepare_common(task, hdr))
1434 return;
1435
1436 rpc_call_start(task);
1437}
1438
d67ae825
TH		 1439static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1440{
1441 struct nfs_pgio_header *hdr = data;
1442
6de7e12f
AS		 1443 if (nfs4_setup_sequence(hdr->ds_clp,
1444 &hdr->args.seq_args,
1445 &hdr->res.seq_res,
1446 task))
d67ae825
TH		 1447 return;
1448
bb21ce0a 1449 ff_layout_read_prepare_common(task, hdr);
d67ae825
TH		 1450}
1451
1452static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1453{
1454 struct nfs_pgio_header *hdr = data;
1455
1456 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1457
1458 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1459 task->tk_status == 0) {
1460 nfs4_sequence_done(task, &hdr->res.seq_res);
1461 return;
1462 }
1463
1464 /* Note this may cause RPC to be resent */
1465 hdr->mds_ops->rpc_call_done(task, hdr);
1466}
1467
1468static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1469{
1470 struct nfs_pgio_header *hdr = data;
1471
37e9ed22 1472 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1473 rpc_count_iostats_metrics(task,
1474 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1475}
1476
4d0ac221
TM		 1477static void ff_layout_read_release(void *data)
1478{
1479 struct nfs_pgio_header *hdr = data;
1480
1481 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1482 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1483 ff_layout_send_layouterror(hdr->lseg);
42f86b44 1484 pnfs_read_resend_pnfs(hdr);
f0922a6c 1485 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
42f86b44 1486 ff_layout_reset_read(hdr);
4d0ac221
TM		 1487 pnfs_generic_rw_release(data);
1488}
1489
1490
d67ae825
TH		 1491static int ff_layout_write_done_cb(struct rpc_task *task,
1492 struct nfs_pgio_header *hdr)
1493{
2e18d4d8 1494 loff_t end_offs = 0;
d67ae825
TH		 1495 int err;
1496
1497 trace_nfs4_pnfs_write(hdr, task->tk_status);
dd52128a 1498 if (task->tk_status < 0)
d67ae825
TH		 1499 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1500 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1501 hdr->res.op_status, OP_WRITE,
1502 task->tk_status);
d67ae825
TH		 1503 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1504 hdr->ds_clp, hdr->lseg,
1505 hdr->pgio_mirror_idx);
1506
7b0df92a
TM		 1507 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1508 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1509 switch (err) {
1510 case -NFS4ERR_RESET_TO_PNFS:
7b0df92a 1511 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
7cc8c5cd 1512 return task->tk_status;
d67ae825 1513 case -NFS4ERR_RESET_TO_MDS:
7b0df92a 1514 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1515 return task->tk_status;
1516 case -EAGAIN:
d67ae825
TH		 1517 return -EAGAIN;
1518 }
1519
1520 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1521 hdr->res.verf->committed == NFS_DATA_SYNC)
2e18d4d8
TM		 1522 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1523
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);
d67ae825 1526
54204010
PT		 1527 /* zero out fattr since we don't care DS attr at all */
1528 hdr->fattr.valid = 0;
69f230d9
PT		 1529 if (task->tk_status >= 0)
1530 nfs_writeback_update_inode(hdr);
1531
d67ae825
TH		 1532 return 0;
1533}
1534
1535static int ff_layout_commit_done_cb(struct rpc_task *task,
1536 struct nfs_commit_data *data)
1537{
d67ae825
TH		 1538 int err;
1539
1540 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
dd52128a 1541 if (task->tk_status < 0)
d67ae825
TH		 1542 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1543 data->args.offset, data->args.count,
dd52128a
TM		 1544 data->res.op_status, OP_COMMIT,
1545 task->tk_status);
d67ae825
TH		 1546 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1547 data->lseg, data->ds_commit_index);
1548
1549 switch (err) {
1550 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1551 pnfs_generic_prepare_to_resend_writes(data);
1552 return -EAGAIN;
d67ae825 1553 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1554 pnfs_generic_prepare_to_resend_writes(data);
1555 return -EAGAIN;
1556 case -EAGAIN:
1557 rpc_restart_call_prepare(task);
1558 return -EAGAIN;
