]> 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: Allow applications to speed up readdir+statx() using AT_STATX_DONT_SYNC
[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
9c455a8c 35static const struct pnfs_commit_ops ff_layout_commit_ops;
54e4a0df
TM		 36static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
37 struct nfs_pgio_header *hdr);
230bc962
TM		 38static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
39 struct nfs42_layoutstat_devinfo *devinfo,
40 int dev_limit);
41static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
42 const struct nfs42_layoutstat_devinfo *devinfo,
43 struct nfs4_ff_layout_mirror *mirror);
54e4a0df 44
d67ae825
TH		 45static struct pnfs_layout_hdr *
46ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
47{
48 struct nfs4_flexfile_layout *ffl;
49
50 ffl = kzalloc(sizeof(*ffl), gfp_flags);
51 if (ffl) {
c21e7168 52 pnfs_init_ds_commit_info(&ffl->commit_info);
d67ae825 53 INIT_LIST_HEAD(&ffl->error_list);
266d12d4 54 INIT_LIST_HEAD(&ffl->mirrors);
1c8d477a 55 ffl->last_report_time = ktime_get();
9c455a8c 56 ffl->commit_info.ops = &ff_layout_commit_ops;
d67ae825
TH		 57 return &ffl->generic_hdr;
58 } else
59 return NULL;
60}
61
62static void
63ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
64{
cf6605d1 65 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
d67ae825
TH		 66 struct nfs4_ff_layout_ds_err *err, *n;
67
cf6605d1 68 list_for_each_entry_safe(err, n, &ffl->error_list, list) {
d67ae825
TH		 69 list_del(&err->list);
70 kfree(err);
71 }
cf6605d1 72 kfree_rcu(ffl, generic_hdr.plh_rcu);
d67ae825
TH		 73}
74
93b717fd 75static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
d67ae825
TH		 76{
77 __be32 *p;
78
79 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
80 if (unlikely(p == NULL))
81 return -ENOBUFS;
93b717fd
TM		 82 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
83 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
d67ae825
TH		 84 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
85 p[0], p[1], p[2], p[3]);
86 return 0;
87}
88
89static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
90{
91 __be32 *p;
92
93 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
94 if (unlikely(!p))
95 return -ENOBUFS;
96 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
97 nfs4_print_deviceid(devid);
98 return 0;
99}
100
101static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
102{
103 __be32 *p;
104
105 p = xdr_inline_decode(xdr, 4);
106 if (unlikely(!p))
107 return -ENOBUFS;
108 fh->size = be32_to_cpup(p++);
109 if (fh->size > sizeof(struct nfs_fh)) {
110 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
111 fh->size);
112 return -EOVERFLOW;
113 }
114 /* fh.data */
115 p = xdr_inline_decode(xdr, fh->size);
116 if (unlikely(!p))
117 return -ENOBUFS;
118 memcpy(&fh->data, p, fh->size);
119 dprintk("%s: fh len %d\n", __func__, fh->size);
120
121 return 0;
122}
123
124/*
125 * Currently only stringified uids and gids are accepted.
126 * I.e., kerberos is not supported to the DSes, so no pricipals.
127 *
128 * That means that one common function will suffice, but when
129 * principals are added, this should be split to accomodate
130 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
131 */
132static int
133decode_name(struct xdr_stream *xdr, u32 *id)
134{
135 __be32 *p;
136 int len;
137
138 /* opaque_length(4)*/
139 p = xdr_inline_decode(xdr, 4);
140 if (unlikely(!p))
141 return -ENOBUFS;
142 len = be32_to_cpup(p++);
143 if (len < 0)
144 return -EINVAL;
145
146 dprintk("%s: len %u\n", __func__, len);
147
148 /* opaque body */
149 p = xdr_inline_decode(xdr, len);
150 if (unlikely(!p))
151 return -ENOBUFS;
152
153 if (!nfs_map_string_to_numeric((char *)p, len, id))
154 return -EINVAL;
155
156 return 0;
157}
158
266d12d4
TM		 159static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
160 const struct nfs4_ff_layout_mirror *m2)
161{
162 int i, j;
163
164 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
165 return false;
166 for (i = 0; i < m1->fh_versions_cnt; i++) {
167 bool found_fh = false;
86fb449b 168 for (j = 0; j < m2->fh_versions_cnt; j++) {
266d12d4
TM		 169 if (nfs_compare_fh(&m1->fh_versions[i],
170 &m2->fh_versions[j]) == 0) {
171 found_fh = true;
172 break;
173 }
174 }
175 if (!found_fh)
176 return false;
177 }
178 return true;
179}
180
181static struct nfs4_ff_layout_mirror *
182ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
183 struct nfs4_ff_layout_mirror *mirror)
184{
185 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
186 struct nfs4_ff_layout_mirror *pos;
187 struct inode *inode = lo->plh_inode;
188
189 spin_lock(&inode->i_lock);
190 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
65990d1a 191 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
266d12d4
TM		 192 continue;
193 if (!ff_mirror_match_fh(mirror, pos))
194 continue;
81a090b9 195 if (refcount_inc_not_zero(&pos->ref)) {
266d12d4
TM		 196 spin_unlock(&inode->i_lock);
197 return pos;
198 }
199 }
200 list_add(&mirror->mirrors, &ff_layout->mirrors);
201 mirror->layout = lo;
202 spin_unlock(&inode->i_lock);
203 return mirror;
204}
205
e3b1df2d 206static void
266d12d4
TM		 207ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
208{
209 struct inode *inode;
210 if (mirror->layout == NULL)
211 return;
212 inode = mirror->layout->plh_inode;
213 spin_lock(&inode->i_lock);
214 list_del(&mirror->mirrors);
215 spin_unlock(&inode->i_lock);
216 mirror->layout = NULL;
217}
218
28a0d72c
TM		 219static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
220{
221 struct nfs4_ff_layout_mirror *mirror;
222
223 mirror = kzalloc(sizeof(*mirror), gfp_flags);
224 if (mirror != NULL) {
225 spin_lock_init(&mirror->lock);
81a090b9 226 refcount_set(&mirror->ref, 1);
266d12d4 227 INIT_LIST_HEAD(&mirror->mirrors);
28a0d72c
TM		 228 }
229 return mirror;
230}
231
232static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
233{
a52458b4 234 const struct cred *cred;
3064b686 235
266d12d4 236 ff_layout_remove_mirror(mirror);
28a0d72c 237 kfree(mirror->fh_versions);
3064b686 238 cred = rcu_access_pointer(mirror->ro_cred);
a52458b4 239 put_cred(cred);
3064b686 240 cred = rcu_access_pointer(mirror->rw_cred);
a52458b4 241 put_cred(cred);
28a0d72c
TM		 242 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
243 kfree(mirror);
244}
245
246static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
247{
81a090b9 248 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
28a0d72c
TM		 249 ff_layout_free_mirror(mirror);
250}
251
d67ae825
TH		 252static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
253{
329651b1
TM		 254 u32 i;
255
256 for (i = 0; i < fls->mirror_array_cnt; i++)
257 ff_layout_put_mirror(fls->mirror_array[i]);
d67ae825
TH		 258}
259
d67ae825
TH		 260static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
261{
262 if (fls) {
263 ff_layout_free_mirror_array(fls);
264 kfree(fls);
265 }
266}
267
660d1eb2
TM		 268static bool
269ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
270 struct pnfs_layout_segment *l2)
271{
272 const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
273 const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
274 u32 i;
275
276 if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
277 return false;
278 for (i = 0; i < fl1->mirror_array_cnt; i++) {
279 if (fl1->mirror_array[i] != fl2->mirror_array[i])
280 return false;
281 }
282 return true;
283}
284
0762ed2c
TM		 285static bool
286ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
287 const struct pnfs_layout_range *l2)
288{
289 u64 end1, end2;
290
291 if (l1->iomode != l2->iomode)
292 return l1->iomode != IOMODE_READ;
293 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
294 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
295 if (end1 < l2->offset)
296 return false;
297 if (end2 < l1->offset)
298 return true;
299 return l2->offset <= l1->offset;
300}
301
302static bool
303ff_lseg_merge(struct pnfs_layout_segment *new,
304 struct pnfs_layout_segment *old)
305{
306 u64 new_end, old_end;
307
ee26bdd6
JL		 308 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
309 return false;
0762ed2c
TM		 310 if (new->pls_range.iomode != old->pls_range.iomode)
311 return false;
312 old_end = pnfs_calc_offset_end(old->pls_range.offset,
313 old->pls_range.length);
314 if (old_end < new->pls_range.offset)
315 return false;
316 new_end = pnfs_calc_offset_end(new->pls_range.offset,
317 new->pls_range.length);
318 if (new_end < old->pls_range.offset)
319 return false;
660d1eb2
TM		 320 if (!ff_lseg_match_mirrors(new, old))
321 return false;
0762ed2c
TM		 322
323 /* Mergeable: copy info from 'old' to 'new' */
324 if (new_end < old_end)
325 new_end = old_end;
326 if (new->pls_range.offset < old->pls_range.offset)
327 new->pls_range.offset = old->pls_range.offset;
328 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
329 new_end);
330 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
331 set_bit(NFS_LSEG_ROC, &new->pls_flags);
0762ed2c
TM		 332 return true;
333}
334
335static void
336ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
337 struct pnfs_layout_segment *lseg,
338 struct list_head *free_me)
339{
340 pnfs_generic_layout_insert_lseg(lo, lseg,
341 ff_lseg_range_is_after,
342 ff_lseg_merge,
343 free_me);
344}
345
d67ae825
TH		 346static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
347{
d67ae825
TH		 348 int i, j;
349
350 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
351 for (j = i + 1; j < fls->mirror_array_cnt; j++)
352 if (fls->mirror_array[i]->efficiency <
455b6ee6
FF		 353 fls->mirror_array[j]->efficiency)
354 swap(fls->mirror_array[i],
355 fls->mirror_array[j]);
d67ae825
TH		 356 }
357}
358
359static struct pnfs_layout_segment *
360ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
361 struct nfs4_layoutget_res *lgr,
362 gfp_t gfp_flags)
363{
