projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
pNFS/flexfiles: Fix layoutstats handling during read failovers
[linux-2.6-block.git] / fs / nfs / flexfilelayout / flexfilelayout.c
CommitLineData
d67ae825
TH		 1/*
2 * Module for pnfs flexfile layout driver.
3 *
4 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
5 *
6 * Tao Peng <bergwolf@primarydata.com>
7 */
8
9#include <linux/nfs_fs.h>
10#include <linux/nfs_page.h>
11#include <linux/module.h>
97f68c6b 12#include <linux/sched/mm.h>
d67ae825
TH		 13
14#include <linux/sunrpc/metrics.h>
d67ae825
TH		 15
16#include "flexfilelayout.h"
17#include "../nfs4session.h"
40c64c26 18#include "../nfs4idmap.h"
d67ae825
TH		 19#include "../internal.h"
20#include "../delegation.h"
21#include "../nfs4trace.h"
22#include "../iostat.h"
23#include "../nfs.h"
ad4dc53e 24#include "../nfs42.h"
d67ae825
TH		 25
26#define NFSDBG_FACILITY NFSDBG_PNFS_LD
27
28#define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
5b9b3c85
TM		 29#define FF_LAYOUTRETURN_MAXERR 20
30
54e4a0df
TM		 31static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
32 struct nfs_pgio_header *hdr);
230bc962
TM		 33static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
34 struct nfs42_layoutstat_devinfo *devinfo,
35 int dev_limit);
36static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
37 const struct nfs42_layoutstat_devinfo *devinfo,
38 struct nfs4_ff_layout_mirror *mirror);
54e4a0df 39
d67ae825
TH		 40static struct pnfs_layout_hdr *
41ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
42{
43 struct nfs4_flexfile_layout *ffl;
44
45 ffl = kzalloc(sizeof(*ffl), gfp_flags);
46 if (ffl) {
47 INIT_LIST_HEAD(&ffl->error_list);
266d12d4 48 INIT_LIST_HEAD(&ffl->mirrors);
1c8d477a 49 ffl->last_report_time = ktime_get();
d67ae825
TH		 50 return &ffl->generic_hdr;
51 } else
52 return NULL;
53}
54
55static void
56ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
57{
58 struct nfs4_ff_layout_ds_err *err, *n;
59
60 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
61 list) {
62 list_del(&err->list);
63 kfree(err);
64 }
65 kfree(FF_LAYOUT_FROM_HDR(lo));
66}
67
93b717fd 68static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
d67ae825
TH		 69{
70 __be32 *p;
71
72 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
73 if (unlikely(p == NULL))
74 return -ENOBUFS;
93b717fd
TM		 75 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
76 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
d67ae825
TH		 77 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
78 p[0], p[1], p[2], p[3]);
79 return 0;
80}
81
82static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
83{
84 __be32 *p;
85
86 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
87 if (unlikely(!p))
88 return -ENOBUFS;
89 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
90 nfs4_print_deviceid(devid);
91 return 0;
92}
93
94static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
95{
96 __be32 *p;
97
98 p = xdr_inline_decode(xdr, 4);
99 if (unlikely(!p))
100 return -ENOBUFS;
101 fh->size = be32_to_cpup(p++);
102 if (fh->size > sizeof(struct nfs_fh)) {
103 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
104 fh->size);
105 return -EOVERFLOW;
106 }
107 /* fh.data */
108 p = xdr_inline_decode(xdr, fh->size);
109 if (unlikely(!p))
110 return -ENOBUFS;
111 memcpy(&fh->data, p, fh->size);
112 dprintk("%s: fh len %d\n", __func__, fh->size);
113
114 return 0;
115}
116
117/*
118 * Currently only stringified uids and gids are accepted.
119 * I.e., kerberos is not supported to the DSes, so no pricipals.