1559 }
1560
bc28e1c2 1561 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
d67ae825
TH		 1562
1563 return 0;
1564}
1565
37e9ed22
TM		 1566static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1567 struct nfs_pgio_header *hdr)
d67ae825 1568{
37e9ed22
TM		 1569 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1570 return;
0b7baf94 1571 nfs4_ff_layout_stat_io_start_write(hdr->inode,
abcb7bfc 1572 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1573 hdr->args.count,
1574 task->tk_start);
37e9ed22
TM		 1575}
1576
1577static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1578 struct nfs_pgio_header *hdr)
1579{
1580 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1581 return;
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);
056f9ad6 1586 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1587}
abcb7bfc 1588
37e9ed22
TM		 1589static int ff_layout_write_prepare_common(struct rpc_task *task,
1590 struct nfs_pgio_header *hdr)
1591{
d67ae825
TH		 1592 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1593 rpc_exit(task, -EIO);
1594 return -EIO;
1595 }
1596
37e9ed22 1597 ff_layout_write_record_layoutstats_start(task, hdr);
d67ae825
TH		 1598 return 0;
1599}
1600
1601static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1602{
1603 struct nfs_pgio_header *hdr = data;
1604
1605 if (ff_layout_write_prepare_common(task, hdr))
1606 return;
1607
1608 rpc_call_start(task);
1609}
1610
1611static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1612{
1613 struct nfs_pgio_header *hdr = data;
1614
6de7e12f
AS		 1615 if (nfs4_setup_sequence(hdr->ds_clp,
1616 &hdr->args.seq_args,
1617 &hdr->res.seq_res,
1618 task))
d67ae825
TH		 1619 return;
1620
bb21ce0a 1621 ff_layout_write_prepare_common(task, hdr);
d67ae825
TH		 1622}
1623
1624static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1625{
1626 struct nfs_pgio_header *hdr = data;
1627
1628 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1629 task->tk_status == 0) {
1630 nfs4_sequence_done(task, &hdr->res.seq_res);
1631 return;
1632 }
1633
1634 /* Note this may cause RPC to be resent */
1635 hdr->mds_ops->rpc_call_done(task, hdr);
1636}
1637
1638static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1639{
1640 struct nfs_pgio_header *hdr = data;
1641
37e9ed22 1642 ff_layout_write_record_layoutstats_done(task, hdr);
d67ae825
TH		 1643 rpc_count_iostats_metrics(task,
1644 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1645}
1646
4d0ac221
TM		 1647static void ff_layout_write_release(void *data)
1648{
1649 struct nfs_pgio_header *hdr = data;
1650
1651 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1652 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1653 ff_layout_send_layouterror(hdr->lseg);
7b0df92a 1654 ff_layout_reset_write(hdr, true);
f0922a6c 1655 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
7b0df92a 1656 ff_layout_reset_write(hdr, false);
4d0ac221
TM		 1657 pnfs_generic_rw_release(data);
1658}
1659
37e9ed22 1660static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
abcb7bfc
TM		 1661 struct nfs_commit_data *cdata)
1662{
37e9ed22
TM		 1663 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1664 return;
0b7baf94 1665 nfs4_ff_layout_stat_io_start_write(cdata->inode,
abcb7bfc 1666 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
e76d28dd 1667 0, task->tk_start);
abcb7bfc
TM		 1668}
1669
37e9ed22
TM		 1670static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1671 struct nfs_commit_data *cdata)
1672{
1673 struct nfs_page *req;
1674 __u64 count = 0;
1675
1676 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1677 return;
1678
1679 if (task->tk_status == 0) {
1680 list_for_each_entry(req, &cdata->pages, wb_list)
1681 count += req->wb_bytes;
1682 }
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);
056f9ad6 1686 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
37e9ed22
TM		 1687}
1688
1689static void ff_layout_commit_prepare_common(struct rpc_task *task,
1690 struct nfs_commit_data *cdata)
1691{
1692 ff_layout_commit_record_layoutstats_start(task, cdata);
1693}
1694
d67ae825
TH		 1695static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1696{