364 struct pnfs_layout_segment *ret;
365 struct nfs4_ff_layout_segment *fls = NULL;
366 struct xdr_stream stream;
367 struct xdr_buf buf;
368 struct page *scratch;
369 u64 stripe_unit;
370 u32 mirror_array_cnt;
371 __be32 *p;
372 int i, rc;
373
374 dprintk("--> %s\n", __func__);
375 scratch = alloc_page(gfp_flags);
376 if (!scratch)
377 return ERR_PTR(-ENOMEM);
378
379 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
380 lgr->layoutp->len);
381 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
382
383 /* stripe unit and mirror_array_cnt */
384 rc = -EIO;
385 p = xdr_inline_decode(&stream, 8 + 4);
386 if (!p)
387 goto out_err_free;
388
389 p = xdr_decode_hyper(p, &stripe_unit);
390 mirror_array_cnt = be32_to_cpup(p++);
391 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
392 stripe_unit, mirror_array_cnt);
393
394 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
395 mirror_array_cnt == 0)
396 goto out_err_free;
397
398 rc = -ENOMEM;
329651b1
TM		 399 fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
400 gfp_flags);
d67ae825
TH		 401 if (!fls)
402 goto out_err_free;
403
404 fls->mirror_array_cnt = mirror_array_cnt;
405 fls->stripe_unit = stripe_unit;
d67ae825
TH		 406
407 for (i = 0; i < fls->mirror_array_cnt; i++) {
266d12d4 408 struct nfs4_ff_layout_mirror *mirror;
97f68c6b 409 struct cred *kcred;
762bb7e9 410 const struct cred __rcu *cred;
8276c902
N		 411 kuid_t uid;
412 kgid_t gid;
90a0be00 413 u32 ds_count, fh_count, id;
d67ae825
TH		 414 int j;
415
416 rc = -EIO;
417 p = xdr_inline_decode(&stream, 4);
418 if (!p)
419 goto out_err_free;
420 ds_count = be32_to_cpup(p);
421
422 /* FIXME: allow for striping? */
423 if (ds_count != 1)
424 goto out_err_free;
425
28a0d72c 426 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
d67ae825
TH		 427 if (fls->mirror_array[i] == NULL) {
428 rc = -ENOMEM;
429 goto out_err_free;
430 }
431
d67ae825
TH		 432 fls->mirror_array[i]->ds_count = ds_count;
433
434 /* deviceid */
65990d1a 435 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
d67ae825
TH		 436 if (rc)
437 goto out_err_free;
438
d67ae825
TH		 439 /* efficiency */
440 rc = -EIO;
441 p = xdr_inline_decode(&stream, 4);
442 if (!p)
443 goto out_err_free;
444 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
445
446 /* stateid */
93b717fd 447 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
d67ae825
TH		 448 if (rc)
449 goto out_err_free;
450
451 /* fh */
662f9a10 452 rc = -EIO;
d67ae825
TH		 453 p = xdr_inline_decode(&stream, 4);
454 if (!p)
455 goto out_err_free;
456 fh_count = be32_to_cpup(p);
457
458 fls->mirror_array[i]->fh_versions =
6396bb22 459 kcalloc(fh_count, sizeof(struct nfs_fh),
d67ae825
TH		 460 gfp_flags);
461 if (fls->mirror_array[i]->fh_versions == NULL) {
462 rc = -ENOMEM;
463 goto out_err_free;
464 }
465
466 for (j = 0; j < fh_count; j++) {
467 rc = decode_nfs_fh(&stream,
468 &fls->mirror_array[i]->fh_versions[j]);
469 if (rc)
470 goto out_err_free;
471 }
472
473 fls->mirror_array[i]->fh_versions_cnt = fh_count;
474
475 /* user */
90a0be00 476 rc = decode_name(&stream, &id);
d67ae825
TH		 477 if (rc)
478 goto out_err_free;
479
8276c902 480 uid = make_kuid(&init_user_ns, id);
90a0be00 481
d67ae825 482 /* group */
90a0be00 483 rc = decode_name(&stream, &id);
d67ae825
TH		 484 if (rc)
485 goto out_err_free;
486
8276c902 487 gid = make_kgid(&init_user_ns, id);
90a0be00 488
97f68c6b
N		 489 if (gfp_flags & __GFP_FS)
490 kcred = prepare_kernel_cred(NULL);
491 else {
492 unsigned int nofs_flags = memalloc_nofs_save();
493 kcred = prepare_kernel_cred(NULL);
494 memalloc_nofs_restore(nofs_flags);
495 }
496 rc = -ENOMEM;
497 if (!kcred)
498 goto out_err_free;
8276c902
N		 499 kcred->fsuid = uid;
500 kcred->fsgid = gid;
762bb7e9 501 cred = RCU_INITIALIZER(kcred);
90a0be00 502
3064b686
JL		 503 if (lgr->range.iomode == IOMODE_READ)
504 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
505 else
506 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
90a0be00 507
266d12d4
TM		 508 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
509 if (mirror != fls->mirror_array[i]) {
90a0be00 510 /* swap cred ptrs so free_mirror will clean up old */
3064b686
JL		 511 if (lgr->range.iomode == IOMODE_READ) {
512 cred = xchg(&mirror->ro_cred, cred);
513 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
514 } else {
515 cred = xchg(&mirror->rw_cred, cred);
516 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
517 }
266d12d4
TM		 518 ff_layout_free_mirror(fls->mirror_array[i]);
519 fls->mirror_array[i] = mirror;
520 }
521
3064b686
JL		 522 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
523 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
8276c902
N		 524 from_kuid(&init_user_ns, uid),
525 from_kgid(&init_user_ns, gid));
d67ae825
TH		 526 }
527
c0f5f505 528 p = xdr_inline_decode(&stream, 4);
d0379a5d
TM		 529 if (!p)
530 goto out_sort_mirrors;
531 fls->flags = be32_to_cpup(p);
532
533 p = xdr_inline_decode(&stream, 4);
534 if (!p)
535 goto out_sort_mirrors;
536 for (i=0; i < fls->mirror_array_cnt; i++)
537 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
c0f5f505 538
d0379a5d 539out_sort_mirrors:
d67ae825 540 ff_layout_sort_mirrors(fls);
d67ae825
TH		 541 ret = &fls->generic_hdr;
542 dprintk("<-- %s (success)\n", __func__);
543out_free_page:
544 __free_page(scratch);
545 return ret;
546out_err_free:
547 _ff_layout_free_lseg(fls);
548 ret = ERR_PTR(rc);
549 dprintk("<-- %s (%d)\n", __func__, rc);
550 goto out_free_page;
551}
552
d67ae825
TH		 553static void
554ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
555{
556 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825
TH		 557
558 dprintk("--> %s\n", __func__);
559
d67ae825
TH		 560 if (lseg->pls_range.iomode == IOMODE_RW) {
561 struct nfs4_flexfile_layout *ffl;
562 struct inode *inode;
563
564 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
565 inode = ffl->generic_hdr.plh_inode;
566 spin_lock(&inode->i_lock);
a9901899 567 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
d67ae825
TH		 568 spin_unlock(&inode->i_lock);
569 }
570 _ff_layout_free_lseg(fls);
571}
572
abcb7bfc 573static void
e76d28dd 574nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 575{
9bbd9bb4
PT		 576 /* first IO request? */
577 if (atomic_inc_return(&timer->n_ops) == 1) {
e76d28dd 578 timer->start_time = now;
abcb7bfc
TM		 579 }
580}
581
582static ktime_t
e76d28dd 583nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 584{
e76d28dd 585 ktime_t start;
abcb7bfc 586
9bbd9bb4
PT		 587 if (atomic_dec_return(&timer->n_ops) < 0)
588 WARN_ON_ONCE(1);
589
9bbd9bb4
PT		 590 start = timer->start_time;
591 timer->start_time = now;
abcb7bfc
TM		 592 return ktime_sub(now, start);
593}
594
97ba375b 595static bool
d983803d 596nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
e76d28dd
TM		 597 struct nfs4_ff_layoutstat *layoutstat,
598 ktime_t now)
abcb7bfc 599{
bbf58bf3 600 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
1c8d477a 601 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
d983803d 602
e76d28dd 603 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
1f3a8e49 604 if (!mirror->start_time)
9bbd9bb4 605 mirror->start_time = now;
d0379a5d
TM		 606 if (mirror->report_interval != 0)
607 report_interval = (s64)mirror->report_interval * 1000LL;
608 else if (layoutstats_timer != 0)
bbf58bf3 609 report_interval = (s64)layoutstats_timer * 1000LL;
1c8d477a 610 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
bbf58bf3 611 report_interval) {
1c8d477a 612 ffl->last_report_time = now;
97ba375b
PT		 613 return true;
614 }
615
616 return false;
abcb7bfc
TM		 617}
618
619static void
620nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
621 __u64 requested)
622{
623 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
624
625 iostat->ops_requested++;
626 iostat->bytes_requested += requested;
627}
628
629static void
630nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
631 __u64 requested,
632 __u64 completed,
e76d28dd
TM		 633 ktime_t time_completed,
634 ktime_t time_started)
abcb7bfc
TM		 635{
636 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
e76d28dd 637 ktime_t completion_time = ktime_sub(time_completed, time_started);
abcb7bfc
TM		 638 ktime_t timer;
639
640 iostat->ops_completed++;
641 iostat->bytes_completed += completed;
642 iostat->bytes_not_delivered += requested - completed;
643
e76d28dd 644 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
abcb7bfc
TM		 645 iostat->total_busy_time =
646 ktime_add(iostat->total_busy_time, timer);
647 iostat->aggregate_completion_time =
e76d28dd
TM		 648 ktime_add(iostat->aggregate_completion_time,
649 completion_time);
abcb7bfc
TM		 650}
651
652static void
0b7baf94
TM		 653nfs4_ff_layout_stat_io_start_read(struct inode *inode,
654 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 655 __u64 requested, ktime_t now)
abcb7bfc 656{
97ba375b
PT		 657 bool report;
658
abcb7bfc 659 spin_lock(&mirror->lock);
e76d28dd 660 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
abcb7bfc 661 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
06946c6a 662 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 663 spin_unlock(&mirror->lock);
97ba375b
PT		 664
665 if (report)
0b7baf94 666 pnfs_report_layoutstat(inode, GFP_KERNEL);
abcb7bfc
TM		 667}
668
669static void
670nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
671 struct nfs4_ff_layout_mirror *mirror,
672 __u64 requested,
673 __u64 completed)
674{
675 spin_lock(&mirror->lock);
676 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
677 requested, completed,
e76d28dd 678 ktime_get(), task->tk_start);
06946c6a 679 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 680 spin_unlock(&mirror->lock);
681}
682
683static void
0b7baf94
TM		 684nfs4_ff_layout_stat_io_start_write(struct inode *inode,
685 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 686 __u64 requested, ktime_t now)
