]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/nfs/flexfilelayout/flexfilelayout.c
projects / mirror_ubuntu-jammy-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
NFS/pNFS: Fix pnfs_layout_mark_request_commit() invalid layout segment handling
[mirror_ubuntu-jammy-kernel.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
260static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
261{
262 int ret = 0;
263
264 dprintk("--> %s\n", __func__);
265
266 /* FIXME: remove this check when layout segment support is added */
267 if (lgr->range.offset != 0 ||
268 lgr->range.length != NFS4_MAX_UINT64) {
269 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
270 __func__);
271 ret = -EINVAL;
272 }
273
274 dprintk("--> %s returns %d\n", __func__, ret);
275 return ret;
276}
277
278static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
279{
280 if (fls) {
281 ff_layout_free_mirror_array(fls);
282 kfree(fls);
283 }
284}
285
0762ed2c
TM		 286static bool
287ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
288 const struct pnfs_layout_range *l2)
289{
290 u64 end1, end2;
291
292 if (l1->iomode != l2->iomode)
293 return l1->iomode != IOMODE_READ;
294 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
295 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
296 if (end1 < l2->offset)
297 return false;
298 if (end2 < l1->offset)
299 return true;
300 return l2->offset <= l1->offset;
301}
302
303static bool
304ff_lseg_merge(struct pnfs_layout_segment *new,
305 struct pnfs_layout_segment *old)
306{
307 u64 new_end, old_end;
308
ee26bdd6
JL		 309 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
310 return false;
0762ed2c
TM		 311 if (new->pls_range.iomode != old->pls_range.iomode)
312 return false;
313 old_end = pnfs_calc_offset_end(old->pls_range.offset,
314 old->pls_range.length);
315 if (old_end < new->pls_range.offset)
316 return false;
317 new_end = pnfs_calc_offset_end(new->pls_range.offset,
318 new->pls_range.length);
319 if (new_end < old->pls_range.offset)
320 return false;
321
322 /* Mergeable: copy info from 'old' to 'new' */
323 if (new_end < old_end)
324 new_end = old_end;
325 if (new->pls_range.offset < old->pls_range.offset)
326 new->pls_range.offset = old->pls_range.offset;
327 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
328 new_end);
329 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
330 set_bit(NFS_LSEG_ROC, &new->pls_flags);
0762ed2c
TM		 331 return true;
332}
333
334static void
335ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
336 struct pnfs_layout_segment *lseg,
337 struct list_head *free_me)
338{
339 pnfs_generic_layout_insert_lseg(lo, lseg,
340 ff_lseg_range_is_after,
341 ff_lseg_merge,
342 free_me);
343}
344
d67ae825
TH		 345static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
346{
d67ae825
TH		 347 int i, j;
348
349 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
350 for (j = i + 1; j < fls->mirror_array_cnt; j++)
351 if (fls->mirror_array[i]->efficiency <
455b6ee6
FF		 352 fls->mirror_array[j]->efficiency)
353 swap(fls->mirror_array[i],
354 fls->mirror_array[j]);
d67ae825
TH		 355 }
356}
357
358static struct pnfs_layout_segment *
359ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
360 struct nfs4_layoutget_res *lgr,
361 gfp_t gfp_flags)
362{
363 struct pnfs_layout_segment *ret;
364 struct nfs4_ff_layout_segment *fls = NULL;
365 struct xdr_stream stream;
366 struct xdr_buf buf;
367 struct page *scratch;
368 u64 stripe_unit;
369 u32 mirror_array_cnt;
370 __be32 *p;
371 int i, rc;
372
373 dprintk("--> %s\n", __func__);
374 scratch = alloc_page(gfp_flags);
375 if (!scratch)
376 return ERR_PTR(-ENOMEM);
377
378 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
379 lgr->layoutp->len);
380 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
381
382 /* stripe unit and mirror_array_cnt */
383 rc = -EIO;
384 p = xdr_inline_decode(&stream, 8 + 4);
385 if (!p)
386 goto out_err_free;
387
388 p = xdr_decode_hyper(p, &stripe_unit);
389 mirror_array_cnt = be32_to_cpup(p++);
390 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
391 stripe_unit, mirror_array_cnt);
392
393 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
394 mirror_array_cnt == 0)
395 goto out_err_free;
396
397 rc = -ENOMEM;
329651b1
TM		 398 fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
399 gfp_flags);
d67ae825
TH		 400 if (!fls)
401 goto out_err_free;
402
403 fls->mirror_array_cnt = mirror_array_cnt;
404 fls->stripe_unit = stripe_unit;
d67ae825
TH		 405
406 for (i = 0; i < fls->mirror_array_cnt; i++) {
266d12d4 407 struct nfs4_ff_layout_mirror *mirror;
97f68c6b 408 struct cred *kcred;
762bb7e9 409 const struct cred __rcu *cred;
8276c902
N		 410 kuid_t uid;
411 kgid_t gid;
90a0be00 412 u32 ds_count, fh_count, id;
d67ae825
TH		 413 int j;
414
415 rc = -EIO;
416 p = xdr_inline_decode(&stream, 4);
417 if (!p)
418 goto out_err_free;
419 ds_count = be32_to_cpup(p);
420
421 /* FIXME: allow for striping? */
422 if (ds_count != 1)
423 goto out_err_free;
424
28a0d72c 425 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
d67ae825
TH		 426 if (fls->mirror_array[i] == NULL) {
427 rc = -ENOMEM;
428 goto out_err_free;
429 }
430
d67ae825
TH		 431 fls->mirror_array[i]->ds_count = ds_count;
432
433 /* deviceid */
65990d1a 434 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
d67ae825
TH		 435 if (rc)
436 goto out_err_free;
437
d67ae825
TH		 438 /* efficiency */
439 rc = -EIO;
440 p = xdr_inline_decode(&stream, 4);
441 if (!p)
442 goto out_err_free;
443 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
444
445 /* stateid */
93b717fd 446 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
d67ae825
TH		 447 if (rc)
448 goto out_err_free;
449
450 /* fh */
662f9a10 451 rc = -EIO;
d67ae825
TH		 452 p = xdr_inline_decode(&stream, 4);
453 if (!p)
454 goto out_err_free;
455 fh_count = be32_to_cpup(p);
456
457 fls->mirror_array[i]->fh_versions =
6396bb22 458 kcalloc(fh_count, sizeof(struct nfs_fh),
d67ae825
TH		 459 gfp_flags);
460 if (fls->mirror_array[i]->fh_versions == NULL) {
461 rc = -ENOMEM;
462 goto out_err_free;
463 }
464
465 for (j = 0; j < fh_count; j++) {
466 rc = decode_nfs_fh(&stream,
467 &fls->mirror_array[i]->fh_versions[j]);
468 if (rc)
469 goto out_err_free;
470 }
471
472 fls->mirror_array[i]->fh_versions_cnt = fh_count;
473
474 /* user */
90a0be00 475 rc = decode_name(&stream, &id);
d67ae825
TH		 476 if (rc)
477 goto out_err_free;
478
8276c902 479 uid = make_kuid(&init_user_ns, id);
90a0be00 480
d67ae825 481 /* group */
90a0be00 482 rc = decode_name(&stream, &id);
d67ae825
TH		 483 if (rc)
484 goto out_err_free;
485
8276c902 486 gid = make_kgid(&init_user_ns, id);
90a0be00 487
97f68c6b
N		 488 if (gfp_flags & __GFP_FS)
489 kcred = prepare_kernel_cred(NULL);
490 else {
491 unsigned int nofs_flags = memalloc_nofs_save();
492 kcred = prepare_kernel_cred(NULL);
493 memalloc_nofs_restore(nofs_flags);
494 }
495 rc = -ENOMEM;
496 if (!kcred)
497 goto out_err_free;
8276c902
N		 498 kcred->fsuid = uid;
499 kcred->fsgid = gid;
762bb7e9 500 cred = RCU_INITIALIZER(kcred);
90a0be00 501
3064b686
JL		 502 if (lgr->range.iomode == IOMODE_READ)
503 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
504 else
505 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
90a0be00 506
266d12d4
TM		 507 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
508 if (mirror != fls->mirror_array[i]) {
90a0be00 509 /* swap cred ptrs so free_mirror will clean up old */
3064b686
JL		 510 if (lgr->range.iomode == IOMODE_READ) {
511 cred = xchg(&mirror->ro_cred, cred);
512 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
513 } else {
514 cred = xchg(&mirror->rw_cred, cred);
515 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
516 }
266d12d4
TM		 517 ff_layout_free_mirror(fls->mirror_array[i]);
518 fls->mirror_array[i] = mirror;
519 }
520
3064b686
JL		 521 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
522 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
8276c902
N		 523 from_kuid(&init_user_ns, uid),
524 from_kgid(&init_user_ns, gid));
d67ae825
TH		 525 }
526
c0f5f505 527 p = xdr_inline_decode(&stream, 4);
d0379a5d
TM		 528 if (!p)
529 goto out_sort_mirrors;
530 fls->flags = be32_to_cpup(p);
531
532 p = xdr_inline_decode(&stream, 4);
533 if (!p)
534 goto out_sort_mirrors;
535 for (i=0; i < fls->mirror_array_cnt; i++)
536 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
c0f5f505 537
d0379a5d 538out_sort_mirrors:
d67ae825
TH		 539 ff_layout_sort_mirrors(fls);
540 rc = ff_layout_check_layout(lgr);
541 if (rc)
542 goto out_err_free;
d67ae825
TH		 543 ret = &fls->generic_hdr;
544 dprintk("<-- %s (success)\n", __func__);
545out_free_page:
546 __free_page(scratch);
547 return ret;
548out_err_free:
549 _ff_layout_free_lseg(fls);
550 ret = ERR_PTR(rc);
551 dprintk("<-- %s (%d)\n", __func__, rc);
552 goto out_free_page;
553}
554
d67ae825
TH		 555static void
556ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
557{
558 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825
TH		 559
560 dprintk("--> %s\n", __func__);
561
d67ae825
TH		 562 if (lseg->pls_range.iomode == IOMODE_RW) {
563 struct nfs4_flexfile_layout *ffl;
564 struct inode *inode;
565
566 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
567 inode = ffl->generic_hdr.plh_inode;
568 spin_lock(&inode->i_lock);
a9901899 569 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
d67ae825
TH		 570 spin_unlock(&inode->i_lock);
571 }
572 _ff_layout_free_lseg(fls);
573}
574
abcb7bfc 575static void
e76d28dd 576nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 577{
9bbd9bb4
PT		 578 /* first IO request? */
579 if (atomic_inc_return(&timer->n_ops) == 1) {
e76d28dd 580 timer->start_time = now;
abcb7bfc
TM		 581 }
582}
583
584static ktime_t
e76d28dd 585nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 586{
e76d28dd 587 ktime_t start;
abcb7bfc 588
9bbd9bb4
PT		 589 if (atomic_dec_return(&timer->n_ops) < 0)
590 WARN_ON_ONCE(1);
591
9bbd9bb4
PT		 592 start = timer->start_time;
593 timer->start_time = now;
abcb7bfc
TM		 594 return ktime_sub(now, start);
595}
596
97ba375b 597static bool
d983803d 598nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
e76d28dd
TM		 599 struct nfs4_ff_layoutstat *layoutstat,
600 ktime_t now)
abcb7bfc 601{
bbf58bf3 602 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
1c8d477a 603 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
d983803d 604
e76d28dd 605 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
1f3a8e49 606 if (!mirror->start_time)
9bbd9bb4 607 mirror->start_time = now;
d0379a5d
TM		 608 if (mirror->report_interval != 0)
609 report_interval = (s64)mirror->report_interval * 1000LL;
610 else if (layoutstats_timer != 0)
bbf58bf3 611 report_interval = (s64)layoutstats_timer * 1000LL;
1c8d477a 612 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
bbf58bf3 613 report_interval) {
1c8d477a 614 ffl->last_report_time = now;
97ba375b
PT		 615 return true;
616 }
617
618 return false;
abcb7bfc
TM		 619}
620
621static void
622nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
623 __u64 requested)
624{
625 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
626
627 iostat->ops_requested++;
628 iostat->bytes_requested += requested;
629}
630
631static void
632nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
633 __u64 requested,
634 __u64 completed,
e76d28dd
TM		 635 ktime_t time_completed,
636 ktime_t time_started)
abcb7bfc
TM		 637{
638 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
e76d28dd 639 ktime_t completion_time = ktime_sub(time_completed, time_started);
abcb7bfc
TM		 640 ktime_t timer;
641
642 iostat->ops_completed++;
643 iostat->bytes_completed += completed;
644 iostat->bytes_not_delivered += requested - completed;
645
e76d28dd 646 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
abcb7bfc
TM		 647 iostat->total_busy_time =
648 ktime_add(iostat->total_busy_time, timer);
649 iostat->aggregate_completion_time =
e76d28dd
TM		 650 ktime_add(iostat->aggregate_completion_time,
651 completion_time);
abcb7bfc
TM		 652}
653
654static void
0b7baf94
TM		 655nfs4_ff_layout_stat_io_start_read(struct inode *inode,
656 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 657 __u64 requested, ktime_t now)
abcb7bfc 658{
97ba375b
PT		 659 bool report;
660
abcb7bfc 661 spin_lock(&mirror->lock);
e76d28dd 662 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
abcb7bfc 663 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
06946c6a 664 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 665 spin_unlock(&mirror->lock);
97ba375b
PT		 666
667 if (report)
0b7baf94 668 pnfs_report_layoutstat(inode, GFP_KERNEL);
abcb7bfc
TM		 669}
670
671static void
672nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
673 struct nfs4_ff_layout_mirror *mirror,
674 __u64 requested,
675 __u64 completed)
676{
677 spin_lock(&mirror->lock);
678 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
679 requested, completed,
e76d28dd 680 ktime_get(), task->tk_start);
06946c6a 681 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 682 spin_unlock(&mirror->lock);
683}
684
685static void
0b7baf94
TM		 686nfs4_ff_layout_stat_io_start_write(struct inode *inode,
687 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 688 __u64 requested, ktime_t now)
abcb7bfc 689{
97ba375b
PT		 690 bool report;
691
abcb7bfc 692 spin_lock(&mirror->lock);
e76d28dd 693 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
abcb7bfc 694 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
06946c6a 695 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 696 spin_unlock(&mirror->lock);
97ba375b
PT		 697
698 if (report)
0b7baf94 699 pnfs_report_layoutstat(inode, GFP_NOIO);
abcb7bfc
TM		 700}
701
702static void
703nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
704 struct nfs4_ff_layout_mirror *mirror,
705 __u64 requested,
706 __u64 completed,
707 enum nfs3_stable_how committed)
708{
709 if (committed == NFS_UNSTABLE)
710 requested = completed = 0;
711
712 spin_lock(&mirror->lock);
713 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
e76d28dd 714 requested, completed, ktime_get(), task->tk_start);
06946c6a 715 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 716 spin_unlock(&mirror->lock);
717}
718
76c66905
TM		 719static void
720ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
721{
722 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
723
724 if (devid)
725 nfs4_mark_deviceid_unavailable(devid);
726}
727
728static void
729ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
730{
731 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
732
733 if (devid)
734 nfs4_mark_deviceid_available(devid);
735}
736
d67ae825 737static struct nfs4_pnfs_ds *
76c66905
TM		 738ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
739 int start_idx, int *best_idx,
740 bool check_device)
d67ae825 741{
13544412 742 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
2444ff27 743 struct nfs4_ff_layout_mirror *mirror;
d67ae825 744 struct nfs4_pnfs_ds *ds;
3dc14735 745 bool fail_return = false;
d67ae825
TH		 746 int idx;
747
76c66905 748 /* mirrors are initially sorted by efficiency */
13544412 749 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
3dc14735 750 if (idx+1 == fls->mirror_array_cnt)
76c66905 751 fail_return = !check_device;
2444ff27
TM		 752
753 mirror = FF_LAYOUT_COMP(lseg, idx);
76c66905
TM		 754 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
755 if (!ds)
2444ff27
TM		 756 continue;
757
76c66905
TM		 758 if (check_device &&
759 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
760 continue;
761
762 *best_idx = idx;
763 return ds;
d67ae825
TH		 764 }
765
766 return NULL;
767}
768
76c66905
TM		 769static struct nfs4_pnfs_ds *
770ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
771 int start_idx, int *best_idx)
772{
773 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
774}
775
776static struct nfs4_pnfs_ds *
777ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
778 int start_idx, int *best_idx)
779{
780 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
781}
782
783static struct nfs4_pnfs_ds *
784ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
785 int start_idx, int *best_idx)
786{
787 struct nfs4_pnfs_ds *ds;
788
789 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
790 if (ds)
791 return ds;
792 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
793}
794
c7d73af2
TH		 795static void
796ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
797 struct nfs_page *req,
798 bool strict_iomode)
799{
c7d73af2
TH		 800 pnfs_put_lseg(pgio->pg_lseg);
801 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 802 nfs_req_openctx(req),
c7d73af2
TH		 803 0,
804 NFS4_MAX_UINT64,
805 IOMODE_READ,
806 strict_iomode,
807 GFP_KERNEL);
808 if (IS_ERR(pgio->pg_lseg)) {
809 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
810 pgio->pg_lseg = NULL;
811 }
c7d73af2
TH		 812}
813
d67ae825
TH		 814static void
815ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
816 struct nfs_page *req)
817{
818 struct nfs_pgio_mirror *pgm;
819 struct nfs4_ff_layout_mirror *mirror;
820 struct nfs4_pnfs_ds *ds;
821 int ds_idx;
822
3dc14735 823retry:
b3230e80 824 pnfs_generic_pg_check_layout(pgio);
d67ae825 825 /* Use full layout for now */
1c1aeaf1 826 if (!pgio->pg_lseg) {
c7d73af2 827 ff_layout_pg_get_read(pgio, req, false);
1c1aeaf1
TM		 828 if (!pgio->pg_lseg)
829 goto out_nolseg;
830 }
831 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