abcb7bfc 1697 ff_layout_commit_prepare_common(task, data);
d67ae825
TH		 1698 rpc_call_start(task);
1699}
1700
1701static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1702{
1703 struct nfs_commit_data *wdata = data;
1704
6de7e12f
AS		 1705 if (nfs4_setup_sequence(wdata->ds_clp,
1706 &wdata->args.seq_args,
1707 &wdata->res.seq_res,
1708 task))
abcb7bfc
TM		 1709 return;
1710 ff_layout_commit_prepare_common(task, data);
1711}
1712
1713static void ff_layout_commit_done(struct rpc_task *task, void *data)
141b9b59
PT		 1714{
1715 pnfs_generic_write_commit_done(task, data);
1716}
1717
1718static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
abcb7bfc
TM		 1719{
1720 struct nfs_commit_data *cdata = data;
abcb7bfc 1721
37e9ed22 1722 ff_layout_commit_record_layoutstats_done(task, cdata);
d67ae825
TH		 1723 rpc_count_iostats_metrics(task,
1724 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1725}
1726
4d0ac221
TM		 1727static void ff_layout_commit_release(void *data)
1728{
1729 struct nfs_commit_data *cdata = data;
1730
1731 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1732 pnfs_generic_commit_release(data);
1733}
1734
d67ae825
TH		 1735static 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,
4d0ac221 1739 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1740};
1741
1742static 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,
4d0ac221 1746 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1747};
1748
1749static 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,
4d0ac221 1753 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1754};
1755
1756static 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,
4d0ac221 1760 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1761};
1762
1763static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1764 .rpc_call_prepare = ff_layout_commit_prepare_v3,
abcb7bfc 1765 .rpc_call_done = ff_layout_commit_done,
d67ae825 1766 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1767 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1768};
1769
1770static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1771 .rpc_call_prepare = ff_layout_commit_prepare_v4,
abcb7bfc 1772 .rpc_call_done = ff_layout_commit_done,
d67ae825 1773 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1774 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1775};
1776
1777static enum pnfs_try_status
1778ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1779{
1780 struct pnfs_layout_segment *lseg = hdr->lseg;
1781 struct nfs4_pnfs_ds *ds;
1782 struct rpc_clnt *ds_clnt;
2444ff27 1783 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1784 const struct cred *ds_cred;
d67ae825
TH		 1785 loff_t offset = hdr->args.offset;
1786 u32 idx = hdr->pgio_mirror_idx;
1787 int vers;
1788 struct nfs_fh *fh;
1789
5b5e0928 1790 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
d67ae825
TH		 1791 __func__, hdr->inode->i_ino,
1792 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1793
2444ff27
TM		 1794 mirror = FF_LAYOUT_COMP(lseg, idx);
1795 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
d67ae825
TH		 1796 if (!ds)
1797 goto out_failed;
1798
561d6f8a 1799 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1800 hdr->inode);
1801 if (IS_ERR(ds_clnt))
1802 goto out_failed;
1803
312cd4cb 1804 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1805 if (!ds_cred)
d67ae825
TH		 1806 goto out_failed;
1807
626d48b1 1808 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1809
1810 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
212bf41d 1811 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
d67ae825 1812
75bf47eb 1813 hdr->pgio_done_cb = ff_layout_read_done_cb;
212bf41d 1814 refcount_inc(&ds->ds_clp->cl_count);
d67ae825 1815 hdr->ds_clp = ds->ds_clp;
749da527 1816 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1817 if (fh)
1818 hdr->args.fh = fh;
bb21ce0a 1819
4cbc8a57 1820 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1821
d67ae825
TH		 1822 /*
1823 * Note that if we ever decide to split across DSes,
1824 * then we may need to handle dense-like offsets.