abcb7bfc 687{
97ba375b
PT		 688 bool report;
689
abcb7bfc 690 spin_lock(&mirror->lock);
e76d28dd 691 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
abcb7bfc 692 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
06946c6a 693 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 694 spin_unlock(&mirror->lock);
97ba375b
PT		 695
696 if (report)
0b7baf94 697 pnfs_report_layoutstat(inode, GFP_NOIO);
abcb7bfc
TM		 698}
699
700static void
701nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
702 struct nfs4_ff_layout_mirror *mirror,
703 __u64 requested,
704 __u64 completed,
705 enum nfs3_stable_how committed)
706{
707 if (committed == NFS_UNSTABLE)
708 requested = completed = 0;
709
710 spin_lock(&mirror->lock);
711 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
e76d28dd 712 requested, completed, ktime_get(), task->tk_start);
06946c6a 713 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 714 spin_unlock(&mirror->lock);
715}
716
76c66905
TM		 717static void
718ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
719{
720 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
721
722 if (devid)
723 nfs4_mark_deviceid_unavailable(devid);
724}
725
726static void
727ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
728{
729 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
730
731 if (devid)
732 nfs4_mark_deviceid_available(devid);
733}
734
d67ae825 735static struct nfs4_pnfs_ds *
76c66905
TM		 736ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
737 int start_idx, int *best_idx,
738 bool check_device)
d67ae825 739{
13544412 740 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
2444ff27 741 struct nfs4_ff_layout_mirror *mirror;
d67ae825 742 struct nfs4_pnfs_ds *ds;
3dc14735 743 bool fail_return = false;
d67ae825
TH		 744 int idx;
745
76c66905 746 /* mirrors are initially sorted by efficiency */
13544412 747 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
3dc14735 748 if (idx+1 == fls->mirror_array_cnt)
76c66905 749 fail_return = !check_device;
2444ff27
TM		 750
751 mirror = FF_LAYOUT_COMP(lseg, idx);
76c66905
TM		 752 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
753 if (!ds)
2444ff27
TM		 754 continue;
755
76c66905
TM		 756 if (check_device &&
757 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
758 continue;
759
760 *best_idx = idx;
761 return ds;
d67ae825
TH		 762 }
763
764 return NULL;
765}
766
76c66905
TM		 767static struct nfs4_pnfs_ds *
768ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
769 int start_idx, int *best_idx)
770{
771 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
772}
773
774static struct nfs4_pnfs_ds *
775ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
776 int start_idx, int *best_idx)
777{
778 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
779}
780
781static struct nfs4_pnfs_ds *
782ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
783 int start_idx, int *best_idx)
784{
785 struct nfs4_pnfs_ds *ds;
786
787 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
788 if (ds)
789 return ds;
790 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
791}
792
c7d73af2
TH		 793static void
794ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
795 struct nfs_page *req,
796 bool strict_iomode)
797{
c7d73af2
TH		 798 pnfs_put_lseg(pgio->pg_lseg);
799 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 800 nfs_req_openctx(req),
cbd7be43
TM		 801 req_offset(req),
802 req->wb_bytes,
c7d73af2
TH		 803 IOMODE_READ,
804 strict_iomode,
805 GFP_KERNEL);
806 if (IS_ERR(pgio->pg_lseg)) {
807 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
808 pgio->pg_lseg = NULL;
809 }
c7d73af2
TH		 810}
811
e1e54ab7
TM		 812static void
813ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
814 struct nfs_page *req)
815{
816 pnfs_generic_pg_check_layout(pgio);
817 pnfs_generic_pg_check_range(pgio, req);
818}
819
d67ae825
TH		 820static void
821ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
822 struct nfs_page *req)
823{
824 struct nfs_pgio_mirror *pgm;
825 struct nfs4_ff_layout_mirror *mirror;
826 struct nfs4_pnfs_ds *ds;
827 int ds_idx;
828
3dc14735 829retry:
e1e54ab7 830 ff_layout_pg_check_layout(pgio, req);
d67ae825 831 /* Use full layout for now */
1c1aeaf1 832 if (!pgio->pg_lseg) {
c7d73af2 833 ff_layout_pg_get_read(pgio, req, false);
1c1aeaf1
TM		 834 if (!pgio->pg_lseg)
835 goto out_nolseg;
836 }
837 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
c7d73af2 838 ff_layout_pg_get_read(pgio, req, true);
1c1aeaf1
TM		 839 if (!pgio->pg_lseg)
840 goto out_nolseg;
841 }
d67ae825 842
13544412 843 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
446ca219 844 if (!ds) {
3dc14735 845 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 846 goto out_mds;
3dc14735
TM		 847 pnfs_put_lseg(pgio->pg_lseg);
848 pgio->pg_lseg = NULL;
849 /* Sleep for 1 second before retrying */
850 ssleep(1);
851 goto retry;
446ca219
TH		 852 }
853
d67ae825
TH		 854 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
855
856 pgio->pg_mirror_idx = ds_idx;
857
858 /* read always uses only one mirror - idx 0 for pgio layer */
859 pgm = &pgio->pg_mirrors[0];
860 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
861
7af46292
TM		 862 if (NFS_SERVER(pgio->pg_inode)->flags &
863 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
864 pgio->pg_maxretrans = io_maxretrans;
d67ae825 865 return;
1c1aeaf1
TM		 866out_nolseg:
867 if (pgio->pg_error < 0)
868 return;
d67ae825 869out_mds:
d5b9216f
TM		 870 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
871 0, NFS4_MAX_UINT64, IOMODE_READ,
872 NFS_I(pgio->pg_inode)->layout,
873 pgio->pg_lseg);
d67ae825
TH		 874 pnfs_put_lseg(pgio->pg_lseg);
875 pgio->pg_lseg = NULL;
7af46292 876 pgio->pg_maxretrans = 0;
d67ae825
TH		 877 nfs_pageio_reset_read_mds(pgio);
878}
879
880static void
881ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
882 struct nfs_page *req)
883{
884 struct nfs4_ff_layout_mirror *mirror;
885 struct nfs_pgio_mirror *pgm;
d67ae825
TH		 886 struct nfs4_pnfs_ds *ds;
887 int i;
d67ae825 888
3dc14735 889retry:
e1e54ab7 890 ff_layout_pg_check_layout(pgio, req);
d600ad1f 891 if (!pgio->pg_lseg) {
d67ae825 892 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 893 nfs_req_openctx(req),
cbd7be43
TM		 894 req_offset(req),
895 req->wb_bytes,
d67ae825 896 IOMODE_RW,
c7d73af2 897 false,
d67ae825 898 GFP_NOFS);
d600ad1f
PT		 899 if (IS_ERR(pgio->pg_lseg)) {
900 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
901 pgio->pg_lseg = NULL;
902 return;
903 }
904 }
d67ae825
TH		 905 /* If no lseg, fall back to write through mds */
906 if (pgio->pg_lseg == NULL)
907 goto out_mds;
908
d67ae825 909 /* Use a direct mapping of ds_idx to pgio mirror_idx */
8b040137
TM		 910 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
911 goto out_eagain;
d67ae825
TH		 912
913 for (i = 0; i < pgio->pg_mirror_count; i++) {
2444ff27 914 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
2444ff27 915 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
446ca219 916 if (!ds) {
3dc14735 917 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 918 goto out_mds;
3dc14735
TM		 919 pnfs_put_lseg(pgio->pg_lseg);
920 pgio->pg_lseg = NULL;
921 /* Sleep for 1 second before retrying */
922 ssleep(1);
923 goto retry;
446ca219 924 }
d67ae825 925 pgm = &pgio->pg_mirrors[i];
d67ae825
TH		 926 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
927 }
928
7af46292
TM		 929 if (NFS_SERVER(pgio->pg_inode)->flags &
930 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
931 pgio->pg_maxretrans = io_maxretrans;
d67ae825 932 return;
8b040137
TM		 933out_eagain:
934 pnfs_generic_pg_cleanup(pgio);
935 pgio->pg_error = -EAGAIN;
936 return;
d67ae825 937out_mds:
d5b9216f
TM		 938 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
939 0, NFS4_MAX_UINT64, IOMODE_RW,
940 NFS_I(pgio->pg_inode)->layout,
941 pgio->pg_lseg);
d67ae825
TH		 942 pnfs_put_lseg(pgio->pg_lseg);
943 pgio->pg_lseg = NULL;
7af46292 944 pgio->pg_maxretrans = 0;
d67ae825 945 nfs_pageio_reset_write_mds(pgio);
8b040137 946 pgio->pg_error = -EAGAIN;
d67ae825
TH		 947}
948
949static unsigned int
950ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
951 struct nfs_page *req)
952{
d600ad1f 953 if (!pgio->pg_lseg) {
d67ae825 954 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 955 nfs_req_openctx(req),
d67ae825
TH		 956 0,
957 NFS4_MAX_UINT64,
958 IOMODE_RW,
c7d73af2 959 false,
d67ae825 960 GFP_NOFS);
d600ad1f
PT		 961 if (IS_ERR(pgio->pg_lseg)) {
962 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
963 pgio->pg_lseg = NULL;
964 goto out;
965 }
966 }
d67ae825
TH		 967 if (pgio->pg_lseg)
968 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
969
d5b9216f
TM		 970 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
971 0, NFS4_MAX_UINT64, IOMODE_RW,
972 NFS_I(pgio->pg_inode)->layout,
973 pgio->pg_lseg);
d67ae825 974 /* no lseg means that pnfs is not in use, so no mirroring here */
d67ae825 975 nfs_pageio_reset_write_mds(pgio);
d600ad1f 976out:
d67ae825
TH		 977 return 1;
978}
979
980static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
981 .pg_init = ff_layout_pg_init_read,
982 .pg_test = pnfs_generic_pg_test,
983 .pg_doio = pnfs_generic_pg_readpages,
984 .pg_cleanup = pnfs_generic_pg_cleanup,
985};
986
987static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
988 .pg_init = ff_layout_pg_init_write,
989 .pg_test = pnfs_generic_pg_test,
990 .pg_doio = pnfs_generic_pg_writepages,
991 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
992 .pg_cleanup = pnfs_generic_pg_cleanup,
993};
994
995static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
996{
997 struct rpc_task *task = &hdr->task;
998
999 pnfs_layoutcommit_inode(hdr->inode, false);
1000
1001 if (retry_pnfs) {
1002 dprintk("%s Reset task %5u for i/o through pNFS "