c7d73af2 832 ff_layout_pg_get_read(pgio, req, true);
1c1aeaf1
TM		 833 if (!pgio->pg_lseg)
834 goto out_nolseg;
835 }
d67ae825 836
13544412 837 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
446ca219 838 if (!ds) {
3dc14735 839 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 840 goto out_mds;
3dc14735
TM		 841 pnfs_put_lseg(pgio->pg_lseg);
842 pgio->pg_lseg = NULL;
843 /* Sleep for 1 second before retrying */
844 ssleep(1);
845 goto retry;
446ca219
TH		 846 }
847
d67ae825
TH		 848 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
849
850 pgio->pg_mirror_idx = ds_idx;
851
852 /* read always uses only one mirror - idx 0 for pgio layer */
853 pgm = &pgio->pg_mirrors[0];
854 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
855
7af46292
TM		 856 if (NFS_SERVER(pgio->pg_inode)->flags &
857 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
858 pgio->pg_maxretrans = io_maxretrans;
d67ae825 859 return;
1c1aeaf1
TM		 860out_nolseg:
861 if (pgio->pg_error < 0)
862 return;
d67ae825 863out_mds:
d5b9216f
TM		 864 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
865 0, NFS4_MAX_UINT64, IOMODE_READ,
866 NFS_I(pgio->pg_inode)->layout,
867 pgio->pg_lseg);
d67ae825
TH		 868 pnfs_put_lseg(pgio->pg_lseg);
869 pgio->pg_lseg = NULL;
7af46292 870 pgio->pg_maxretrans = 0;
d67ae825
TH		 871 nfs_pageio_reset_read_mds(pgio);
872}
873
874static void
875ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
876 struct nfs_page *req)
877{
878 struct nfs4_ff_layout_mirror *mirror;
879 struct nfs_pgio_mirror *pgm;
d67ae825
TH		 880 struct nfs4_pnfs_ds *ds;
881 int i;
d67ae825 882
3dc14735 883retry:
b3230e80 884 pnfs_generic_pg_check_layout(pgio);
d600ad1f 885 if (!pgio->pg_lseg) {
d67ae825 886 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 887 nfs_req_openctx(req),
d67ae825
TH		 888 0,
889 NFS4_MAX_UINT64,
890 IOMODE_RW,
c7d73af2 891 false,
d67ae825 892 GFP_NOFS);
d600ad1f
PT		 893 if (IS_ERR(pgio->pg_lseg)) {
894 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
895 pgio->pg_lseg = NULL;
896 return;
897 }
898 }
d67ae825
TH		 899 /* If no lseg, fall back to write through mds */
900 if (pgio->pg_lseg == NULL)
901 goto out_mds;
902
d67ae825
TH		 903 /* Use a direct mapping of ds_idx to pgio mirror_idx */
904 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
905 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
906 goto out_mds;
907
908 for (i = 0; i < pgio->pg_mirror_count; i++) {
2444ff27 909 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
2444ff27 910 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
446ca219 911 if (!ds) {
3dc14735 912 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 913 goto out_mds;
3dc14735
TM		 914 pnfs_put_lseg(pgio->pg_lseg);
915 pgio->pg_lseg = NULL;
916 /* Sleep for 1 second before retrying */
917 ssleep(1);
918 goto retry;
446ca219 919 }
d67ae825 920 pgm = &pgio->pg_mirrors[i];
d67ae825
TH		 921 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
922 }
923
7af46292
TM		 924 if (NFS_SERVER(pgio->pg_inode)->flags &
925 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
926 pgio->pg_maxretrans = io_maxretrans;
d67ae825
TH		 927 return;
928
929out_mds:
d5b9216f
TM		 930 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
931 0, NFS4_MAX_UINT64, IOMODE_RW,
932 NFS_I(pgio->pg_inode)->layout,
933 pgio->pg_lseg);
d67ae825
TH		 934 pnfs_put_lseg(pgio->pg_lseg);
935 pgio->pg_lseg = NULL;
7af46292 936 pgio->pg_maxretrans = 0;
d67ae825
TH		 937 nfs_pageio_reset_write_mds(pgio);
938}
939
940static unsigned int
941ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
942 struct nfs_page *req)
943{
d600ad1f 944 if (!pgio->pg_lseg) {
d67ae825 945 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
9fcd5960 946 nfs_req_openctx(req),
d67ae825
TH		 947 0,
948 NFS4_MAX_UINT64,
949 IOMODE_RW,
c7d73af2 950 false,
d67ae825 951 GFP_NOFS);
d600ad1f
PT		 952 if (IS_ERR(pgio->pg_lseg)) {
953 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
954 pgio->pg_lseg = NULL;
955 goto out;
956 }
957 }
d67ae825
TH		 958 if (pgio->pg_lseg)
959 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
960
d5b9216f
TM		 961 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
962 0, NFS4_MAX_UINT64, IOMODE_RW,
963 NFS_I(pgio->pg_inode)->layout,
964 pgio->pg_lseg);
d67ae825 965 /* no lseg means that pnfs is not in use, so no mirroring here */
d67ae825 966 nfs_pageio_reset_write_mds(pgio);
d600ad1f 967out:
d67ae825
TH		 968 return 1;
969}
970
971static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
972 .pg_init = ff_layout_pg_init_read,
973 .pg_test = pnfs_generic_pg_test,
974 .pg_doio = pnfs_generic_pg_readpages,
975 .pg_cleanup = pnfs_generic_pg_cleanup,
976};
977
978static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
979 .pg_init = ff_layout_pg_init_write,
980 .pg_test = pnfs_generic_pg_test,
981 .pg_doio = pnfs_generic_pg_writepages,
982 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
983 .pg_cleanup = pnfs_generic_pg_cleanup,
984};
985
986static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
987{
988 struct rpc_task *task = &hdr->task;
989
990 pnfs_layoutcommit_inode(hdr->inode, false);
991
992 if (retry_pnfs) {
993 dprintk("%s Reset task %5u for i/o through pNFS "
994 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
995 hdr->task.tk_pid,
996 hdr->inode->i_sb->s_id,
997 (unsigned long long)NFS_FILEID(hdr->inode),
998 hdr->args.count,
999 (unsigned long long)hdr->args.offset);
1000
dc602dd7 1001 hdr->completion_ops->reschedule_io(hdr);
d67ae825
TH		 1002 return;
1003 }
1004
1005 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1006 dprintk("%s Reset task %5u for i/o through MDS "
1007 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1008 hdr->task.tk_pid,
1009 hdr->inode->i_sb->s_id,
1010 (unsigned long long)NFS_FILEID(hdr->inode),
1011 hdr->args.count,
1012 (unsigned long long)hdr->args.offset);
1013
d5b9216f
TM		 1014 trace_pnfs_mds_fallback_write_done(hdr->inode,
1015 hdr->args.offset, hdr->args.count,
1016 IOMODE_RW, NFS_I(hdr->inode)->layout,
1017 hdr->lseg);
d67ae825
TH		 1018 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1019 }
1020}
1021
1022static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1023{
1024 struct rpc_task *task = &hdr->task;
1025
1026 pnfs_layoutcommit_inode(hdr->inode, false);
1027
1028 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1029 dprintk("%s Reset task %5u for i/o through MDS "
1030 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1031 hdr->task.tk_pid,
1032 hdr->inode->i_sb->s_id,
1033 (unsigned long long)NFS_FILEID(hdr->inode),
1034 hdr->args.count,
1035 (unsigned long long)hdr->args.offset);
1036
d5b9216f
TM		 1037 trace_pnfs_mds_fallback_read_done(hdr->inode,
1038 hdr->args.offset, hdr->args.count,
1039 IOMODE_READ, NFS_I(hdr->inode)->layout,
1040 hdr->lseg);
d67ae825
TH		 1041 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1042 }
1043}
1044
1045static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1046 struct nfs4_state *state,
1047 struct nfs_client *clp,
1048 struct pnfs_layout_segment *lseg,
1049 int idx)
1050{
1051 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1052 struct inode *inode = lo->plh_inode;
d67ae825 1053 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
d67ae825
TH		 1054 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1055
d67ae825 1056 switch (task->tk_status) {
d67ae825
TH		 1057 case -NFS4ERR_BADSESSION:
1058 case -NFS4ERR_BADSLOT:
1059 case -NFS4ERR_BAD_HIGH_SLOT:
1060 case -NFS4ERR_DEADSESSION:
1061 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1062 case -NFS4ERR_SEQ_FALSE_RETRY:
1063 case -NFS4ERR_SEQ_MISORDERED:
1064 dprintk("%s ERROR %d, Reset session. Exchangeid "
1065 "flags 0x%x\n", __func__, task->tk_status,
1066 clp->cl_exchange_flags);
1067 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1068 break;
1069 case -NFS4ERR_DELAY:
1070 case -NFS4ERR_GRACE:
1071 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1072 break;
1073 case -NFS4ERR_RETRY_UNCACHED_REP:
1074 break;
1075 /* Invalidate Layout errors */
1076 case -NFS4ERR_PNFS_NO_LAYOUT:
1077 case -ESTALE: /* mapped NFS4ERR_STALE */
1078 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1079 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1080 case -NFS4ERR_FHEXPIRED:
1081 case -NFS4ERR_WRONG_TYPE:
1082 dprintk("%s Invalid layout error %d\n", __func__,
1083 task->tk_status);
1084 /*
1085 * Destroy layout so new i/o will get a new layout.
1086 * Layout will not be destroyed until all current lseg
1087 * references are put. Mark layout as invalid to resend failed
1088 * i/o and all i/o waiting on the slot table to the MDS until
1089 * layout is destroyed and a new valid layout is obtained.