1825 */
1826 hdr->args.offset = offset;
1827 hdr->mds_offset = offset;
1828
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);
a52458b4 1834 put_cred(ds_cred);
d67ae825
TH		 1835 return PNFS_ATTEMPTED;
1836
1837out_failed:
3b13b4b3 1838 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825 1839 return PNFS_TRY_AGAIN;
d5b9216f
TM		 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);
d67ae825
TH		 1843 return PNFS_NOT_ATTEMPTED;
1844}
1845
1846/* Perform async writes. */
1847static enum pnfs_try_status
1848ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1849{
1850 struct pnfs_layout_segment *lseg = hdr->lseg;
1851 struct nfs4_pnfs_ds *ds;
1852 struct rpc_clnt *ds_clnt;
2444ff27 1853 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1854 const struct cred *ds_cred;
d67ae825
TH		 1855 loff_t offset = hdr->args.offset;
1856 int vers;
1857 struct nfs_fh *fh;
1858 int idx = hdr->pgio_mirror_idx;
1859
2444ff27
TM		 1860 mirror = FF_LAYOUT_COMP(lseg, idx);
1861 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825 1862 if (!ds)
37f8aa16 1863 goto out_failed;
d67ae825 1864
561d6f8a 1865 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1866 hdr->inode);
1867 if (IS_ERR(ds_clnt))
37f8aa16 1868 goto out_failed;
d67ae825 1869
312cd4cb 1870 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1871 if (!ds_cred)
37f8aa16 1872 goto out_failed;
d67ae825 1873
626d48b1 1874 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825 1875
5b5e0928 1876 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
d67ae825 1877 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
212bf41d 1878 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1879 vers);
1880
1881 hdr->pgio_done_cb = ff_layout_write_done_cb;
212bf41d 1882 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1883 hdr->ds_clp = ds->ds_clp;
1884 hdr->ds_commit_idx = idx;
749da527 1885 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1886 if (fh)
1887 hdr->args.fh = fh;
1888
4cbc8a57 1889 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1890
d67ae825
TH		 1891 /*
1892 * Note that if we ever decide to split across DSes,
1893 * then we may need to handle dense-like offsets.
1894 */
1895 hdr->args.offset = offset;
1896
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);
a52458b4 1902 put_cred(ds_cred);
d67ae825 1903 return PNFS_ATTEMPTED;
37f8aa16
TM		 1904
1905out_failed:
1906 if (ff_layout_avoid_mds_available_ds(lseg))
1907 return PNFS_TRY_AGAIN;
d5b9216f
TM		 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);
37f8aa16 1911 return PNFS_NOT_ATTEMPTED;
d67ae825
TH		 1912}
1913
d67ae825
TH		 1914static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1915{
1916 return i;
1917}
1918
1919static struct nfs_fh *
1920select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1921{
1922 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1923
1924 /* FIXME: Assume that there is only one NFS version available
1925 * for the DS.