1003 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1004 hdr->task.tk_pid,
1005 hdr->inode->i_sb->s_id,
1006 (unsigned long long)NFS_FILEID(hdr->inode),
1007 hdr->args.count,
1008 (unsigned long long)hdr->args.offset);
1009
dc602dd7 1010 hdr->completion_ops->reschedule_io(hdr);
d67ae825
TH		 1011 return;
1012 }
1013
1014 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1015 dprintk("%s Reset task %5u for i/o through MDS "
1016 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1017 hdr->task.tk_pid,
1018 hdr->inode->i_sb->s_id,
1019 (unsigned long long)NFS_FILEID(hdr->inode),
1020 hdr->args.count,
1021 (unsigned long long)hdr->args.offset);
1022
d5b9216f
TM		 1023 trace_pnfs_mds_fallback_write_done(hdr->inode,
1024 hdr->args.offset, hdr->args.count,
1025 IOMODE_RW, NFS_I(hdr->inode)->layout,
1026 hdr->lseg);
d67ae825
TH		 1027 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1028 }
1029}
1030
1031static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1032{
1033 struct rpc_task *task = &hdr->task;
1034
1035 pnfs_layoutcommit_inode(hdr->inode, false);
1036
1037 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1038 dprintk("%s Reset task %5u for i/o through MDS "
1039 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1040 hdr->task.tk_pid,
1041 hdr->inode->i_sb->s_id,
1042 (unsigned long long)NFS_FILEID(hdr->inode),
1043 hdr->args.count,
1044 (unsigned long long)hdr->args.offset);
1045
d5b9216f
TM		 1046 trace_pnfs_mds_fallback_read_done(hdr->inode,
1047 hdr->args.offset, hdr->args.count,
1048 IOMODE_READ, NFS_I(hdr->inode)->layout,
1049 hdr->lseg);
d67ae825
TH		 1050 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1051 }
1052}
1053
1054static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1055 struct nfs4_state *state,
1056 struct nfs_client *clp,
1057 struct pnfs_layout_segment *lseg,
1058 int idx)
1059{
1060 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1061 struct inode *inode = lo->plh_inode;
d67ae825 1062 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
d67ae825
TH		 1063 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1064
d67ae825 1065 switch (task->tk_status) {
d67ae825
TH		 1066 case -NFS4ERR_BADSESSION:
1067 case -NFS4ERR_BADSLOT:
1068 case -NFS4ERR_BAD_HIGH_SLOT:
1069 case -NFS4ERR_DEADSESSION:
1070 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1071 case -NFS4ERR_SEQ_FALSE_RETRY:
1072 case -NFS4ERR_SEQ_MISORDERED:
1073 dprintk("%s ERROR %d, Reset session. Exchangeid "
1074 "flags 0x%x\n", __func__, task->tk_status,
1075 clp->cl_exchange_flags);
1076 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1077 break;
1078 case -NFS4ERR_DELAY:
1079 case -NFS4ERR_GRACE:
1080 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1081 break;
1082 case -NFS4ERR_RETRY_UNCACHED_REP:
1083 break;
1084 /* Invalidate Layout errors */
1085 case -NFS4ERR_PNFS_NO_LAYOUT:
1086 case -ESTALE: /* mapped NFS4ERR_STALE */
1087 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1088 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1089 case -NFS4ERR_FHEXPIRED:
1090 case -NFS4ERR_WRONG_TYPE:
1091 dprintk("%s Invalid layout error %d\n", __func__,
1092 task->tk_status);
1093 /*
1094 * Destroy layout so new i/o will get a new layout.
1095 * Layout will not be destroyed until all current lseg
1096 * references are put. Mark layout as invalid to resend failed
1097 * i/o and all i/o waiting on the slot table to the MDS until
1098 * layout is destroyed and a new valid layout is obtained.
1099 */
1100 pnfs_destroy_layout(NFS_I(inode));
1101 rpc_wake_up(&tbl->slot_tbl_waitq);
1102 goto reset;
1103 /* RPC connection errors */
1104 case -ECONNREFUSED:
1105 case -EHOSTDOWN:
1106 case -EHOSTUNREACH:
1107 case -ENETUNREACH:
1108 case -EIO:
1109 case -ETIMEDOUT:
1110 case -EPIPE:
1111 dprintk("%s DS connection error %d\n", __func__,
1112 task->tk_status);
1c48cee8
WAA		 1113 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1114 &devid->deviceid);
d67ae825
TH		 1115 rpc_wake_up(&tbl->slot_tbl_waitq);
1116 /* fall through */
1117 default:
3b13b4b3 1118 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1119 return -NFS4ERR_RESET_TO_PNFS;
1120reset:
1121 dprintk("%s Retry through MDS. Error %d\n", __func__,
1122 task->tk_status);
1123 return -NFS4ERR_RESET_TO_MDS;
1124 }
d67ae825
TH		 1125 task->tk_status = 0;
1126 return -EAGAIN;
d67ae825
TH		 1127}
1128
1129/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1130static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1131 struct pnfs_layout_segment *lseg,
1132 int idx)
1133{
1134 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1135
90816d1d
TM		 1136 switch (task->tk_status) {
1137 /* File access problems. Don't mark the device as unavailable */
1138 case -EACCES:
1139 case -ESTALE:
1140 case -EISDIR:
1141 case -EBADHANDLE:
1142 case -ELOOP:
1143 case -ENOSPC:
1144 break;
1145 case -EJUKEBOX:
1146 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1147 goto out_retry;
1148 default:
d67ae825
TH		 1149 dprintk("%s DS connection error %d\n", __func__,
1150 task->tk_status);
1c48cee8
WAA		 1151 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1152 &devid->deviceid);
d67ae825 1153 }
90816d1d
TM		 1154 /* FIXME: Need to prevent infinite looping here. */
1155 return -NFS4ERR_RESET_TO_PNFS;
1156out_retry:
d67ae825 1157 task->tk_status = 0;
7eeea167 1158 rpc_restart_call_prepare(task);
d67ae825
TH		 1159 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1160 return -EAGAIN;
1161}
1162
1163static int ff_layout_async_handle_error(struct rpc_task *task,
1164 struct nfs4_state *state,
1165 struct nfs_client *clp,
1166 struct pnfs_layout_segment *lseg,
1167 int idx)
1168{
1169 int vers = clp->cl_nfs_mod->rpc_vers->number;
1170
76c66905
TM		 1171 if (task->tk_status >= 0) {
1172 ff_layout_mark_ds_reachable(lseg, idx);
df3ab232 1173 return 0;
76c66905 1174 }
df3ab232
TM		 1175
1176 /* Handle the case of an invalid layout segment */
1177 if (!pnfs_is_valid_lseg(lseg))
1178 return -NFS4ERR_RESET_TO_PNFS;
1179
d67ae825
TH		 1180 switch (vers) {
1181 case 3:
1182 return ff_layout_async_handle_error_v3(task, lseg, idx);
1183 case 4:
1184 return ff_layout_async_handle_error_v4(task, state, clp,
1185 lseg, idx);
1186 default:
1187 /* should never happen */
1188 WARN_ON_ONCE(1);
1189 return 0;
1190 }
1191}
1192
1193static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1194 int idx, u64 offset, u64 length,
088f3e68 1195 u32 *op_status, int opnum, int error)
d67ae825
TH		 1196{
1197 struct nfs4_ff_layout_mirror *mirror;
088f3e68 1198 u32 status = *op_status;
d67ae825
TH		 1199 int err;
1200
dd52128a
TM		 1201 if (status == 0) {
1202 switch (error) {
1203 case -ETIMEDOUT:
1204 case -EPFNOSUPPORT:
1205 case -EPROTONOSUPPORT:
1206 case -EOPNOTSUPP:
1207 case -ECONNREFUSED:
1208 case -ECONNRESET:
1209 case -EHOSTDOWN:
1210 case -EHOSTUNREACH:
1211 case -ENETUNREACH:
1212 case -EADDRINUSE:
1213 case -ENOBUFS:
1214 case -EPIPE:
1215 case -EPERM:
088f3e68 1216 *op_status = status = NFS4ERR_NXIO;
dd52128a
TM		 1217 break;
1218 case -EACCES:
088f3e68 1219 *op_status = status = NFS4ERR_ACCESS;
dd52128a
TM		 1220 break;
1221 default:
1222 return;
1223 }
1224 }
1225
194a0dc8
TM		 1226 mirror = FF_LAYOUT_COMP(lseg, idx);
1227 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1228 mirror, offset, length, status, opnum,
1229 GFP_NOIO);
1230
c22eeb86
PT		 1231 switch (status) {
1232 case NFS4ERR_DELAY:
1233 case NFS4ERR_GRACE:
c22eeb86 1234 break;
194a0dc8
TM		 1235 case NFS4ERR_NXIO:
1236 ff_layout_mark_ds_unreachable(lseg, idx);
1237 /* Fallthrough */
1238 default:
1239 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1240 lseg);
c22eeb86
PT		 1241 }
1242
d67ae825
TH		 1243 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1244}
1245
1246/* NFS_PROTO call done callback routines */
d67ae825
TH		 1247static int ff_layout_read_done_cb(struct rpc_task *task,
1248 struct nfs_pgio_header *hdr)
1249{
166bd5b8 1250 int new_idx = hdr->pgio_mirror_idx;
d67ae825
TH		 1251 int err;
1252
088f3e68 1253 if (task->tk_status < 0) {
d67ae825
TH		 1254 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1255 hdr->args.offset, hdr->args.count,
088f3e68 1256 &hdr->res.op_status, OP_READ,
dd52128a 1257 task->tk_status);
088f3e68
TM		 1258 trace_ff_layout_read_error(hdr);
1259 }
1260
d67ae825
TH		 1261 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1262 hdr->ds_clp, hdr->lseg,
1263 hdr->pgio_mirror_idx);
1264
0722dc9f 1265 trace_nfs4_pnfs_read(hdr, err);
42f86b44
TM		 1266 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1267 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1268 switch (err) {
1269 case -NFS4ERR_RESET_TO_PNFS:
13544412
TM		 1270 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1271 hdr->pgio_mirror_idx + 1,
166bd5b8 1272 &new_idx))
f0922a6c 1273 goto out_layouterror;
42f86b44 1274 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
d67ae825
TH		 1275 return task->tk_status;
1276 case -NFS4ERR_RESET_TO_MDS:
42f86b44 1277 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1278 return task->tk_status;
1279 case -EAGAIN:
13544412 1280 goto out_eagain;
d67ae825
TH		 1281 }
1282
1283 return 0;
f0922a6c 1284out_layouterror:
166bd5b8 1285 ff_layout_read_record_layoutstats_done(task, hdr);
f0922a6c 1286 ff_layout_send_layouterror(hdr->lseg);
166bd5b8 1287 hdr->pgio_mirror_idx = new_idx;
13544412
TM		 1288out_eagain:
1289 rpc_restart_call_prepare(task);
1290 return -EAGAIN;
d67ae825
TH		 1291}
1292
c0f5f505
TM		 1293static bool
1294ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1295{
1296 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1297}
1298
d67ae825
TH		 1299/*
1300 * We reference the rpc_cred of the first WRITE that triggers the need for
1301 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1302 * rfc5661 is not clear about which credential should be used.