1090 */
1091 pnfs_destroy_layout(NFS_I(inode));
1092 rpc_wake_up(&tbl->slot_tbl_waitq);
1093 goto reset;
1094 /* RPC connection errors */
1095 case -ECONNREFUSED:
1096 case -EHOSTDOWN:
1097 case -EHOSTUNREACH:
1098 case -ENETUNREACH:
1099 case -EIO:
1100 case -ETIMEDOUT:
1101 case -EPIPE:
1102 dprintk("%s DS connection error %d\n", __func__,
1103 task->tk_status);
1c48cee8
WAA		 1104 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1105 &devid->deviceid);
d67ae825
TH		 1106 rpc_wake_up(&tbl->slot_tbl_waitq);
1107 /* fall through */
1108 default:
3b13b4b3 1109 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1110 return -NFS4ERR_RESET_TO_PNFS;
1111reset:
1112 dprintk("%s Retry through MDS. Error %d\n", __func__,
1113 task->tk_status);
1114 return -NFS4ERR_RESET_TO_MDS;
1115 }
d67ae825
TH		 1116 task->tk_status = 0;
1117 return -EAGAIN;
d67ae825
TH		 1118}
1119
1120/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1121static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1122 struct pnfs_layout_segment *lseg,
1123 int idx)
1124{
1125 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1126
90816d1d
TM		 1127 switch (task->tk_status) {
1128 /* File access problems. Don't mark the device as unavailable */
1129 case -EACCES:
1130 case -ESTALE:
1131 case -EISDIR:
1132 case -EBADHANDLE:
1133 case -ELOOP:
1134 case -ENOSPC:
1135 break;
1136 case -EJUKEBOX:
1137 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1138 goto out_retry;
1139 default:
d67ae825
TH		 1140 dprintk("%s DS connection error %d\n", __func__,
1141 task->tk_status);
1c48cee8
WAA		 1142 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1143 &devid->deviceid);
d67ae825 1144 }
90816d1d
TM		 1145 /* FIXME: Need to prevent infinite looping here. */
1146 return -NFS4ERR_RESET_TO_PNFS;
1147out_retry:
d67ae825 1148 task->tk_status = 0;
7eeea167 1149 rpc_restart_call_prepare(task);
d67ae825
TH		 1150 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1151 return -EAGAIN;
1152}
1153
1154static int ff_layout_async_handle_error(struct rpc_task *task,
1155 struct nfs4_state *state,
1156 struct nfs_client *clp,
1157 struct pnfs_layout_segment *lseg,
1158 int idx)
1159{
1160 int vers = clp->cl_nfs_mod->rpc_vers->number;
1161
76c66905
TM		 1162 if (task->tk_status >= 0) {
1163 ff_layout_mark_ds_reachable(lseg, idx);
df3ab232 1164 return 0;
76c66905 1165 }
df3ab232
TM		 1166
1167 /* Handle the case of an invalid layout segment */
1168 if (!pnfs_is_valid_lseg(lseg))
1169 return -NFS4ERR_RESET_TO_PNFS;
1170
d67ae825
TH		 1171 switch (vers) {
1172 case 3:
1173 return ff_layout_async_handle_error_v3(task, lseg, idx);
1174 case 4:
1175 return ff_layout_async_handle_error_v4(task, state, clp,
1176 lseg, idx);
1177 default:
1178 /* should never happen */
1179 WARN_ON_ONCE(1);
1180 return 0;
1181 }
1182}
1183
1184static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1185 int idx, u64 offset, u64 length,
088f3e68 1186 u32 *op_status, int opnum, int error)
d67ae825
TH		 1187{
1188 struct nfs4_ff_layout_mirror *mirror;
088f3e68 1189 u32 status = *op_status;
d67ae825
TH		 1190 int err;
1191
dd52128a
TM		 1192 if (status == 0) {
1193 switch (error) {
1194 case -ETIMEDOUT:
1195 case -EPFNOSUPPORT:
1196 case -EPROTONOSUPPORT:
1197 case -EOPNOTSUPP:
1198 case -ECONNREFUSED:
1199 case -ECONNRESET:
1200 case -EHOSTDOWN:
1201 case -EHOSTUNREACH:
1202 case -ENETUNREACH:
1203 case -EADDRINUSE:
1204 case -ENOBUFS:
1205 case -EPIPE:
1206 case -EPERM:
088f3e68 1207 *op_status = status = NFS4ERR_NXIO;
dd52128a
TM		 1208 break;
1209 case -EACCES:
088f3e68 1210 *op_status = status = NFS4ERR_ACCESS;
dd52128a
TM		 1211 break;
1212 default:
1213 return;
1214 }
1215 }
1216
194a0dc8
TM		 1217 mirror = FF_LAYOUT_COMP(lseg, idx);
1218 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1219 mirror, offset, length, status, opnum,
1220 GFP_NOIO);
1221
c22eeb86
PT		 1222 switch (status) {
1223 case NFS4ERR_DELAY:
1224 case NFS4ERR_GRACE:
c22eeb86 1225 break;
194a0dc8
TM		 1226 case NFS4ERR_NXIO:
1227 ff_layout_mark_ds_unreachable(lseg, idx);
1228 /* Fallthrough */
1229 default:
1230 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1231 lseg);
c22eeb86
PT		 1232 }
1233
d67ae825
TH		 1234 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1235}
1236
1237/* NFS_PROTO call done callback routines */
d67ae825
TH		 1238static int ff_layout_read_done_cb(struct rpc_task *task,
1239 struct nfs_pgio_header *hdr)
1240{
166bd5b8 1241 int new_idx = hdr->pgio_mirror_idx;
d67ae825
TH		 1242 int err;
1243
088f3e68 1244 if (task->tk_status < 0) {
d67ae825
TH		 1245 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1246 hdr->args.offset, hdr->args.count,
088f3e68 1247 &hdr->res.op_status, OP_READ,
dd52128a 1248 task->tk_status);
088f3e68
TM		 1249 trace_ff_layout_read_error(hdr);
1250 }
1251
d67ae825
TH		 1252 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1253 hdr->ds_clp, hdr->lseg,
1254 hdr->pgio_mirror_idx);
1255
0722dc9f 1256 trace_nfs4_pnfs_read(hdr, err);
42f86b44
TM		 1257 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1258 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1259 switch (err) {
1260 case -NFS4ERR_RESET_TO_PNFS:
13544412
TM		 1261 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1262 hdr->pgio_mirror_idx + 1,
166bd5b8 1263 &new_idx))
f0922a6c 1264 goto out_layouterror;
42f86b44 1265 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
d67ae825
TH		 1266 return task->tk_status;
1267 case -NFS4ERR_RESET_TO_MDS:
42f86b44 1268 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1269 return task->tk_status;
1270 case -EAGAIN:
13544412 1271 goto out_eagain;
d67ae825
TH		 1272 }
1273
1274 return 0;
f0922a6c 1275out_layouterror:
166bd5b8 1276 ff_layout_read_record_layoutstats_done(task, hdr);
f0922a6c 1277 ff_layout_send_layouterror(hdr->lseg);
166bd5b8 1278 hdr->pgio_mirror_idx = new_idx;
13544412
TM		 1279out_eagain:
1280 rpc_restart_call_prepare(task);
1281 return -EAGAIN;
d67ae825
TH		 1282}
1283
c0f5f505
TM		 1284static bool
1285ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1286{
1287 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1288}
1289
d67ae825
TH		 1290/*
1291 * We reference the rpc_cred of the first WRITE that triggers the need for
1292 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1293 * rfc5661 is not clear about which credential should be used.
1294 *
1295 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1296 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1297 * we always send layoutcommit after DS writes.
1298 */
1299static void
bc28e1c2
TM		 1300ff_layout_set_layoutcommit(struct inode *inode,
1301 struct pnfs_layout_segment *lseg,
1302 loff_t end_offset)
d67ae825 1303{
bc28e1c2 1304 if (!ff_layout_need_layoutcommit(lseg))
c0f5f505
TM		 1305 return;
1306
bc28e1c2
TM		 1307 pnfs_set_layoutcommit(inode, lseg, end_offset);
1308 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1309 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
d67ae825
TH		 1310}
1311
37e9ed22
TM		 1312static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1313 struct nfs_pgio_header *hdr)
d67ae825 1314{
37e9ed22
TM		 1315 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1316 return;
0b7baf94 1317 nfs4_ff_layout_stat_io_start_read(hdr->inode,
abcb7bfc 1318 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1319 hdr->args.count,
1320 task->tk_start);
37e9ed22
TM		 1321}
1322
1323static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1324 struct nfs_pgio_header *hdr)
1325{
1326 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1327 return;
1328 nfs4_ff_layout_stat_io_end_read(task,
1329 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1330 hdr->args.count,
1331 hdr->res.count);
056f9ad6 1332 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1333}
abcb7bfc 1334
37e9ed22
TM		 1335static int ff_layout_read_prepare_common(struct rpc_task *task,
1336 struct nfs_pgio_header *hdr)
1337{
d67ae825
TH		 1338 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1339 rpc_exit(task, -EIO);
1340 return -EIO;
1341 }
d67ae825 1342
37e9ed22 1343 ff_layout_read_record_layoutstats_start(task, hdr);
d67ae825
TH		 1344 return 0;
1345}
1346
1347/*
1348 * Call ops for the async read/write cases
1349 * In the case of dense layouts, the offset needs to be reset to its
1350 * original value.