120 *
121 * That means that one common function will suffice, but when
122 * principals are added, this should be split to accomodate
123 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
124 */
125static int
126decode_name(struct xdr_stream *xdr, u32 *id)
127{
128 __be32 *p;
129 int len;
130
131 /* opaque_length(4)*/
132 p = xdr_inline_decode(xdr, 4);
133 if (unlikely(!p))
134 return -ENOBUFS;
135 len = be32_to_cpup(p++);
136 if (len < 0)
137 return -EINVAL;
138
139 dprintk("%s: len %u\n", __func__, len);
140
141 /* opaque body */
142 p = xdr_inline_decode(xdr, len);
143 if (unlikely(!p))
144 return -ENOBUFS;
145
146 if (!nfs_map_string_to_numeric((char *)p, len, id))
147 return -EINVAL;
148
149 return 0;
150}
151
266d12d4
TM		 152static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
153 const struct nfs4_ff_layout_mirror *m2)
154{
155 int i, j;
156
157 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
158 return false;
159 for (i = 0; i < m1->fh_versions_cnt; i++) {
160 bool found_fh = false;
86fb449b 161 for (j = 0; j < m2->fh_versions_cnt; j++) {
266d12d4
TM		 162 if (nfs_compare_fh(&m1->fh_versions[i],
163 &m2->fh_versions[j]) == 0) {
164 found_fh = true;
165 break;
166 }
167 }
168 if (!found_fh)
169 return false;
170 }
171 return true;
172}
173
174static struct nfs4_ff_layout_mirror *
175ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
176 struct nfs4_ff_layout_mirror *mirror)
177{
178 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
179 struct nfs4_ff_layout_mirror *pos;
180 struct inode *inode = lo->plh_inode;
181
182 spin_lock(&inode->i_lock);
183 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
65990d1a 184 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
266d12d4
TM		 185 continue;
186 if (!ff_mirror_match_fh(mirror, pos))
187 continue;
81a090b9 188 if (refcount_inc_not_zero(&pos->ref)) {
266d12d4
TM		 189 spin_unlock(&inode->i_lock);
190 return pos;
191 }
192 }
193 list_add(&mirror->mirrors, &ff_layout->mirrors);
194 mirror->layout = lo;
195 spin_unlock(&inode->i_lock);
196 return mirror;
197}
198
e3b1df2d 199static void
266d12d4
TM		 200ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
201{
202 struct inode *inode;
203 if (mirror->layout == NULL)
204 return;
205 inode = mirror->layout->plh_inode;
206 spin_lock(&inode->i_lock);
207 list_del(&mirror->mirrors);
208 spin_unlock(&inode->i_lock);
209 mirror->layout = NULL;
210}
211
28a0d72c
TM		 212static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
213{
214 struct nfs4_ff_layout_mirror *mirror;
215
216 mirror = kzalloc(sizeof(*mirror), gfp_flags);
217 if (mirror != NULL) {
218 spin_lock_init(&mirror->lock);
81a090b9 219 refcount_set(&mirror->ref, 1);
266d12d4 220 INIT_LIST_HEAD(&mirror->mirrors);
28a0d72c
TM		 221 }
222 return mirror;
223}
224
225static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
226{
a52458b4 227 const struct cred *cred;
3064b686 228
266d12d4 229 ff_layout_remove_mirror(mirror);
28a0d72c 230 kfree(mirror->fh_versions);
3064b686 231 cred = rcu_access_pointer(mirror->ro_cred);
a52458b4 232 put_cred(cred);
3064b686 233 cred = rcu_access_pointer(mirror->rw_cred);
a52458b4 234 put_cred(cred);
28a0d72c
TM		 235 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
236 kfree(mirror);
237}
238
239static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
240{
81a090b9 241 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
28a0d72c
TM		 242 ff_layout_free_mirror(mirror);
243}
244
d67ae825
TH		 245static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
246{
247 int i;
248
249 if (fls->mirror_array) {
250 for (i = 0; i < fls->mirror_array_cnt; i++) {
251 /* normally mirror_ds is freed in
252 * .free_deviceid_node but we still do it here
253 * for .alloc_lseg error path */
28a0d72c 254 ff_layout_put_mirror(fls->mirror_array[i]);
d67ae825
TH		 255 }
256 kfree(fls->mirror_array);
257 fls->mirror_array = NULL;
258 }
259}
260
261static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
262{
263 int ret = 0;
264
265 dprintk("--> %s\n", __func__);
266
267 /* FIXME: remove this check when layout segment support is added */
268 if (lgr->range.offset != 0 ||