1303 *
1304 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1305 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1306 * we always send layoutcommit after DS writes.
1307 */
1308static void
bc28e1c2
TM		 1309ff_layout_set_layoutcommit(struct inode *inode,
1310 struct pnfs_layout_segment *lseg,
1311 loff_t end_offset)
d67ae825 1312{
bc28e1c2 1313 if (!ff_layout_need_layoutcommit(lseg))
c0f5f505
TM		 1314 return;
1315
bc28e1c2
TM		 1316 pnfs_set_layoutcommit(inode, lseg, end_offset);
1317 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1318 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
d67ae825
TH		 1319}
1320
37e9ed22
TM		 1321static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1322 struct nfs_pgio_header *hdr)
d67ae825 1323{
37e9ed22
TM		 1324 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1325 return;
0b7baf94 1326 nfs4_ff_layout_stat_io_start_read(hdr->inode,
abcb7bfc 1327 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1328 hdr->args.count,
1329 task->tk_start);
37e9ed22
TM		 1330}
1331
1332static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1333 struct nfs_pgio_header *hdr)
1334{
1335 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1336 return;
1337 nfs4_ff_layout_stat_io_end_read(task,
1338 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1339 hdr->args.count,
1340 hdr->res.count);
056f9ad6 1341 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1342}
abcb7bfc 1343
37e9ed22
TM		 1344static int ff_layout_read_prepare_common(struct rpc_task *task,
1345 struct nfs_pgio_header *hdr)
1346{
d67ae825
TH		 1347 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1348 rpc_exit(task, -EIO);
1349 return -EIO;
1350 }
d67ae825 1351
37e9ed22 1352 ff_layout_read_record_layoutstats_start(task, hdr);
d67ae825
TH		 1353 return 0;
1354}
1355
1356/*
1357 * Call ops for the async read/write cases
1358 * In the case of dense layouts, the offset needs to be reset to its
1359 * original value.
1360 */
1361static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1362{
1363 struct nfs_pgio_header *hdr = data;
1364
1365 if (ff_layout_read_prepare_common(task, hdr))
1366 return;
1367
1368 rpc_call_start(task);
1369}
1370
d67ae825
TH		 1371static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1372{
1373 struct nfs_pgio_header *hdr = data;
1374
6de7e12f
AS		 1375 if (nfs4_setup_sequence(hdr->ds_clp,
1376 &hdr->args.seq_args,
1377 &hdr->res.seq_res,
1378 task))
d67ae825
TH		 1379 return;
1380
bb21ce0a 1381 ff_layout_read_prepare_common(task, hdr);
d67ae825
TH		 1382}
1383
1384static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1385{
1386 struct nfs_pgio_header *hdr = data;
1387
1388 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1389
1390 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1391 task->tk_status == 0) {
1392 nfs4_sequence_done(task, &hdr->res.seq_res);
1393 return;
1394 }
1395
1396 /* Note this may cause RPC to be resent */
1397 hdr->mds_ops->rpc_call_done(task, hdr);
1398}
1399
1400static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1401{
1402 struct nfs_pgio_header *hdr = data;
1403
37e9ed22 1404 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1405 rpc_count_iostats_metrics(task,
1406 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1407}
1408
4d0ac221
TM		 1409static void ff_layout_read_release(void *data)
1410{
1411 struct nfs_pgio_header *hdr = data;
1412
1413 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1414 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1415 ff_layout_send_layouterror(hdr->lseg);
42f86b44 1416 pnfs_read_resend_pnfs(hdr);
f0922a6c 1417 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
42f86b44 1418 ff_layout_reset_read(hdr);
4d0ac221
TM		 1419 pnfs_generic_rw_release(data);
1420}
1421
1422
d67ae825
TH		 1423static int ff_layout_write_done_cb(struct rpc_task *task,
1424 struct nfs_pgio_header *hdr)
1425{
2e18d4d8 1426 loff_t end_offs = 0;
d67ae825
TH		 1427 int err;
1428
088f3e68 1429 if (task->tk_status < 0) {
d67ae825
TH		 1430 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1431 hdr->args.offset, hdr->args.count,
088f3e68 1432 &hdr->res.op_status, OP_WRITE,
dd52128a 1433 task->tk_status);
088f3e68
TM		 1434 trace_ff_layout_write_error(hdr);
1435 }
1436
d67ae825
TH		 1437 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1438 hdr->ds_clp, hdr->lseg,
1439 hdr->pgio_mirror_idx);
1440
0722dc9f 1441 trace_nfs4_pnfs_write(hdr, err);
7b0df92a
TM		 1442 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1443 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1444 switch (err) {
1445 case -NFS4ERR_RESET_TO_PNFS:
7b0df92a 1446 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
7cc8c5cd 1447 return task->tk_status;
d67ae825 1448 case -NFS4ERR_RESET_TO_MDS:
7b0df92a 1449 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1450 return task->tk_status;
1451 case -EAGAIN:
d67ae825
TH		 1452 return -EAGAIN;
1453 }
1454
1455 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1456 hdr->res.verf->committed == NFS_DATA_SYNC)
2e18d4d8
TM		 1457 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1458
1459 /* Note: if the write is unstable, don't set end_offs until commit */
1460 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
d67ae825 1461
54204010
PT		 1462 /* zero out fattr since we don't care DS attr at all */
1463 hdr->fattr.valid = 0;
69f230d9
PT		 1464 if (task->tk_status >= 0)
1465 nfs_writeback_update_inode(hdr);
1466
d67ae825
TH		 1467 return 0;
1468}
1469
1470static int ff_layout_commit_done_cb(struct rpc_task *task,
1471 struct nfs_commit_data *data)
1472{
d67ae825
TH		 1473 int err;
1474
088f3e68 1475 if (task->tk_status < 0) {
d67ae825
TH		 1476 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1477 data->args.offset, data->args.count,
088f3e68 1478 &data->res.op_status, OP_COMMIT,
dd52128a 1479 task->tk_status);
088f3e68
TM		 1480 trace_ff_layout_commit_error(data);
1481 }
1482
d67ae825
TH		 1483 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1484 data->lseg, data->ds_commit_index);
1485
0722dc9f 1486 trace_nfs4_pnfs_commit_ds(data, err);
d67ae825
TH		 1487 switch (err) {
1488 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1489 pnfs_generic_prepare_to_resend_writes(data);
1490 return -EAGAIN;
d67ae825 1491 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1492 pnfs_generic_prepare_to_resend_writes(data);
1493 return -EAGAIN;
1494 case -EAGAIN:
1495 rpc_restart_call_prepare(task);
1496 return -EAGAIN;
1497 }
1498
bc28e1c2 1499 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
d67ae825
TH		 1500
1501 return 0;
1502}
1503
37e9ed22
TM		 1504static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1505 struct nfs_pgio_header *hdr)
d67ae825 1506{
37e9ed22
TM		 1507 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1508 return;
0b7baf94 1509 nfs4_ff_layout_stat_io_start_write(hdr->inode,
abcb7bfc 1510 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1511 hdr->args.count,
1512 task->tk_start);
37e9ed22
TM		 1513}
1514
1515static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1516 struct nfs_pgio_header *hdr)
1517{
1518 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1519 return;
1520 nfs4_ff_layout_stat_io_end_write(task,
1521 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1522 hdr->args.count, hdr->res.count,
1523 hdr->res.verf->committed);
056f9ad6 1524 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1525}
abcb7bfc 1526
37e9ed22
TM		 1527static int ff_layout_write_prepare_common(struct rpc_task *task,
1528 struct nfs_pgio_header *hdr)
1529{
d67ae825
TH		 1530 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1531 rpc_exit(task, -EIO);
1532 return -EIO;
1533 }
1534
37e9ed22 1535 ff_layout_write_record_layoutstats_start(task, hdr);
d67ae825
TH		 1536 return 0;
1537}
1538
1539static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1540{
1541 struct nfs_pgio_header *hdr = data;
1542
1543 if (ff_layout_write_prepare_common(task, hdr))
1544 return;
1545
1546 rpc_call_start(task);
1547}
1548
1549static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1550{
1551 struct nfs_pgio_header *hdr = data;
1552
6de7e12f
AS		 1553 if (nfs4_setup_sequence(hdr->ds_clp,
1554 &hdr->args.seq_args,
1555 &hdr->res.seq_res,
1556 task))
d67ae825
TH		 1557 return;
1558
bb21ce0a 1559 ff_layout_write_prepare_common(task, hdr);
d67ae825
TH		 1560}
1561
1562static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1563{
1564 struct nfs_pgio_header *hdr = data;
1565
1566 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1567 task->tk_status == 0) {
1568 nfs4_sequence_done(task, &hdr->res.seq_res);
1569 return;
1570 }
1571
1572 /* Note this may cause RPC to be resent */
1573 hdr->mds_ops->rpc_call_done(task, hdr);
1574}
1575
1576static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1577{
1578 struct nfs_pgio_header *hdr = data;
1579
37e9ed22 1580 ff_layout_write_record_layoutstats_done(task, hdr);
d67ae825
TH		 1581 rpc_count_iostats_metrics(task,
1582 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1583}
1584
4d0ac221
TM		 1585static void ff_layout_write_release(void *data)
1586{
1587 struct nfs_pgio_header *hdr = data;
1588
1589 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1590 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1591 ff_layout_send_layouterror(hdr->lseg);
7b0df92a 1592 ff_layout_reset_write(hdr, true);
f0922a6c 1593 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
7b0df92a 1594 ff_layout_reset_write(hdr, false);