1926 */
1927 return &flseg->mirror_array[i]->fh_versions[0];
1928}
1929
1930static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1931{
1932 struct pnfs_layout_segment *lseg = data->lseg;
1933 struct nfs4_pnfs_ds *ds;
1934 struct rpc_clnt *ds_clnt;
2444ff27 1935 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1936 const struct cred *ds_cred;
d67ae825 1937 u32 idx;
57f3f4c0 1938 int vers, ret;
d67ae825
TH		 1939 struct nfs_fh *fh;
1940
4b75053e
TM		 1941 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1942 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1943 goto out_err;
1944
d67ae825 1945 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
2444ff27
TM		 1946 mirror = FF_LAYOUT_COMP(lseg, idx);
1947 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825
TH		 1948 if (!ds)
1949 goto out_err;
1950
561d6f8a 1951 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1952 data->inode);
1953 if (IS_ERR(ds_clnt))
1954 goto out_err;
1955
312cd4cb 1956 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
57f3f4c0 1957 if (!ds_cred)
d67ae825
TH		 1958 goto out_err;
1959
626d48b1 1960 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1961
1962 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
212bf41d 1963 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1964 vers);
1965 data->commit_done_cb = ff_layout_commit_done_cb;
1966 data->cred = ds_cred;
212bf41d 1967 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1968 data->ds_clp = ds->ds_clp;
1969 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1970 if (fh)
1971 data->args.fh = fh;
abcb7bfc 1972
57f3f4c0 1973 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
d67ae825
TH		 1974 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1975 &ff_layout_commit_call_ops_v4,
1976 how, RPC_TASK_SOFTCONN);
a52458b4 1977 put_cred(ds_cred);
57f3f4c0 1978 return ret;
d67ae825
TH		 1979out_err:
1980 pnfs_generic_prepare_to_resend_writes(data);
1981 pnfs_generic_commit_release(data);
1982 return -EAGAIN;
1983}
1984
1985static int
1986ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1987 int how, struct nfs_commit_info *cinfo)
1988{
1989 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1990 ff_layout_initiate_commit);
1991}
1992
1993static struct pnfs_ds_commit_info *
1994ff_layout_get_ds_info(struct inode *inode)
1995{
1996 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1997
1998 if (layout == NULL)
1999 return NULL;
2000
2001 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2002}
2003
2004static void
fc87701b 2005ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
d67ae825
TH		 2006{
2007 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2008 id_node));
2009}
2010
5ba6a09e 2011static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
5b9b3c85
TM		 2012 const struct nfs4_layoutreturn_args *args,
2013 const struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825 2014{
d67ae825 2015 __be32 *start;
d67ae825
TH		 2016
2017 start = xdr_reserve_space(xdr, 4);
2018 if (unlikely(!start))
2019 return -E2BIG;
2020
5b9b3c85 2021 *start = cpu_to_be32(ff_args->num_errors);
d67ae825 2022 /* This assume we always return _ALL_ layouts */
5b9b3c85 2023 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
d67ae825
TH		 2024}
2025
230bc962
TM		 2026static void
2027encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2028{
ab6e9aaf 2029 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
230bc962
TM		 2030}
2031
2032static void
2033ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2034 const nfs4_stateid *stateid,
2035 const struct nfs42_layoutstat_devinfo *devinfo)
2036{
2037 __be32 *p;
2038
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);
2049}
2050
2051static void
2052ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2053 const nfs4_stateid *stateid,
2054 const struct nfs42_layoutstat_devinfo *devinfo)
2055{
2056 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2057 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2058 devinfo->ld_private.data);
2059}
2060
d67ae825 2061/* report nothing for now */
230bc962
TM		 2062static 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)
d67ae825
TH		 2065{
2066 __be32 *p;
230bc962 2067 int i;
d67ae825
TH		 2068
2069 p = xdr_reserve_space(xdr, 4);
230bc962
TM		 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]);
2075}
2076
2077static void
2078ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2079 unsigned int num_entries)
2080{
2081 unsigned int i;
2082
2083 for (i = 0; i < num_entries; i++) {
2084 if (!devinfo[i].ld_private.ops)
2085 continue;
2086 if (!devinfo[i].ld_private.ops->free)
2087 continue;