1351 */
1352static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1353{
1354 struct nfs_pgio_header *hdr = data;
1355
1356 if (ff_layout_read_prepare_common(task, hdr))
1357 return;
1358
1359 rpc_call_start(task);
1360}
1361
d67ae825
TH		 1362static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1363{
1364 struct nfs_pgio_header *hdr = data;
1365
6de7e12f
AS		 1366 if (nfs4_setup_sequence(hdr->ds_clp,
1367 &hdr->args.seq_args,
1368 &hdr->res.seq_res,
1369 task))
d67ae825
TH		 1370 return;
1371
bb21ce0a 1372 ff_layout_read_prepare_common(task, hdr);
d67ae825
TH		 1373}
1374
1375static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1376{
1377 struct nfs_pgio_header *hdr = data;
1378
1379 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1380
1381 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1382 task->tk_status == 0) {
1383 nfs4_sequence_done(task, &hdr->res.seq_res);
1384 return;
1385 }
1386
1387 /* Note this may cause RPC to be resent */
1388 hdr->mds_ops->rpc_call_done(task, hdr);
1389}
1390
1391static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1392{
1393 struct nfs_pgio_header *hdr = data;
1394
37e9ed22 1395 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1396 rpc_count_iostats_metrics(task,
1397 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1398}
1399
4d0ac221
TM		 1400static void ff_layout_read_release(void *data)
1401{
1402 struct nfs_pgio_header *hdr = data;
1403
1404 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1405 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1406 ff_layout_send_layouterror(hdr->lseg);
42f86b44 1407 pnfs_read_resend_pnfs(hdr);
f0922a6c 1408 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
42f86b44 1409 ff_layout_reset_read(hdr);
4d0ac221
TM		 1410 pnfs_generic_rw_release(data);
1411}
1412
1413
d67ae825
TH		 1414static int ff_layout_write_done_cb(struct rpc_task *task,
1415 struct nfs_pgio_header *hdr)
1416{
2e18d4d8 1417 loff_t end_offs = 0;
d67ae825
TH		 1418 int err;
1419
088f3e68 1420 if (task->tk_status < 0) {
d67ae825
TH		 1421 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1422 hdr->args.offset, hdr->args.count,
088f3e68 1423 &hdr->res.op_status, OP_WRITE,
dd52128a 1424 task->tk_status);
088f3e68
TM		 1425 trace_ff_layout_write_error(hdr);
1426 }
1427
d67ae825
TH		 1428 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1429 hdr->ds_clp, hdr->lseg,
1430 hdr->pgio_mirror_idx);
1431
0722dc9f 1432 trace_nfs4_pnfs_write(hdr, err);
7b0df92a
TM		 1433 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1434 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1435 switch (err) {
1436 case -NFS4ERR_RESET_TO_PNFS:
7b0df92a 1437 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
7cc8c5cd 1438 return task->tk_status;
d67ae825 1439 case -NFS4ERR_RESET_TO_MDS:
7b0df92a 1440 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1441 return task->tk_status;
1442 case -EAGAIN:
d67ae825
TH		 1443 return -EAGAIN;
1444 }
1445
1446 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1447 hdr->res.verf->committed == NFS_DATA_SYNC)
2e18d4d8
TM		 1448 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1449
1450 /* Note: if the write is unstable, don't set end_offs until commit */
1451 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
d67ae825 1452
54204010
PT		 1453 /* zero out fattr since we don't care DS attr at all */
1454 hdr->fattr.valid = 0;
69f230d9
PT		 1455 if (task->tk_status >= 0)
1456 nfs_writeback_update_inode(hdr);
1457
d67ae825
TH		 1458 return 0;
1459}
1460
1461static int ff_layout_commit_done_cb(struct rpc_task *task,
1462 struct nfs_commit_data *data)
1463{
d67ae825
TH		 1464 int err;
1465
088f3e68 1466 if (task->tk_status < 0) {
d67ae825
TH		 1467 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1468 data->args.offset, data->args.count,
088f3e68 1469 &data->res.op_status, OP_COMMIT,
dd52128a 1470 task->tk_status);
088f3e68
TM		 1471 trace_ff_layout_commit_error(data);
1472 }
1473
d67ae825
TH		 1474 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1475 data->lseg, data->ds_commit_index);
1476
0722dc9f 1477 trace_nfs4_pnfs_commit_ds(data, err);
d67ae825
TH		 1478 switch (err) {
1479 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1480 pnfs_generic_prepare_to_resend_writes(data);
1481 return -EAGAIN;
d67ae825 1482 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1483 pnfs_generic_prepare_to_resend_writes(data);
1484 return -EAGAIN;
1485 case -EAGAIN:
1486 rpc_restart_call_prepare(task);
1487 return -EAGAIN;
1488 }
1489
bc28e1c2 1490 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
d67ae825
TH		 1491
1492 return 0;
1493}
1494
37e9ed22
TM		 1495static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1496 struct nfs_pgio_header *hdr)
d67ae825 1497{
37e9ed22
TM		 1498 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1499 return;
0b7baf94 1500 nfs4_ff_layout_stat_io_start_write(hdr->inode,
abcb7bfc 1501 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1502 hdr->args.count,
1503 task->tk_start);
37e9ed22
TM		 1504}
1505
1506static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1507 struct nfs_pgio_header *hdr)
1508{
1509 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1510 return;
1511 nfs4_ff_layout_stat_io_end_write(task,
1512 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1513 hdr->args.count, hdr->res.count,
1514 hdr->res.verf->committed);
056f9ad6 1515 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1516}
abcb7bfc 1517
37e9ed22
TM		 1518static int ff_layout_write_prepare_common(struct rpc_task *task,
1519 struct nfs_pgio_header *hdr)
1520{
d67ae825
TH		 1521 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1522 rpc_exit(task, -EIO);
1523 return -EIO;
1524 }
1525
37e9ed22 1526 ff_layout_write_record_layoutstats_start(task, hdr);
d67ae825
TH		 1527 return 0;
1528}
1529
1530static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1531{
1532 struct nfs_pgio_header *hdr = data;
1533
1534 if (ff_layout_write_prepare_common(task, hdr))
1535 return;
1536
1537 rpc_call_start(task);
1538}
1539
1540static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1541{
1542 struct nfs_pgio_header *hdr = data;
1543
6de7e12f
AS		 1544 if (nfs4_setup_sequence(hdr->ds_clp,
1545 &hdr->args.seq_args,
1546 &hdr->res.seq_res,
1547 task))
d67ae825
TH		 1548 return;
1549
bb21ce0a 1550 ff_layout_write_prepare_common(task, hdr);
d67ae825
TH		 1551}
1552
1553static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1554{
1555 struct nfs_pgio_header *hdr = data;
1556
1557 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1558 task->tk_status == 0) {
1559 nfs4_sequence_done(task, &hdr->res.seq_res);
1560 return;
1561 }
1562
1563 /* Note this may cause RPC to be resent */
1564 hdr->mds_ops->rpc_call_done(task, hdr);
1565}
1566
1567static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1568{
1569 struct nfs_pgio_header *hdr = data;
1570
37e9ed22 1571 ff_layout_write_record_layoutstats_done(task, hdr);
d67ae825
TH		 1572 rpc_count_iostats_metrics(task,
1573 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1574}
1575
4d0ac221
TM		 1576static void ff_layout_write_release(void *data)
1577{
1578 struct nfs_pgio_header *hdr = data;
1579
1580 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1581 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1582 ff_layout_send_layouterror(hdr->lseg);
7b0df92a 1583 ff_layout_reset_write(hdr, true);
f0922a6c 1584 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
7b0df92a 1585 ff_layout_reset_write(hdr, false);
4d0ac221
TM		 1586 pnfs_generic_rw_release(data);
1587}
1588
37e9ed22 1589static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
abcb7bfc
TM		 1590 struct nfs_commit_data *cdata)
1591{
37e9ed22
TM		 1592 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1593 return;
0b7baf94 1594 nfs4_ff_layout_stat_io_start_write(cdata->inode,
abcb7bfc 1595 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
e76d28dd 1596 0, task->tk_start);
abcb7bfc
TM		 1597}
1598
37e9ed22
TM		 1599static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1600 struct nfs_commit_data *cdata)
1601{
1602 struct nfs_page *req;
1603 __u64 count = 0;
1604
1605 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1606 return;
1607
1608 if (task->tk_status == 0) {
1609 list_for_each_entry(req, &cdata->pages, wb_list)
1610 count += req->wb_bytes;
1611 }
1612 nfs4_ff_layout_stat_io_end_write(task,
1613 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1614 count, count, NFS_FILE_SYNC);