4d0ac221
TM		 1595 pnfs_generic_rw_release(data);
1596}
1597
37e9ed22 1598static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
abcb7bfc
TM		 1599 struct nfs_commit_data *cdata)
1600{
37e9ed22
TM		 1601 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1602 return;
0b7baf94 1603 nfs4_ff_layout_stat_io_start_write(cdata->inode,
abcb7bfc 1604 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
e76d28dd 1605 0, task->tk_start);
abcb7bfc
TM		 1606}
1607
37e9ed22
TM		 1608static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1609 struct nfs_commit_data *cdata)
1610{
1611 struct nfs_page *req;
1612 __u64 count = 0;
1613
1614 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1615 return;
1616
1617 if (task->tk_status == 0) {
1618 list_for_each_entry(req, &cdata->pages, wb_list)
1619 count += req->wb_bytes;
1620 }
1621 nfs4_ff_layout_stat_io_end_write(task,
1622 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1623 count, count, NFS_FILE_SYNC);
056f9ad6 1624 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
37e9ed22
TM		 1625}
1626
1627static void ff_layout_commit_prepare_common(struct rpc_task *task,
1628 struct nfs_commit_data *cdata)
1629{
1630 ff_layout_commit_record_layoutstats_start(task, cdata);
1631}
1632
d67ae825
TH		 1633static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1634{
abcb7bfc 1635 ff_layout_commit_prepare_common(task, data);
d67ae825
TH		 1636 rpc_call_start(task);
1637}
1638
1639static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1640{
1641 struct nfs_commit_data *wdata = data;
1642
6de7e12f
AS		 1643 if (nfs4_setup_sequence(wdata->ds_clp,
1644 &wdata->args.seq_args,
1645 &wdata->res.seq_res,
1646 task))
abcb7bfc
TM		 1647 return;
1648 ff_layout_commit_prepare_common(task, data);
1649}
1650
1651static void ff_layout_commit_done(struct rpc_task *task, void *data)
141b9b59
PT		 1652{
1653 pnfs_generic_write_commit_done(task, data);
1654}
1655
1656static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
abcb7bfc
TM		 1657{
1658 struct nfs_commit_data *cdata = data;
abcb7bfc 1659
37e9ed22 1660 ff_layout_commit_record_layoutstats_done(task, cdata);
d67ae825
TH		 1661 rpc_count_iostats_metrics(task,
1662 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1663}
1664
4d0ac221
TM		 1665static void ff_layout_commit_release(void *data)
1666{
1667 struct nfs_commit_data *cdata = data;
1668
1669 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1670 pnfs_generic_commit_release(data);
1671}
1672
d67ae825
TH		 1673static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1674 .rpc_call_prepare = ff_layout_read_prepare_v3,
1675 .rpc_call_done = ff_layout_read_call_done,
1676 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1677 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1678};
1679
1680static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1681 .rpc_call_prepare = ff_layout_read_prepare_v4,
1682 .rpc_call_done = ff_layout_read_call_done,
1683 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1684 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1685};
1686
1687static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1688 .rpc_call_prepare = ff_layout_write_prepare_v3,
1689 .rpc_call_done = ff_layout_write_call_done,
1690 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1691 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1692};
1693
1694static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1695 .rpc_call_prepare = ff_layout_write_prepare_v4,
1696 .rpc_call_done = ff_layout_write_call_done,
1697 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1698 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1699};
1700
1701static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1702 .rpc_call_prepare = ff_layout_commit_prepare_v3,
abcb7bfc 1703 .rpc_call_done = ff_layout_commit_done,
d67ae825 1704 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1705 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1706};
1707
1708static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1709 .rpc_call_prepare = ff_layout_commit_prepare_v4,
abcb7bfc 1710 .rpc_call_done = ff_layout_commit_done,
d67ae825 1711 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1712 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1713};
1714
1715static enum pnfs_try_status
1716ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1717{
1718 struct pnfs_layout_segment *lseg = hdr->lseg;
1719 struct nfs4_pnfs_ds *ds;
1720 struct rpc_clnt *ds_clnt;
2444ff27 1721 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1722 const struct cred *ds_cred;
d67ae825
TH		 1723 loff_t offset = hdr->args.offset;
1724 u32 idx = hdr->pgio_mirror_idx;
1725 int vers;
1726 struct nfs_fh *fh;
1727
5b5e0928 1728 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
d67ae825
TH		 1729 __func__, hdr->inode->i_ino,
1730 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1731
2444ff27
TM		 1732 mirror = FF_LAYOUT_COMP(lseg, idx);
1733 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
d67ae825
TH		 1734 if (!ds)
1735 goto out_failed;
1736
561d6f8a 1737 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1738 hdr->inode);
1739 if (IS_ERR(ds_clnt))
1740 goto out_failed;
1741
312cd4cb 1742 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1743 if (!ds_cred)
d67ae825
TH		 1744 goto out_failed;
1745
626d48b1 1746 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1747
1748 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
212bf41d 1749 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
d67ae825 1750
75bf47eb 1751 hdr->pgio_done_cb = ff_layout_read_done_cb;
212bf41d 1752 refcount_inc(&ds->ds_clp->cl_count);
d67ae825 1753 hdr->ds_clp = ds->ds_clp;
749da527 1754 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1755 if (fh)
1756 hdr->args.fh = fh;
bb21ce0a 1757
4cbc8a57 1758 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1759
d67ae825
TH		 1760 /*
1761 * Note that if we ever decide to split across DSes,
1762 * then we may need to handle dense-like offsets.
1763 */
1764 hdr->args.offset = offset;
1765 hdr->mds_offset = offset;
1766
1767 /* Perform an asynchronous read to ds */
1768 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1769 vers == 3 ? &ff_layout_read_call_ops_v3 :
1770 &ff_layout_read_call_ops_v4,
1771 0, RPC_TASK_SOFTCONN);
a52458b4 1772 put_cred(ds_cred);
d67ae825
TH		 1773 return PNFS_ATTEMPTED;
1774
1775out_failed:
3b13b4b3 1776 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825 1777 return PNFS_TRY_AGAIN;
d5b9216f
TM		 1778 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1779 hdr->args.offset, hdr->args.count,
1780 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
d67ae825
TH		 1781 return PNFS_NOT_ATTEMPTED;
1782}
1783
1784/* Perform async writes. */
1785static enum pnfs_try_status
1786ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1787{
1788 struct pnfs_layout_segment *lseg = hdr->lseg;
1789 struct nfs4_pnfs_ds *ds;
1790 struct rpc_clnt *ds_clnt;
2444ff27 1791 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1792 const struct cred *ds_cred;
d67ae825
TH		 1793 loff_t offset = hdr->args.offset;
1794 int vers;
1795 struct nfs_fh *fh;
1796 int idx = hdr->pgio_mirror_idx;
1797
2444ff27
TM		 1798 mirror = FF_LAYOUT_COMP(lseg, idx);
1799 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825 1800 if (!ds)
37f8aa16 1801 goto out_failed;
d67ae825 1802
561d6f8a 1803 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1804 hdr->inode);
1805 if (IS_ERR(ds_clnt))
37f8aa16 1806 goto out_failed;
d67ae825 1807
312cd4cb 1808 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1809 if (!ds_cred)
37f8aa16 1810 goto out_failed;
d67ae825 1811
626d48b1 1812 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825 1813
5b5e0928 1814 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
d67ae825 1815 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
212bf41d 1816 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1817 vers);
1818
1819 hdr->pgio_done_cb = ff_layout_write_done_cb;
212bf41d 1820 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1821 hdr->ds_clp = ds->ds_clp;
1822 hdr->ds_commit_idx = idx;
749da527 1823 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1824 if (fh)
1825 hdr->args.fh = fh;
1826
4cbc8a57 1827 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1828
d67ae825
TH		 1829 /*
1830 * Note that if we ever decide to split across DSes,
1831 * then we may need to handle dense-like offsets.
1832 */
1833 hdr->args.offset = offset;
1834
1835 /* Perform an asynchronous write */
1836 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1837 vers == 3 ? &ff_layout_write_call_ops_v3 :
1838 &ff_layout_write_call_ops_v4,
1839 sync, RPC_TASK_SOFTCONN);
a52458b4 1840 put_cred(ds_cred);
d67ae825 1841 return PNFS_ATTEMPTED;
37f8aa16
TM		 1842
1843out_failed:
1844 if (ff_layout_avoid_mds_available_ds(lseg))
1845 return PNFS_TRY_AGAIN;
d5b9216f
TM		 1846 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1847 hdr->args.offset, hdr->args.count,
1848 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
37f8aa16 1849 return PNFS_NOT_ATTEMPTED;
d67ae825
TH		 1850}
1851
d67ae825
TH		 1852static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1853{
1854 return i;
1855}
1856
1857static struct nfs_fh *
1858select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1859{
1860 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1861
1862 /* FIXME: Assume that there is only one NFS version available
1863 * for the DS.