2088 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2089 }
d67ae825
TH		 2090}
2091
2092static struct nfs4_deviceid_node *
2093ff_layout_alloc_deviceid_node(struct nfs_server *server,
2094 struct pnfs_device *pdev, gfp_t gfp_flags)
2095{
2096 struct nfs4_ff_layout_ds *dsaddr;
2097
2098 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2099 if (!dsaddr)
2100 return NULL;
2101 return &dsaddr->id_node;
2102}
2103
2104static void
94e5c571 2105ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
5b9b3c85
TM		 2106 const void *voidargs,
2107 const struct nfs4_xdr_opaque_data *ff_opaque)
d67ae825 2108{
5b9b3c85 2109 const struct nfs4_layoutreturn_args *args = voidargs;
5ba6a09e 2110 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
d9152114
TM		 2111 struct xdr_buf tmp_buf = {
2112 .head = {
2113 [0] = {
2114 .iov_base = page_address(ff_args->pages[0]),
2115 },
2116 },
2117 .buflen = PAGE_SIZE,
2118 };
2119 struct xdr_stream tmp_xdr;
d67ae825
TH		 2120 __be32 *start;
2121
2122 dprintk("%s: Begin\n", __func__);
d67ae825 2123
0ccc61b1 2124 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
d9152114
TM		 2125
2126 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2127 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2128
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);
082fa37d 2132
d67ae825
TH		 2133 dprintk("%s: Return\n", __func__);
2134}
2135
5b9b3c85
TM		 2136static void
2137ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2138{
2139 struct nfs4_flexfile_layoutreturn_args *ff_args;
2140
2141 if (!args->data)
2142 return;
2143 ff_args = args->data;
2144 args->data = NULL;
2145
2146 ff_layout_free_ds_ioerr(&ff_args->errors);
230bc962 2147 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
5b9b3c85 2148
d9152114 2149 put_page(ff_args->pages[0]);
5b9b3c85
TM		 2150 kfree(ff_args);
2151}
2152
2e54b9b1 2153static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
5b9b3c85
TM		 2154 .encode = ff_layout_encode_layoutreturn,
2155 .free = ff_layout_free_layoutreturn,
2156};
2157
2158static int
2159ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2160{
2161 struct nfs4_flexfile_layoutreturn_args *ff_args;
230bc962 2162 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
5b9b3c85
TM		 2163
2164 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2165 if (!ff_args)
d9152114
TM		 2166 goto out_nomem;
2167 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2168 if (!ff_args->pages[0])
2169 goto out_nomem_free;
5b9b3c85
TM		 2170
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);
2175
230bc962
TM		 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);
2180
5b9b3c85
TM		 2181 args->ld_private->ops = &layoutreturn_ops;
2182 args->ld_private->data = ff_args;
2183 return 0;
d9152114
TM		 2184out_nomem_free:
2185 kfree(ff_args);
2186out_nomem:
2187 return -ENOMEM;
5b9b3c85
TM		 2188}
2189
f0922a6c
TM		 2190#ifdef CONFIG_NFS_V4_2
2191void
2192ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2193{
2194 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2195 struct nfs42_layout_error *errors;
2196 LIST_HEAD(head);
2197
2198 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2199 return;
2200 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2201 if (list_empty(&head))
2202 return;
2203
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;
2208 size_t n = 0;
2209
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;
2217 n++;
2218 if (!list_is_last(&pos->list, &head) &&
2219 n < NFS42_LAYOUTERROR_MAX)
2220 continue;
2221 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2222 break;
2223 n = 0;
2224 }
2225 kfree(errors);
2226 }
2227 ff_layout_free_ds_ioerr(&head);
2228}
2229#else
2230void
2231ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2232{
2233}
2234#endif
2235
27c43064
PT		 2236static int
2237ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2238{
2239 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2240
2241 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2242}
2243
2244static size_t
2245ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2246 const int buflen)
2247{
2248 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2249 const struct in6_addr *addr = &sin6->sin6_addr;
2250
2251 /*
2252 * RFC 4291, Section 2.2.2
2253 *
2254 * Shorthanded ANY address
2255 */
2256 if (ipv6_addr_any(addr))
2257 return snprintf(buf, buflen, "::");
2258
2259 /*
2260 * RFC 4291, Section 2.2.2
2261 *
2262 * Shorthanded loopback address
2263 */
2264 if (ipv6_addr_loopback(addr))
2265 return snprintf(buf, buflen, "::1");
2266
2267 /*
2268 * RFC 4291, Section 2.2.3
2269 *
2270 * Special presentation address format for mapped v4
2271 * addresses.