056f9ad6 1615 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
37e9ed22
TM		 1616}
1617
1618static void ff_layout_commit_prepare_common(struct rpc_task *task,
1619 struct nfs_commit_data *cdata)
1620{
1621 ff_layout_commit_record_layoutstats_start(task, cdata);
1622}
1623
d67ae825
TH		 1624static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1625{
abcb7bfc 1626 ff_layout_commit_prepare_common(task, data);
d67ae825
TH		 1627 rpc_call_start(task);
1628}
1629
1630static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1631{
1632 struct nfs_commit_data *wdata = data;
1633
6de7e12f
AS		 1634 if (nfs4_setup_sequence(wdata->ds_clp,
1635 &wdata->args.seq_args,
1636 &wdata->res.seq_res,
1637 task))
abcb7bfc
TM		 1638 return;
1639 ff_layout_commit_prepare_common(task, data);
1640}
1641
1642static void ff_layout_commit_done(struct rpc_task *task, void *data)
141b9b59
PT		 1643{
1644 pnfs_generic_write_commit_done(task, data);
1645}
1646
1647static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
abcb7bfc
TM		 1648{
1649 struct nfs_commit_data *cdata = data;
abcb7bfc 1650
37e9ed22 1651 ff_layout_commit_record_layoutstats_done(task, cdata);
d67ae825
TH		 1652 rpc_count_iostats_metrics(task,
1653 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1654}
1655
4d0ac221
TM		 1656static void ff_layout_commit_release(void *data)
1657{
1658 struct nfs_commit_data *cdata = data;
1659
1660 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1661 pnfs_generic_commit_release(data);
1662}
1663
d67ae825
TH		 1664static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1665 .rpc_call_prepare = ff_layout_read_prepare_v3,
1666 .rpc_call_done = ff_layout_read_call_done,
1667 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1668 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1669};
1670
1671static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1672 .rpc_call_prepare = ff_layout_read_prepare_v4,
1673 .rpc_call_done = ff_layout_read_call_done,
1674 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1675 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1676};
1677
1678static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1679 .rpc_call_prepare = ff_layout_write_prepare_v3,
1680 .rpc_call_done = ff_layout_write_call_done,
1681 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1682 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1683};
1684
1685static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1686 .rpc_call_prepare = ff_layout_write_prepare_v4,
1687 .rpc_call_done = ff_layout_write_call_done,
1688 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1689 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1690};
1691
1692static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1693 .rpc_call_prepare = ff_layout_commit_prepare_v3,
abcb7bfc 1694 .rpc_call_done = ff_layout_commit_done,
d67ae825 1695 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1696 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1697};
1698
1699static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1700 .rpc_call_prepare = ff_layout_commit_prepare_v4,
abcb7bfc 1701 .rpc_call_done = ff_layout_commit_done,
d67ae825 1702 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1703 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1704};
1705
1706static enum pnfs_try_status
1707ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1708{
1709 struct pnfs_layout_segment *lseg = hdr->lseg;
1710 struct nfs4_pnfs_ds *ds;
1711 struct rpc_clnt *ds_clnt;
2444ff27 1712 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1713 const struct cred *ds_cred;
d67ae825
TH		 1714 loff_t offset = hdr->args.offset;
1715 u32 idx = hdr->pgio_mirror_idx;
1716 int vers;
1717 struct nfs_fh *fh;
1718
5b5e0928 1719 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
d67ae825
TH		 1720 __func__, hdr->inode->i_ino,
1721 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1722
2444ff27
TM		 1723 mirror = FF_LAYOUT_COMP(lseg, idx);
1724 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
d67ae825
TH		 1725 if (!ds)
1726 goto out_failed;
1727
561d6f8a 1728 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1729 hdr->inode);
1730 if (IS_ERR(ds_clnt))
1731 goto out_failed;
1732
312cd4cb 1733 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1734 if (!ds_cred)
d67ae825
TH		 1735 goto out_failed;
1736
626d48b1 1737 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1738
1739 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
212bf41d 1740 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
d67ae825 1741
75bf47eb 1742 hdr->pgio_done_cb = ff_layout_read_done_cb;
212bf41d 1743 refcount_inc(&ds->ds_clp->cl_count);
d67ae825 1744 hdr->ds_clp = ds->ds_clp;
749da527 1745 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1746 if (fh)
1747 hdr->args.fh = fh;
bb21ce0a 1748
4cbc8a57 1749 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1750
d67ae825
TH		 1751 /*
1752 * Note that if we ever decide to split across DSes,
1753 * then we may need to handle dense-like offsets.
1754 */
1755 hdr->args.offset = offset;
1756 hdr->mds_offset = offset;
1757
1758 /* Perform an asynchronous read to ds */
1759 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1760 vers == 3 ? &ff_layout_read_call_ops_v3 :
1761 &ff_layout_read_call_ops_v4,
1762 0, RPC_TASK_SOFTCONN);
a52458b4 1763 put_cred(ds_cred);
d67ae825
TH		 1764 return PNFS_ATTEMPTED;
1765
1766out_failed:
3b13b4b3 1767 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825 1768 return PNFS_TRY_AGAIN;
d5b9216f
TM		 1769 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1770 hdr->args.offset, hdr->args.count,
1771 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
d67ae825
TH		 1772 return PNFS_NOT_ATTEMPTED;
1773}
1774
1775/* Perform async writes. */
1776static enum pnfs_try_status
1777ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1778{
1779 struct pnfs_layout_segment *lseg = hdr->lseg;
1780 struct nfs4_pnfs_ds *ds;
1781 struct rpc_clnt *ds_clnt;
2444ff27 1782 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1783 const struct cred *ds_cred;
d67ae825
TH		 1784 loff_t offset = hdr->args.offset;
1785 int vers;
1786 struct nfs_fh *fh;
1787 int idx = hdr->pgio_mirror_idx;
1788
2444ff27
TM		 1789 mirror = FF_LAYOUT_COMP(lseg, idx);
1790 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825 1791 if (!ds)
37f8aa16 1792 goto out_failed;
d67ae825 1793
561d6f8a 1794 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1795 hdr->inode);
1796 if (IS_ERR(ds_clnt))
37f8aa16 1797 goto out_failed;
d67ae825 1798
312cd4cb 1799 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1800 if (!ds_cred)
37f8aa16 1801 goto out_failed;
d67ae825 1802
626d48b1 1803 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825 1804
5b5e0928 1805 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
d67ae825 1806 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
212bf41d 1807 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1808 vers);
1809
1810 hdr->pgio_done_cb = ff_layout_write_done_cb;
212bf41d 1811 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1812 hdr->ds_clp = ds->ds_clp;
1813 hdr->ds_commit_idx = idx;
749da527 1814 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1815 if (fh)
1816 hdr->args.fh = fh;
1817
4cbc8a57 1818 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1819
d67ae825
TH		 1820 /*
1821 * Note that if we ever decide to split across DSes,
1822 * then we may need to handle dense-like offsets.
1823 */
1824 hdr->args.offset = offset;
1825
1826 /* Perform an asynchronous write */
1827 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1828 vers == 3 ? &ff_layout_write_call_ops_v3 :
1829 &ff_layout_write_call_ops_v4,
1830 sync, RPC_TASK_SOFTCONN);
a52458b4 1831 put_cred(ds_cred);
d67ae825 1832 return PNFS_ATTEMPTED;
37f8aa16
TM		 1833
1834out_failed:
1835 if (ff_layout_avoid_mds_available_ds(lseg))
1836 return PNFS_TRY_AGAIN;
d5b9216f
TM		 1837 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1838 hdr->args.offset, hdr->args.count,
1839 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
37f8aa16 1840 return PNFS_NOT_ATTEMPTED;
d67ae825
TH		 1841}
1842
d67ae825
TH		 1843static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1844{
1845 return i;
1846}
1847
1848static struct nfs_fh *
1849select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1850{
1851 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1852
1853 /* FIXME: Assume that there is only one NFS version available
1854 * for the DS.