1864 */
1865 return &flseg->mirror_array[i]->fh_versions[0];
1866}
1867
1868static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1869{
1870 struct pnfs_layout_segment *lseg = data->lseg;
1871 struct nfs4_pnfs_ds *ds;
1872 struct rpc_clnt *ds_clnt;
2444ff27 1873 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1874 const struct cred *ds_cred;
d67ae825 1875 u32 idx;
57f3f4c0 1876 int vers, ret;
d67ae825
TH		 1877 struct nfs_fh *fh;
1878
4b75053e
TM		 1879 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1880 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1881 goto out_err;
1882
d67ae825 1883 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
2444ff27
TM		 1884 mirror = FF_LAYOUT_COMP(lseg, idx);
1885 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825
TH		 1886 if (!ds)
1887 goto out_err;
1888
561d6f8a 1889 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1890 data->inode);
1891 if (IS_ERR(ds_clnt))
1892 goto out_err;
1893
312cd4cb 1894 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
57f3f4c0 1895 if (!ds_cred)
d67ae825
TH		 1896 goto out_err;
1897
626d48b1 1898 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1899
1900 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
212bf41d 1901 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1902 vers);
1903 data->commit_done_cb = ff_layout_commit_done_cb;
1904 data->cred = ds_cred;
212bf41d 1905 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1906 data->ds_clp = ds->ds_clp;
1907 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1908 if (fh)
1909 data->args.fh = fh;
abcb7bfc 1910
57f3f4c0 1911 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
d67ae825
TH		 1912 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1913 &ff_layout_commit_call_ops_v4,
1914 how, RPC_TASK_SOFTCONN);
a52458b4 1915 put_cred(ds_cred);
57f3f4c0 1916 return ret;
d67ae825
TH		 1917out_err:
1918 pnfs_generic_prepare_to_resend_writes(data);
1919 pnfs_generic_commit_release(data);
1920 return -EAGAIN;
1921}
1922
1923static int
1924ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1925 int how, struct nfs_commit_info *cinfo)
1926{
1927 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1928 ff_layout_initiate_commit);
1929}
1930
1931static struct pnfs_ds_commit_info *
1932ff_layout_get_ds_info(struct inode *inode)
1933{
1934 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1935
1936 if (layout == NULL)
1937 return NULL;
1938
1939 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1940}
1941
ba827c9a
TM		 1942static void
1943ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1944 struct pnfs_layout_segment *lseg)
1945{
1946 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1947 struct inode *inode = lseg->pls_layout->plh_inode;
1948 struct pnfs_commit_array *array, *new;
1949
1950 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1951 if (new) {
1952 spin_lock(&inode->i_lock);
1953 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1954 spin_unlock(&inode->i_lock);
1955 if (array != new)
1956 pnfs_free_commit_array(new);
1957 }
1958}
1959
a9901899
TM		 1960static void
1961ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1962 struct inode *inode)
1963{
1964 spin_lock(&inode->i_lock);
1965 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1966 spin_unlock(&inode->i_lock);
1967}
1968
d67ae825 1969static void
fc87701b 1970ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
d67ae825
TH		 1971{
1972 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1973 id_node));
1974}
1975
5ba6a09e 1976static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
5b9b3c85
TM		 1977 const struct nfs4_layoutreturn_args *args,
1978 const struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825 1979{
d67ae825 1980 __be32 *start;
d67ae825
TH		 1981
1982 start = xdr_reserve_space(xdr, 4);
1983 if (unlikely(!start))
1984 return -E2BIG;
1985
5b9b3c85 1986 *start = cpu_to_be32(ff_args->num_errors);
d67ae825 1987 /* This assume we always return _ALL_ layouts */
5b9b3c85 1988 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
d67ae825
TH		 1989}
1990
230bc962
TM		 1991static void
1992encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1993{
ab6e9aaf 1994 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
230bc962
TM		 1995}
1996
1997static void
1998ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1999 const nfs4_stateid *stateid,
2000 const struct nfs42_layoutstat_devinfo *devinfo)
2001{
2002 __be32 *p;
2003
2004 p = xdr_reserve_space(xdr, 8 + 8);
2005 p = xdr_encode_hyper(p, devinfo->offset);
2006 p = xdr_encode_hyper(p, devinfo->length);
2007 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2008 p = xdr_reserve_space(xdr, 4*8);
2009 p = xdr_encode_hyper(p, devinfo->read_count);
2010 p = xdr_encode_hyper(p, devinfo->read_bytes);
2011 p = xdr_encode_hyper(p, devinfo->write_count);
2012 p = xdr_encode_hyper(p, devinfo->write_bytes);
2013 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2014}
2015
2016static void
2017ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2018 const nfs4_stateid *stateid,
2019 const struct nfs42_layoutstat_devinfo *devinfo)
2020{
2021 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2022 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2023 devinfo->ld_private.data);
2024}
2025
d67ae825 2026/* report nothing for now */
230bc962
TM		 2027static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2028 const struct nfs4_layoutreturn_args *args,
2029 struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825
TH		 2030{
2031 __be32 *p;
230bc962 2032 int i;
d67ae825
TH		 2033
2034 p = xdr_reserve_space(xdr, 4);
230bc962
TM		 2035 *p = cpu_to_be32(ff_args->num_dev);
2036 for (i = 0; i < ff_args->num_dev; i++)
2037 ff_layout_encode_ff_iostat(xdr,
2038 &args->layout->plh_stateid,
2039 &ff_args->devinfo[i]);
2040}
2041
2042static void
2043ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2044 unsigned int num_entries)
2045{
2046 unsigned int i;
2047
2048 for (i = 0; i < num_entries; i++) {
2049 if (!devinfo[i].ld_private.ops)
2050 continue;
2051 if (!devinfo[i].ld_private.ops->free)
2052 continue;
2053 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2054 }
d67ae825
TH		 2055}
2056
2057static struct nfs4_deviceid_node *
2058ff_layout_alloc_deviceid_node(struct nfs_server *server,
2059 struct pnfs_device *pdev, gfp_t gfp_flags)
2060{
2061 struct nfs4_ff_layout_ds *dsaddr;
2062
2063 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2064 if (!dsaddr)
2065 return NULL;
2066 return &dsaddr->id_node;
2067}
2068
2069static void
94e5c571 2070ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
5b9b3c85
TM		 2071 const void *voidargs,
2072 const struct nfs4_xdr_opaque_data *ff_opaque)
d67ae825 2073{
5b9b3c85 2074 const struct nfs4_layoutreturn_args *args = voidargs;
5ba6a09e 2075 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
d9152114
TM		 2076 struct xdr_buf tmp_buf = {
2077 .head = {
2078 [0] = {
2079 .iov_base = page_address(ff_args->pages[0]),
2080 },
2081 },
2082 .buflen = PAGE_SIZE,
2083 };
2084 struct xdr_stream tmp_xdr;
d67ae825
TH		 2085 __be32 *start;
2086
2087 dprintk("%s: Begin\n", __func__);
d67ae825 2088
0ccc61b1 2089 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
d9152114
TM		 2090
2091 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2092 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2093
2094 start = xdr_reserve_space(xdr, 4);
2095 *start = cpu_to_be32(tmp_buf.len);
2096 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
082fa37d 2097
d67ae825
TH		 2098 dprintk("%s: Return\n", __func__);
2099}
2100
5b9b3c85
TM		 2101static void
2102ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2103{
2104 struct nfs4_flexfile_layoutreturn_args *ff_args;
2105
2106 if (!args->data)
2107 return;
2108 ff_args = args->data;
2109 args->data = NULL;
2110
2111 ff_layout_free_ds_ioerr(&ff_args->errors);
230bc962 2112 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
5b9b3c85 2113
d9152114 2114 put_page(ff_args->pages[0]);
5b9b3c85
TM		 2115 kfree(ff_args);
2116}
2117
2e54b9b1 2118static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
5b9b3c85
TM		 2119 .encode = ff_layout_encode_layoutreturn,
2120 .free = ff_layout_free_layoutreturn,
2121};
2122
2123static int
2124ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2125{
2126 struct nfs4_flexfile_layoutreturn_args *ff_args;
230bc962 2127 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
5b9b3c85
TM		 2128
2129 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2130 if (!ff_args)
d9152114
TM		 2131 goto out_nomem;
2132 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2133 if (!ff_args->pages[0])
2134 goto out_nomem_free;
5b9b3c85
TM		 2135
2136 INIT_LIST_HEAD(&ff_args->errors);
2137 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2138 &args->range, &ff_args->errors,
2139 FF_LAYOUTRETURN_MAXERR);
2140
230bc962
TM		 2141 spin_lock(&args->inode->i_lock);
2142 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2143 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2144 spin_unlock(&args->inode->i_lock);
2145
5b9b3c85
TM		 2146 args->ld_private->ops = &layoutreturn_ops;
2147 args->ld_private->data = ff_args;
2148 return 0;
d9152114
TM		 2149out_nomem_free:
2150 kfree(ff_args);
2151out_nomem:
2152 return -ENOMEM;
5b9b3c85
TM		 2153}
2154
f0922a6c
TM		 2155#ifdef CONFIG_NFS_V4_2
2156void
2157ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2158{
2159 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2160 struct nfs42_layout_error *errors;
2161 LIST_HEAD(head);
2162
2163 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2164 return;
2165 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2166 if (list_empty(&head))
2167 return;
2168
2169 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2170 sizeof(*errors), GFP_NOFS);
2171 if (errors != NULL) {
2172 const struct nfs4_ff_layout_ds_err *pos;
2173 size_t n = 0;
2174
2175 list_for_each_entry(pos, &head, list) {
2176 errors[n].offset = pos->offset;
2177 errors[n].length = pos->length;
2178 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2179 errors[n].errors[0].dev_id = pos->deviceid;
2180 errors[n].errors[0].status = pos->status;
2181 errors[n].errors[0].opnum = pos->opnum;
2182 n++;
2183 if (!list_is_last(&pos->list, &head) &&
2184 n < NFS42_LAYOUTERROR_MAX)
2185 continue;
2186 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2187 break;
2188 n = 0;
2189 }
2190 kfree(errors);
2191 }
2192 ff_layout_free_ds_ioerr(&head);
2193}
2194#else
2195void
2196ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2197{
2198}
2199#endif
2200
27c43064
PT		 2201static int
2202ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2203{
2204 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2205
2206 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2207}
2208
2209static size_t
2210ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2211 const int buflen)
2212{
2213 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2214 const struct in6_addr *addr = &sin6->sin6_addr;
2215
2216 /*
2217 * RFC 4291, Section 2.2.2
2218 *
2219 * Shorthanded ANY address
2220 */
2221 if (ipv6_addr_any(addr))
2222 return snprintf(buf, buflen, "::");
2223
2224 /*
2225 * RFC 4291, Section 2.2.2
2226 *
2227 * Shorthanded loopback address
2228 */
2229 if (ipv6_addr_loopback(addr))
2230 return snprintf(buf, buflen, "::1");
2231
2232 /*
2233 * RFC 4291, Section 2.2.3
2234 *
2235 * Special presentation address format for mapped v4
2236 * addresses.