2272 */
2273 if (ipv6_addr_v4mapped(addr))
2274 return snprintf(buf, buflen, "::ffff:%pI4",
2275 &addr->s6_addr32[3]);
2276
2277 /*
2278 * RFC 4291, Section 2.2.1
2279 */
2280 return snprintf(buf, buflen, "%pI6c", addr);
2281}
2282
2283/* Derived from rpc_sockaddr2uaddr */
2284static void
2285ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2286{
2287 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2288 char portbuf[RPCBIND_MAXUADDRPLEN];
2289 char addrbuf[RPCBIND_MAXUADDRLEN];
2290 char *netid;
2291 unsigned short port;
2292 int len, netid_len;
2293 __be32 *p;
2294
2295 switch (sap->sa_family) {
2296 case AF_INET:
2297 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2298 return;
2299 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2300 netid = "tcp";
2301 netid_len = 3;
2302 break;
2303 case AF_INET6:
2304 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2305 return;
2306 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2307 netid = "tcp6";
2308 netid_len = 4;
2309 break;
2310 default:
2311 /* we only support tcp and tcp6 */
2312 WARN_ON_ONCE(1);
2313 return;
2314 }
2315
2316 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2317 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2318
2319 p = xdr_reserve_space(xdr, 4 + netid_len);
2320 xdr_encode_opaque(p, netid, netid_len);
2321
2322 p = xdr_reserve_space(xdr, 4 + len);
2323 xdr_encode_opaque(p, addrbuf, len);
2324}
2325
2326static void
2327ff_layout_encode_nfstime(struct xdr_stream *xdr,
2328 ktime_t t)
2329{
2330 struct timespec64 ts;
2331 __be32 *p;
2332
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);
2337}
2338
2339static void
2340ff_layout_encode_io_latency(struct xdr_stream *xdr,
2341 struct nfs4_ff_io_stat *stat)
2342{
2343 __be32 *p;
2344
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);
2353}
2354
2355static void
422c93c8
TM		 2356ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2357 const struct nfs42_layoutstat_devinfo *devinfo,
2358 struct nfs4_ff_layout_mirror *mirror)
27c43064 2359{
27c43064
PT		 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];
422c93c8 2363 __be32 *p;
27c43064
PT		 2364
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);
27c43064
PT		 2368 /* netaddr4 */
2369 ff_layout_encode_netaddr(xdr, da);
2370 /* nfs_fh4 */
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);
2379 /* nfstime4 */
2380 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2381 /* bool */
2382 p = xdr_reserve_space(xdr, 4);
2383 *p = cpu_to_be32(false);
422c93c8
TM		 2384}
2385
2386static void
2387ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2388 const struct nfs4_xdr_opaque_data *opaque)
2389{
2390 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2391 struct nfs42_layoutstat_devinfo, ld_private);
2392 __be32 *start;
2393
2394 /* layoutupdate length */
2395 start = xdr_reserve_space(xdr, 4);
2396 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
27c43064
PT		 2397
2398 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2399}
2400
422c93c8
TM		 2401static void
2402ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2403{
2404 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2405
2406 ff_layout_put_mirror(mirror);
2407}
2408
2409static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2410 .encode = ff_layout_encode_layoutstats,
2411 .free = ff_layout_free_layoutstats,
2412};
2413
266d12d4 2414static int
08e2e5bc
TM		 2415ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2416 struct nfs42_layoutstat_devinfo *devinfo,
266d12d4 2417 int dev_limit)
ad4dc53e 2418{
266d12d4 2419 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
ad4dc53e
PT		 2420 struct nfs4_ff_layout_mirror *mirror;
2421 struct nfs4_deviceid_node *dev;
266d12d4 2422 int i = 0;
ad4dc53e 2423
266d12d4
TM		 2424 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2425 if (i >= dev_limit)
ad4dc53e 2426 break;