269 lgr->range.length != NFS4_MAX_UINT64) {
270 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
271 __func__);
272 ret = -EINVAL;
273 }
274
275 dprintk("--> %s returns %d\n", __func__, ret);
276 return ret;
277}
278
279static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
280{
281 if (fls) {
282 ff_layout_free_mirror_array(fls);
283 kfree(fls);
284 }
285}
286
0762ed2c
TM		 287static bool
288ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
289 const struct pnfs_layout_range *l2)
290{
291 u64 end1, end2;
292
293 if (l1->iomode != l2->iomode)
294 return l1->iomode != IOMODE_READ;
295 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
296 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
297 if (end1 < l2->offset)
298 return false;
299 if (end2 < l1->offset)
300 return true;
301 return l2->offset <= l1->offset;
302}
303
304static bool
305ff_lseg_merge(struct pnfs_layout_segment *new,
306 struct pnfs_layout_segment *old)
307{
308 u64 new_end, old_end;
309
ee26bdd6
JL		 310 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
311 return false;
0762ed2c
TM		 312 if (new->pls_range.iomode != old->pls_range.iomode)
313 return false;
314 old_end = pnfs_calc_offset_end(old->pls_range.offset,
315 old->pls_range.length);
316 if (old_end < new->pls_range.offset)
317 return false;
318 new_end = pnfs_calc_offset_end(new->pls_range.offset,
319 new->pls_range.length);
320 if (new_end < old->pls_range.offset)
321 return false;
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;
399 fls = kzalloc(sizeof(*fls), gfp_flags);
400 if (!fls)
401 goto out_err_free;
402
403 fls->mirror_array_cnt = mirror_array_cnt;
404 fls->stripe_unit = stripe_unit;
405 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
406 sizeof(fls->mirror_array[0]), gfp_flags);
407 if (fls->mirror_array == NULL)
408 goto out_err_free;
409
410 for (i = 0; i < fls->mirror_array_cnt; i++) {
266d12d4 411 struct nfs4_ff_layout_mirror *mirror;
97f68c6b 412 struct cred *kcred;
762bb7e9 413 const struct cred __rcu *cred;
8276c902
N		 414 kuid_t uid;
415 kgid_t gid;
90a0be00 416 u32 ds_count, fh_count, id;
d67ae825
TH		 417 int j;
418
419 rc = -EIO;
420 p = xdr_inline_decode(&stream, 4);
421 if (!p)
422 goto out_err_free;
423 ds_count = be32_to_cpup(p);
424
425 /* FIXME: allow for striping? */
426 if (ds_count != 1)
427 goto out_err_free;
428
28a0d72c 429 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
d67ae825
TH		 430 if (fls->mirror_array[i] == NULL) {
431 rc = -ENOMEM;
432 goto out_err_free;
433 }
434
d67ae825
TH		 435 fls->mirror_array[i]->ds_count = ds_count;
436
437 /* deviceid */
65990d1a 438 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
d67ae825
TH		 439 if (rc)
440 goto out_err_free;
441
d67ae825
TH		 442 /* efficiency */
443 rc = -EIO;
444 p = xdr_inline_decode(&stream, 4);
445 if (!p)
446 goto out_err_free;
447 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
448
449 /* stateid */
93b717fd 450 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
d67ae825
TH		 451 if (rc)
452 goto out_err_free;
453
454 /* fh */
662f9a10 455 rc = -EIO;
d67ae825
TH		 456 p = xdr_inline_decode(&stream, 4);
457 if (!p)
458 goto out_err_free;
459 fh_count = be32_to_cpup(p);
460
461 fls->mirror_array[i]->fh_versions =
6396bb22 462 kcalloc(fh_count, sizeof(struct nfs_fh),
d67ae825
TH		 463 gfp_flags);
464 if (fls->mirror_array[i]->fh_versions == NULL) {
465 rc = -ENOMEM;
466 goto out_err_free;
467 }
468
469 for (j = 0; j < fh_count; j++) {
470 rc = decode_nfs_fh(&stream,
471 &fls->mirror_array[i]->fh_versions[j]);
472 if (rc)
473 goto out_err_free;
474 }
475
476 fls->mirror_array[i]->fh_versions_cnt = fh_count;
477
478 /* user */
90a0be00 479 rc = decode_name(&stream, &id);
d67ae825
TH		 480 if (rc)
481 goto out_err_free;
482
8276c902 483 uid = make_kuid(&init_user_ns, id);
90a0be00 484
d67ae825 485 /* group */
90a0be00 486 rc = decode_name(&stream, &id);
d67ae825
TH		 487 if (rc)
488 goto out_err_free;
489
8276c902 490 gid = make_kgid(&init_user_ns, id);
90a0be00 491
97f68c6b
N		 492 if (gfp_flags & __GFP_FS)
493 kcred = prepare_kernel_cred(NULL);
494 else {
495 unsigned int nofs_flags = memalloc_nofs_save();