2237 */
2238 if (ipv6_addr_v4mapped(addr))
2239 return snprintf(buf, buflen, "::ffff:%pI4",
2240 &addr->s6_addr32[3]);
2241
2242 /*
2243 * RFC 4291, Section 2.2.1
2244 */
2245 return snprintf(buf, buflen, "%pI6c", addr);
2246}
2247
2248/* Derived from rpc_sockaddr2uaddr */
2249static void
2250ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2251{
2252 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2253 char portbuf[RPCBIND_MAXUADDRPLEN];
2254 char addrbuf[RPCBIND_MAXUADDRLEN];
2255 char *netid;
2256 unsigned short port;
2257 int len, netid_len;
2258 __be32 *p;
2259
2260 switch (sap->sa_family) {
2261 case AF_INET:
2262 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2263 return;
2264 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2265 netid = "tcp";
2266 netid_len = 3;
2267 break;
2268 case AF_INET6:
2269 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2270 return;
2271 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2272 netid = "tcp6";
2273 netid_len = 4;
2274 break;
2275 default:
2276 /* we only support tcp and tcp6 */
2277 WARN_ON_ONCE(1);
2278 return;
2279 }
2280
2281 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2282 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2283
2284 p = xdr_reserve_space(xdr, 4 + netid_len);
2285 xdr_encode_opaque(p, netid, netid_len);
2286
2287 p = xdr_reserve_space(xdr, 4 + len);
2288 xdr_encode_opaque(p, addrbuf, len);
2289}
2290
2291static void
2292ff_layout_encode_nfstime(struct xdr_stream *xdr,
2293 ktime_t t)
2294{
2295 struct timespec64 ts;
2296 __be32 *p;
2297
2298 p = xdr_reserve_space(xdr, 12);
2299 ts = ktime_to_timespec64(t);
2300 p = xdr_encode_hyper(p, ts.tv_sec);
2301 *p++ = cpu_to_be32(ts.tv_nsec);
2302}
2303
2304static void
2305ff_layout_encode_io_latency(struct xdr_stream *xdr,
2306 struct nfs4_ff_io_stat *stat)
2307{
2308 __be32 *p;
2309
2310 p = xdr_reserve_space(xdr, 5 * 8);
2311 p = xdr_encode_hyper(p, stat->ops_requested);
2312 p = xdr_encode_hyper(p, stat->bytes_requested);
2313 p = xdr_encode_hyper(p, stat->ops_completed);
2314 p = xdr_encode_hyper(p, stat->bytes_completed);
2315 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2316 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2317 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2318}
2319
2320static void
422c93c8
TM		 2321ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2322 const struct nfs42_layoutstat_devinfo *devinfo,
2323 struct nfs4_ff_layout_mirror *mirror)
27c43064 2324{
27c43064
PT		 2325 struct nfs4_pnfs_ds_addr *da;
2326 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2327 struct nfs_fh *fh = &mirror->fh_versions[0];
422c93c8 2328 __be32 *p;
27c43064
PT		 2329
2330 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2331 dprintk("%s: DS %s: encoding address %s\n",
2332 __func__, ds->ds_remotestr, da->da_remotestr);
27c43064
PT		 2333 /* netaddr4 */
2334 ff_layout_encode_netaddr(xdr, da);
2335 /* nfs_fh4 */
2336 p = xdr_reserve_space(xdr, 4 + fh->size);
2337 xdr_encode_opaque(p, fh->data, fh->size);
2338 /* ff_io_latency4 read */
2339 spin_lock(&mirror->lock);
2340 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2341 /* ff_io_latency4 write */
2342 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2343 spin_unlock(&mirror->lock);
2344 /* nfstime4 */
2345 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2346 /* bool */
2347 p = xdr_reserve_space(xdr, 4);
2348 *p = cpu_to_be32(false);
422c93c8
TM		 2349}
2350
2351static void
2352ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2353 const struct nfs4_xdr_opaque_data *opaque)
2354{
2355 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2356 struct nfs42_layoutstat_devinfo, ld_private);
2357 __be32 *start;
2358
2359 /* layoutupdate length */
2360 start = xdr_reserve_space(xdr, 4);
2361 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
27c43064
PT		 2362
2363 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2364}
2365
422c93c8
TM		 2366static void
2367ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2368{
2369 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2370
2371 ff_layout_put_mirror(mirror);
2372}
2373
2374static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2375 .encode = ff_layout_encode_layoutstats,
2376 .free = ff_layout_free_layoutstats,
2377};
2378
266d12d4 2379static int
08e2e5bc
TM		 2380ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2381 struct nfs42_layoutstat_devinfo *devinfo,
266d12d4 2382 int dev_limit)
ad4dc53e 2383{
266d12d4 2384 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
ad4dc53e
PT		 2385 struct nfs4_ff_layout_mirror *mirror;
2386 struct nfs4_deviceid_node *dev;
266d12d4 2387 int i = 0;
ad4dc53e 2388
266d12d4
TM		 2389 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2390 if (i >= dev_limit)
ad4dc53e 2391 break;
65990d1a 2392 if (IS_ERR_OR_NULL(mirror->mirror_ds))
27c43064 2393 continue;
06946c6a
TM		 2394 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2395 continue;
266d12d4 2396 /* mirror refcount put in cleanup_layoutstats */
81a090b9 2397 if (!refcount_inc_not_zero(&mirror->ref))
266d12d4
TM		 2398 continue;
2399 dev = &mirror->mirror_ds->id_node;
ad4dc53e 2400 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
266d12d4
TM		 2401 devinfo->offset = 0;
2402 devinfo->length = NFS4_MAX_UINT64;
2f8220c1 2403 spin_lock(&mirror->lock);
d099d7b8 2404 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
ad4dc53e 2405 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
d099d7b8 2406 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
ad4dc53e 2407 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2f8220c1 2408 spin_unlock(&mirror->lock);
ad4dc53e 2409 devinfo->layout_type = LAYOUT_FLEX_FILES;
422c93c8
TM		 2410 devinfo->ld_private.ops = &layoutstat_ops;
2411 devinfo->ld_private.data = mirror;
ad4dc53e 2412
08e2e5bc 2413 devinfo++;
266d12d4 2414 i++;
ad4dc53e 2415 }
266d12d4 2416 return i;
ad4dc53e
PT		 2417}
2418
2419static int
2420ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2421{
266d12d4 2422 struct nfs4_flexfile_layout *ff_layout;
08e2e5bc 2423 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
ad4dc53e 2424
ad4dc53e 2425 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
266d12d4 2426 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
ad4dc53e
PT		 2427 if (!args->devinfo)
2428 return -ENOMEM;
2429
ad4dc53e 2430 spin_lock(&args->inode->i_lock);
08e2e5bc
TM		 2431 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2432 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2433 &args->devinfo[0], dev_count);
ad4dc53e 2434 spin_unlock(&args->inode->i_lock);
46c98c6d
TM		 2435 if (!args->num_dev) {
2436 kfree(args->devinfo);
2437 args->devinfo = NULL;
2438 return -ENOENT;
2439 }
ad4dc53e
PT		 2440
2441 return 0;
2442}
2443
76b2a303
TM		 2444static int
2445ff_layout_set_layoutdriver(struct nfs_server *server,
2446 const struct nfs_fh *dummy)
2447{
2448#if IS_ENABLED(CONFIG_NFS_V4_2)
2449 server->caps |= NFS_CAP_LAYOUTSTATS;
2450#endif
2451 return 0;
2452}
2453
9c455a8c
TM		 2454static const struct pnfs_commit_ops ff_layout_commit_ops = {
2455 .setup_ds_info = ff_layout_setup_ds_info,
2456 .release_ds_info = ff_layout_release_ds_info,
2457 .mark_request_commit = pnfs_layout_mark_request_commit,
2458 .clear_request_commit = pnfs_generic_clear_request_commit,
2459 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2460 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2461 .commit_pagelist = ff_layout_commit_pagelist,
2462};
2463
d67ae825
TH		 2464static struct pnfs_layoutdriver_type flexfilelayout_type = {
2465 .id = LAYOUT_FLEX_FILES,
2466 .name = "LAYOUT_FLEX_FILES",
2467 .owner = THIS_MODULE,
f86c3ac5 2468 .flags = PNFS_LAYOUTGET_ON_OPEN,
28ced9a8 2469 .max_layoutget_response = 4096, /* 1 page or so... */
76b2a303 2470 .set_layoutdriver = ff_layout_set_layoutdriver,
d67ae825
TH		 2471 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2472 .free_layout_hdr = ff_layout_free_layout_hdr,
2473 .alloc_lseg = ff_layout_alloc_lseg,
2474 .free_lseg = ff_layout_free_lseg,
0762ed2c 2475 .add_lseg = ff_layout_add_lseg,
d67ae825
TH		 2476 .pg_read_ops = &ff_layout_pg_read_ops,
2477 .pg_write_ops = &ff_layout_pg_write_ops,
2478 .get_ds_info = ff_layout_get_ds_info,
fc87701b 2479 .free_deviceid_node = ff_layout_free_deviceid_node,
d67ae825
TH		 2480 .read_pagelist = ff_layout_read_pagelist,
2481 .write_pagelist = ff_layout_write_pagelist,
2482 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
5b9b3c85 2483 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
5bb89b47 2484 .sync = pnfs_nfs_generic_sync,
ad4dc53e 2485 .prepare_layoutstats = ff_layout_prepare_layoutstats,
d67ae825
TH		 2486};
2487
2488static int __init nfs4flexfilelayout_init(void)
2489{
2490 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2491 __func__);
2492 return pnfs_register_layoutdriver(&flexfilelayout_type);
2493}
2494
2495static void __exit nfs4flexfilelayout_exit(void)
2496{
2497 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2498 __func__);
2499 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2500}
2501
2502MODULE_ALIAS("nfs-layouttype4-4");
2503
2504MODULE_LICENSE("GPL");
2505MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2506
2507module_init(nfs4flexfilelayout_init);
2508module_exit(nfs4flexfilelayout_exit);
33344e0f
TM		 2509
2510module_param(io_maxretrans, ushort, 0644);
2511MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2512 "retries an I/O request before returning an error. ");
Michael's Linux tree.