65990d1a 2427 if (IS_ERR_OR_NULL(mirror->mirror_ds))
27c43064 2428 continue;
06946c6a
TM		 2429 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2430 continue;
266d12d4 2431 /* mirror refcount put in cleanup_layoutstats */
81a090b9 2432 if (!refcount_inc_not_zero(&mirror->ref))
266d12d4
TM		 2433 continue;
2434 dev = &mirror->mirror_ds->id_node;
ad4dc53e 2435 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
266d12d4
TM		 2436 devinfo->offset = 0;
2437 devinfo->length = NFS4_MAX_UINT64;
2f8220c1 2438 spin_lock(&mirror->lock);
d099d7b8 2439 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
ad4dc53e 2440 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
d099d7b8 2441 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
ad4dc53e 2442 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2f8220c1 2443 spin_unlock(&mirror->lock);
ad4dc53e 2444 devinfo->layout_type = LAYOUT_FLEX_FILES;
422c93c8
TM		 2445 devinfo->ld_private.ops = &layoutstat_ops;
2446 devinfo->ld_private.data = mirror;
ad4dc53e 2447
08e2e5bc 2448 devinfo++;
266d12d4 2449 i++;
ad4dc53e 2450 }
266d12d4 2451 return i;
ad4dc53e
PT		 2452}
2453
2454static int
2455ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2456{
266d12d4 2457 struct nfs4_flexfile_layout *ff_layout;
08e2e5bc 2458 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
ad4dc53e 2459
ad4dc53e 2460 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
266d12d4 2461 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
ad4dc53e
PT		 2462 if (!args->devinfo)
2463 return -ENOMEM;
2464
ad4dc53e 2465 spin_lock(&args->inode->i_lock);
08e2e5bc
TM		 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);
ad4dc53e 2469 spin_unlock(&args->inode->i_lock);
46c98c6d
TM		 2470 if (!args->num_dev) {
2471 kfree(args->devinfo);
2472 args->devinfo = NULL;
2473 return -ENOENT;
2474 }
ad4dc53e
PT		 2475
2476 return 0;
2477}
2478
76b2a303
TM		 2479static int
2480ff_layout_set_layoutdriver(struct nfs_server *server,
2481 const struct nfs_fh *dummy)
2482{
2483#if IS_ENABLED(CONFIG_NFS_V4_2)
2484 server->caps |= NFS_CAP_LAYOUTSTATS;
2485#endif
2486 return 0;
2487}
2488
d67ae825
TH		 2489static struct pnfs_layoutdriver_type flexfilelayout_type = {
2490 .id = LAYOUT_FLEX_FILES,
2491 .name = "LAYOUT_FLEX_FILES",
2492 .owner = THIS_MODULE,
f86c3ac5 2493 .flags = PNFS_LAYOUTGET_ON_OPEN,
28ced9a8 2494 .max_layoutget_response = 4096, /* 1 page or so... */
76b2a303 2495 .set_layoutdriver = ff_layout_set_layoutdriver,
d67ae825
TH		 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,
0762ed2c 2500 .add_lseg = ff_layout_add_lseg,
d67ae825
TH		 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,
fc87701b 2504 .free_deviceid_node = ff_layout_free_deviceid_node,
338d00cf 2505 .mark_request_commit = pnfs_layout_mark_request_commit,
d67ae825
TH		 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,
5b9b3c85 2513 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
5bb89b47 2514 .sync = pnfs_nfs_generic_sync,
ad4dc53e 2515 .prepare_layoutstats = ff_layout_prepare_layoutstats,
d67ae825
TH		 2516};
2517
2518static int __init nfs4flexfilelayout_init(void)
2519{
2520 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2521 __func__);
2522 return pnfs_register_layoutdriver(&flexfilelayout_type);
2523}
2524
2525static void __exit nfs4flexfilelayout_exit(void)
2526{
2527 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2528 __func__);
2529 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2530}
2531
2532MODULE_ALIAS("nfs-layouttype4-4");
2533
2534MODULE_LICENSE("GPL");
2535MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2536
2537module_init(nfs4flexfilelayout_init);
2538module_exit(nfs4flexfilelayout_exit);
33344e0f
TM		 2539
2540module_param(io_maxretrans, ushort, 0644);
2541MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2542 "retries an I/O request before returning an error. ");
Michael's Linux tree.