1855 */
1856 return &flseg->mirror_array[i]->fh_versions[0];
1857}
1858
1859static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1860{
1861 struct pnfs_layout_segment *lseg = data->lseg;
1862 struct nfs4_pnfs_ds *ds;
1863 struct rpc_clnt *ds_clnt;
2444ff27 1864 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1865 const struct cred *ds_cred;
d67ae825 1866 u32 idx;
57f3f4c0 1867 int vers, ret;
d67ae825
TH		 1868 struct nfs_fh *fh;
1869
4b75053e
TM		 1870 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1871 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1872 goto out_err;
1873
d67ae825 1874 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
2444ff27
TM		 1875 mirror = FF_LAYOUT_COMP(lseg, idx);
1876 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825
TH		 1877 if (!ds)
1878 goto out_err;
1879
561d6f8a 1880 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1881 data->inode);
1882 if (IS_ERR(ds_clnt))
1883 goto out_err;
1884
312cd4cb 1885 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
57f3f4c0 1886 if (!ds_cred)
d67ae825
TH		 1887 goto out_err;
1888
626d48b1 1889 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1890
1891 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
212bf41d 1892 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1893 vers);
1894 data->commit_done_cb = ff_layout_commit_done_cb;
1895 data->cred = ds_cred;
212bf41d 1896 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1897 data->ds_clp = ds->ds_clp;
1898 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1899 if (fh)
1900 data->args.fh = fh;
abcb7bfc 1901
57f3f4c0 1902 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
d67ae825
TH		 1903 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1904 &ff_layout_commit_call_ops_v4,
1905 how, RPC_TASK_SOFTCONN);
a52458b4 1906 put_cred(ds_cred);
57f3f4c0 1907 return ret;
d67ae825
TH		 1908out_err:
1909 pnfs_generic_prepare_to_resend_writes(data);
1910 pnfs_generic_commit_release(data);
1911 return -EAGAIN;
1912}
1913
1914static int
1915ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1916 int how, struct nfs_commit_info *cinfo)
1917{
1918 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1919 ff_layout_initiate_commit);
1920}
1921
1922static struct pnfs_ds_commit_info *
1923ff_layout_get_ds_info(struct inode *inode)
1924{
1925 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1926
1927 if (layout == NULL)
1928 return NULL;
1929
1930 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1931}
1932
ba827c9a
TM		 1933static void
1934ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1935 struct pnfs_layout_segment *lseg)
1936{
1937 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1938 struct inode *inode = lseg->pls_layout->plh_inode;
1939 struct pnfs_commit_array *array, *new;
1940
1941 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1942 if (new) {
1943 spin_lock(&inode->i_lock);
1944 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1945 spin_unlock(&inode->i_lock);
1946 if (array != new)
1947 pnfs_free_commit_array(new);
1948 }
1949}
1950
a9901899
TM		 1951static void
1952ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1953 struct inode *inode)
1954{
1955 spin_lock(&inode->i_lock);
1956 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1957 spin_unlock(&inode->i_lock);
1958}
1959
d67ae825 1960static void
fc87701b 1961ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
d67ae825
TH		 1962{
1963 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1964 id_node));
1965}
1966
5ba6a09e 1967static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
5b9b3c85
TM		 1968 const struct nfs4_layoutreturn_args *args,
1969 const struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825 1970{
d67ae825 1971 __be32 *start;
d67ae825
TH		 1972
1973 start = xdr_reserve_space(xdr, 4);
1974 if (unlikely(!start))
1975 return -E2BIG;
1976
5b9b3c85 1977 *start = cpu_to_be32(ff_args->num_errors);
d67ae825 1978 /* This assume we always return _ALL_ layouts */
5b9b3c85 1979 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
d67ae825
TH		 1980}
1981
230bc962
TM		 1982static void
1983encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1984{
ab6e9aaf 1985 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
230bc962
TM		 1986}
1987
1988static void
1989ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1990 const nfs4_stateid *stateid,
1991 const struct nfs42_layoutstat_devinfo *devinfo)
1992{
1993 __be32 *p;
1994
1995 p = xdr_reserve_space(xdr, 8 + 8);
1996 p = xdr_encode_hyper(p, devinfo->offset);
1997 p = xdr_encode_hyper(p, devinfo->length);
1998 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
1999 p = xdr_reserve_space(xdr, 4*8);
2000 p = xdr_encode_hyper(p, devinfo->read_count);
2001 p = xdr_encode_hyper(p, devinfo->read_bytes);
2002 p = xdr_encode_hyper(p, devinfo->write_count);
2003 p = xdr_encode_hyper(p, devinfo->write_bytes);
2004 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2005}
2006
2007static void
2008ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2009 const nfs4_stateid *stateid,
2010 const struct nfs42_layoutstat_devinfo *devinfo)
2011{
2012 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2013 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2014 devinfo->ld_private.data);
2015}
2016
d67ae825 2017/* report nothing for now */
230bc962
TM		 2018static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2019 const struct nfs4_layoutreturn_args *args,
2020 struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825
TH		 2021{
2022 __be32 *p;
230bc962 2023 int i;
d67ae825
TH		 2024
2025 p = xdr_reserve_space(xdr, 4);
230bc962
TM		 2026 *p = cpu_to_be32(ff_args->num_dev);
2027 for (i = 0; i < ff_args->num_dev; i++)
2028 ff_layout_encode_ff_iostat(xdr,
2029 &args->layout->plh_stateid,
2030 &ff_args->devinfo[i]);
2031}
2032
2033static void
2034ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2035 unsigned int num_entries)
2036{
2037 unsigned int i;
2038
2039 for (i = 0; i < num_entries; i++) {
2040 if (!devinfo[i].ld_private.ops)
2041 continue;
2042 if (!devinfo[i].ld_private.ops->free)
2043 continue;
2044 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2045 }
d67ae825
TH		 2046}
2047
2048static struct nfs4_deviceid_node *
2049ff_layout_alloc_deviceid_node(struct nfs_server *server,
2050 struct pnfs_device *pdev, gfp_t gfp_flags)
2051{
2052 struct nfs4_ff_layout_ds *dsaddr;
2053
2054 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2055 if (!dsaddr)
2056 return NULL;
2057 return &dsaddr->id_node;
2058}
2059
2060static void
94e5c571 2061ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
5b9b3c85
TM		 2062 const void *voidargs,
2063 const struct nfs4_xdr_opaque_data *ff_opaque)
d67ae825 2064{
5b9b3c85 2065 const struct nfs4_layoutreturn_args *args = voidargs;
5ba6a09e 2066 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
d9152114
TM		 2067 struct xdr_buf tmp_buf = {
2068 .head = {
2069 [0] = {
2070 .iov_base = page_address(ff_args->pages[0]),
2071 },
2072 },
2073 .buflen = PAGE_SIZE,
2074 };
2075 struct xdr_stream tmp_xdr;
d67ae825
TH		 2076 __be32 *start;
2077
2078 dprintk("%s: Begin\n", __func__);
d67ae825 2079
0ccc61b1 2080 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
d9152114
TM		 2081
2082 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2083 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2084
2085 start = xdr_reserve_space(xdr, 4);
2086 *start = cpu_to_be32(tmp_buf.len);
2087 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
082fa37d 2088
d67ae825
TH		 2089 dprintk("%s: Return\n", __func__);
2090}
2091
5b9b3c85
TM		 2092static void
2093ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2094{
2095 struct nfs4_flexfile_layoutreturn_args *ff_args;
2096
2097 if (!args->data)
2098 return;
2099 ff_args = args->data;
2100 args->data = NULL;
2101
2102 ff_layout_free_ds_ioerr(&ff_args->errors);
230bc962 2103 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
5b9b3c85 2104
d9152114 2105 put_page(ff_args->pages[0]);
5b9b3c85
TM		 2106 kfree(ff_args);
2107}
2108
2e54b9b1 2109static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
5b9b3c85
TM		 2110 .encode = ff_layout_encode_layoutreturn,
2111 .free = ff_layout_free_layoutreturn,
2112};
2113
2114static int
2115ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2116{
2117 struct nfs4_flexfile_layoutreturn_args *ff_args;
230bc962 2118 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
5b9b3c85
TM		 2119
2120 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2121 if (!ff_args)
d9152114
TM		 2122 goto out_nomem;
2123 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2124 if (!ff_args->pages[0])
2125 goto out_nomem_free;
5b9b3c85
TM		 2126
2127 INIT_LIST_HEAD(&ff_args->errors);
2128 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2129 &args->range, &ff_args->errors,
2130 FF_LAYOUTRETURN_MAXERR);
2131
230bc962
TM		 2132 spin_lock(&args->inode->i_lock);
2133 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2134 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2135 spin_unlock(&args->inode->i_lock);
2136
5b9b3c85
TM		 2137 args->ld_private->ops = &layoutreturn_ops;
2138 args->ld_private->data = ff_args;
2139 return 0;
d9152114
TM		 2140out_nomem_free:
2141 kfree(ff_args);
2142out_nomem:
2143 return -ENOMEM;
5b9b3c85
TM		 2144}
2145
f0922a6c
TM		 2146#ifdef CONFIG_NFS_V4_2
2147void
2148ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2149{
2150 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2151 struct nfs42_layout_error *errors;
2152 LIST_HEAD(head);
2153
2154 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2155 return;
2156 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2157 if (list_empty(&head))
2158 return;
2159
2160 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2161 sizeof(*errors), GFP_NOFS);
2162 if (errors != NULL) {
2163 const struct nfs4_ff_layout_ds_err *pos;
2164 size_t n = 0;
2165
2166 list_for_each_entry(pos, &head, list) {
2167 errors[n].offset = pos->offset;
2168 errors[n].length = pos->length;
2169 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2170 errors[n].errors[0].dev_id = pos->deviceid;
2171 errors[n].errors[0].status = pos->status;
2172 errors[n].errors[0].opnum = pos->opnum;
2173 n++;
2174 if (!list_is_last(&pos->list, &head) &&
2175 n < NFS42_LAYOUTERROR_MAX)
2176 continue;
2177 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2178 break;
2179 n = 0;
2180 }
2181 kfree(errors);
2182 }
2183 ff_layout_free_ds_ioerr(&head);
2184}
2185#else
2186void
2187ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2188{
2189}
2190#endif
2191
27c43064
PT		 2192static int
2193ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2194{
2195 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2196
2197 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2198}
2199
2200static size_t
2201ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2202 const int buflen)
2203{
2204 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2205 const struct in6_addr *addr = &sin6->sin6_addr;
2206
2207 /*
2208 * RFC 4291, Section 2.2.2
2209 *
2210 * Shorthanded ANY address
2211 */
2212 if (ipv6_addr_any(addr))
2213 return snprintf(buf, buflen, "::");
2214
2215 /*
2216 * RFC 4291, Section 2.2.2
2217 *
2218 * Shorthanded loopback address
2219 */
2220 if (ipv6_addr_loopback(addr))
2221 return snprintf(buf, buflen, "::1");
2222
2223 /*
2224 * RFC 4291, Section 2.2.3
2225 *
2226 * Special presentation address format for mapped v4
2227 * addresses.
2228 */
2229 if (ipv6_addr_v4mapped(addr))
2230 return snprintf(buf, buflen, "::ffff:%pI4",
2231 &addr->s6_addr32[3]);
2232
2233 /*
2234 * RFC 4291, Section 2.2.1
2235 */
2236 return snprintf(buf, buflen, "%pI6c", addr);
2237}
2238
2239/* Derived from rpc_sockaddr2uaddr */
2240static void
2241ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2242{
2243 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2244 char portbuf[RPCBIND_MAXUADDRPLEN];
2245 char addrbuf[RPCBIND_MAXUADDRLEN];
2246 char *netid;
2247 unsigned short port;
2248 int len, netid_len;
2249 __be32 *p;
2250
2251 switch (sap->sa_family) {
2252 case AF_INET:
2253 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2254 return;
2255 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2256 netid = "tcp";
2257 netid_len = 3;
2258 break;
2259 case AF_INET6:
2260 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2261 return;
2262 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2263 netid = "tcp6";
2264 netid_len = 4;
2265 break;
2266 default:
2267 /* we only support tcp and tcp6 */
2268 WARN_ON_ONCE(1);
2269 return;
2270 }
2271
2272 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2273 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2274
2275 p = xdr_reserve_space(xdr, 4 + netid_len);
2276 xdr_encode_opaque(p, netid, netid_len);
2277
2278 p = xdr_reserve_space(xdr, 4 + len);
2279 xdr_encode_opaque(p, addrbuf, len);
2280}
2281
2282static void
2283ff_layout_encode_nfstime(struct xdr_stream *xdr,
2284 ktime_t t)
2285{
2286 struct timespec64 ts;
2287 __be32 *p;
2288
2289 p = xdr_reserve_space(xdr, 12);
2290 ts = ktime_to_timespec64(t);
2291 p = xdr_encode_hyper(p, ts.tv_sec);
2292 *p++ = cpu_to_be32(ts.tv_nsec);
2293}
2294
2295static void
2296ff_layout_encode_io_latency(struct xdr_stream *xdr,
2297 struct nfs4_ff_io_stat *stat)
2298{
2299 __be32 *p;
2300
2301 p = xdr_reserve_space(xdr, 5 * 8);
2302 p = xdr_encode_hyper(p, stat->ops_requested);
2303 p = xdr_encode_hyper(p, stat->bytes_requested);
2304 p = xdr_encode_hyper(p, stat->ops_completed);
2305 p = xdr_encode_hyper(p, stat->bytes_completed);
2306 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2307 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2308 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2309}
2310
2311static void
422c93c8
TM		 2312ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2313 const struct nfs42_layoutstat_devinfo *devinfo,
2314 struct nfs4_ff_layout_mirror *mirror)
27c43064 2315{
27c43064
PT		 2316 struct nfs4_pnfs_ds_addr *da;
2317 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2318 struct nfs_fh *fh = &mirror->fh_versions[0];
422c93c8 2319 __be32 *p;
27c43064
PT		 2320
2321 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2322 dprintk("%s: DS %s: encoding address %s\n",
2323 __func__, ds->ds_remotestr, da->da_remotestr);
27c43064
PT		 2324 /* netaddr4 */
2325 ff_layout_encode_netaddr(xdr, da);
2326 /* nfs_fh4 */
2327 p = xdr_reserve_space(xdr, 4 + fh->size);
2328 xdr_encode_opaque(p, fh->data, fh->size);
2329 /* ff_io_latency4 read */
2330 spin_lock(&mirror->lock);
2331 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2332 /* ff_io_latency4 write */
2333 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2334 spin_unlock(&mirror->lock);
2335 /* nfstime4 */
2336 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2337 /* bool */
2338 p = xdr_reserve_space(xdr, 4);
2339 *p = cpu_to_be32(false);
422c93c8
TM		 2340}
2341
2342static void
2343ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2344 const struct nfs4_xdr_opaque_data *opaque)
2345{
2346 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2347 struct nfs42_layoutstat_devinfo, ld_private);
2348 __be32 *start;
2349
2350 /* layoutupdate length */
2351 start = xdr_reserve_space(xdr, 4);
2352 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
27c43064
PT		 2353
2354 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2355}
2356
422c93c8
TM		 2357static void
2358ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2359{
2360 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2361
2362 ff_layout_put_mirror(mirror);
2363}
2364
2365static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2366 .encode = ff_layout_encode_layoutstats,
2367 .free = ff_layout_free_layoutstats,
2368};
2369
266d12d4 2370static int
08e2e5bc
TM		 2371ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2372 struct nfs42_layoutstat_devinfo *devinfo,
266d12d4 2373 int dev_limit)
ad4dc53e 2374{
266d12d4 2375 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
ad4dc53e
PT		 2376 struct nfs4_ff_layout_mirror *mirror;
2377 struct nfs4_deviceid_node *dev;
266d12d4 2378 int i = 0;
ad4dc53e 2379
266d12d4
TM		 2380 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2381 if (i >= dev_limit)
ad4dc53e 2382 break;
65990d1a 2383 if (IS_ERR_OR_NULL(mirror->mirror_ds))
27c43064 2384 continue;
06946c6a
TM		 2385 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2386 continue;
266d12d4 2387 /* mirror refcount put in cleanup_layoutstats */
81a090b9 2388 if (!refcount_inc_not_zero(&mirror->ref))
266d12d4
TM		 2389 continue;
2390 dev = &mirror->mirror_ds->id_node;
ad4dc53e 2391 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
266d12d4
TM		 2392 devinfo->offset = 0;
2393 devinfo->length = NFS4_MAX_UINT64;
2f8220c1 2394 spin_lock(&mirror->lock);
d099d7b8 2395 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
ad4dc53e 2396 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
d099d7b8 2397 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
ad4dc53e 2398 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2f8220c1 2399 spin_unlock(&mirror->lock);
ad4dc53e 2400 devinfo->layout_type = LAYOUT_FLEX_FILES;
422c93c8
TM		 2401 devinfo->ld_private.ops = &layoutstat_ops;
2402 devinfo->ld_private.data = mirror;
ad4dc53e 2403
08e2e5bc 2404 devinfo++;
266d12d4 2405 i++;
ad4dc53e 2406 }
266d12d4 2407 return i;
ad4dc53e
PT		 2408}
2409
2410static int
2411ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2412{
266d12d4 2413 struct nfs4_flexfile_layout *ff_layout;
08e2e5bc 2414 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
ad4dc53e 2415
ad4dc53e 2416 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
266d12d4 2417 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
ad4dc53e
PT		 2418 if (!args->devinfo)
2419 return -ENOMEM;
2420
ad4dc53e 2421 spin_lock(&args->inode->i_lock);
08e2e5bc
TM		 2422 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2423 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2424 &args->devinfo[0], dev_count);
ad4dc53e 2425 spin_unlock(&args->inode->i_lock);
46c98c6d
TM		 2426 if (!args->num_dev) {
2427 kfree(args->devinfo);
2428 args->devinfo = NULL;
2429 return -ENOENT;
2430 }
ad4dc53e
PT		 2431
2432 return 0;
2433}
2434
76b2a303
TM		 2435static int
2436ff_layout_set_layoutdriver(struct nfs_server *server,
2437 const struct nfs_fh *dummy)
2438{
2439#if IS_ENABLED(CONFIG_NFS_V4_2)
2440 server->caps |= NFS_CAP_LAYOUTSTATS;
2441#endif
2442 return 0;
2443}
2444
9c455a8c
TM		 2445static const struct pnfs_commit_ops ff_layout_commit_ops = {
2446 .setup_ds_info = ff_layout_setup_ds_info,
2447 .release_ds_info = ff_layout_release_ds_info,
2448 .mark_request_commit = pnfs_layout_mark_request_commit,
2449 .clear_request_commit = pnfs_generic_clear_request_commit,
2450 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2451 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2452 .commit_pagelist = ff_layout_commit_pagelist,
2453};
2454
d67ae825
TH		 2455static struct pnfs_layoutdriver_type flexfilelayout_type = {
2456 .id = LAYOUT_FLEX_FILES,
2457 .name = "LAYOUT_FLEX_FILES",
2458 .owner = THIS_MODULE,
f86c3ac5 2459 .flags = PNFS_LAYOUTGET_ON_OPEN,
28ced9a8 2460 .max_layoutget_response = 4096, /* 1 page or so... */
76b2a303 2461 .set_layoutdriver = ff_layout_set_layoutdriver,
d67ae825
TH		 2462 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2463 .free_layout_hdr = ff_layout_free_layout_hdr,
2464 .alloc_lseg = ff_layout_alloc_lseg,
2465 .free_lseg = ff_layout_free_lseg,
0762ed2c 2466 .add_lseg = ff_layout_add_lseg,
d67ae825
TH		 2467 .pg_read_ops = &ff_layout_pg_read_ops,
2468 .pg_write_ops = &ff_layout_pg_write_ops,
2469 .get_ds_info = ff_layout_get_ds_info,
fc87701b 2470 .free_deviceid_node = ff_layout_free_deviceid_node,
d67ae825
TH		 2471 .read_pagelist = ff_layout_read_pagelist,
2472 .write_pagelist = ff_layout_write_pagelist,
2473 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
5b9b3c85 2474 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
5bb89b47 2475 .sync = pnfs_nfs_generic_sync,
ad4dc53e 2476 .prepare_layoutstats = ff_layout_prepare_layoutstats,
d67ae825
TH		 2477};
2478
2479static int __init nfs4flexfilelayout_init(void)
2480{
2481 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2482 __func__);
2483 return pnfs_register_layoutdriver(&flexfilelayout_type);
2484}
2485
2486static void __exit nfs4flexfilelayout_exit(void)
2487{
2488 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2489 __func__);
2490 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2491}
2492
2493MODULE_ALIAS("nfs-layouttype4-4");
2494
2495MODULE_LICENSE("GPL");
2496MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2497
2498module_init(nfs4flexfilelayout_init);
2499module_exit(nfs4flexfilelayout_exit);
33344e0f
TM		 2500
2501module_param(io_maxretrans, ushort, 0644);
2502MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2503 "retries an I/O request before returning an error. ");
Ubuntu Jammy Linux kernel mirror