496 kcred = prepare_kernel_cred(NULL);
497 memalloc_nofs_restore(nofs_flags);
498 }
499 rc = -ENOMEM;
500 if (!kcred)
501 goto out_err_free;
8276c902
N		 502 kcred->fsuid = uid;
503 kcred->fsgid = gid;
762bb7e9 504 cred = RCU_INITIALIZER(kcred);
90a0be00 505
3064b686
JL		 506 if (lgr->range.iomode == IOMODE_READ)
507 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
508 else
509 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
90a0be00 510
266d12d4
TM		 511 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
512 if (mirror != fls->mirror_array[i]) {
90a0be00 513 /* swap cred ptrs so free_mirror will clean up old */
3064b686
JL		 514 if (lgr->range.iomode == IOMODE_READ) {
515 cred = xchg(&mirror->ro_cred, cred);
516 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
517 } else {
518 cred = xchg(&mirror->rw_cred, cred);
519 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
520 }
266d12d4
TM		 521 ff_layout_free_mirror(fls->mirror_array[i]);
522 fls->mirror_array[i] = mirror;
523 }
524
3064b686
JL		 525 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
526 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
8276c902
N		 527 from_kuid(&init_user_ns, uid),
528 from_kgid(&init_user_ns, gid));
d67ae825
TH		 529 }
530
c0f5f505 531 p = xdr_inline_decode(&stream, 4);
d0379a5d
TM		 532 if (!p)
533 goto out_sort_mirrors;
534 fls->flags = be32_to_cpup(p);
535
536 p = xdr_inline_decode(&stream, 4);
537 if (!p)
538 goto out_sort_mirrors;
539 for (i=0; i < fls->mirror_array_cnt; i++)
540 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
c0f5f505 541
d0379a5d 542out_sort_mirrors:
d67ae825
TH		 543 ff_layout_sort_mirrors(fls);
544 rc = ff_layout_check_layout(lgr);
545 if (rc)
546 goto out_err_free;
d67ae825
TH		 547 ret = &fls->generic_hdr;
548 dprintk("<-- %s (success)\n", __func__);
549out_free_page:
550 __free_page(scratch);
551 return ret;
552out_err_free:
553 _ff_layout_free_lseg(fls);
554 ret = ERR_PTR(rc);
555 dprintk("<-- %s (%d)\n", __func__, rc);
556 goto out_free_page;
557}
558
559static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
560{
561 struct pnfs_layout_segment *lseg;
562
563 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
564 if (lseg->pls_range.iomode == IOMODE_RW)
565 return true;
566
567 return false;
568}
569
570static void
571ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
572{
573 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825
TH		 574
575 dprintk("--> %s\n", __func__);
576
d67ae825
TH		 577 if (lseg->pls_range.iomode == IOMODE_RW) {
578 struct nfs4_flexfile_layout *ffl;
579 struct inode *inode;
580
581 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
582 inode = ffl->generic_hdr.plh_inode;
583 spin_lock(&inode->i_lock);
584 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
585 ffl->commit_info.nbuckets = 0;
586 kfree(ffl->commit_info.buckets);
587 ffl->commit_info.buckets = NULL;
588 }
589 spin_unlock(&inode->i_lock);
590 }
591 _ff_layout_free_lseg(fls);
592}
593
594/* Return 1 until we have multiple lsegs support */
595static int
596ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
597{
598 return 1;
599}
600
abcb7bfc 601static void
e76d28dd 602nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 603{
9bbd9bb4
PT		 604 /* first IO request? */
605 if (atomic_inc_return(&timer->n_ops) == 1) {
e76d28dd 606 timer->start_time = now;
abcb7bfc
TM		 607 }
608}
609
610static ktime_t
e76d28dd 611nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 612{
e76d28dd 613 ktime_t start;
abcb7bfc 614
9bbd9bb4
PT		 615 if (atomic_dec_return(&timer->n_ops) < 0)
616 WARN_ON_ONCE(1);
617
9bbd9bb4
PT		 618 start = timer->start_time;
619 timer->start_time = now;
abcb7bfc
TM		 620 return ktime_sub(now, start);
621}
622
97ba375b 623static bool
d983803d 624nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
e76d28dd
TM		 625 struct nfs4_ff_layoutstat *layoutstat,
626 ktime_t now)
abcb7bfc 627{
bbf58bf3 628 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
1c8d477a 629 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
d983803d 630
e76d28dd 631 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
1f3a8e49 632 if (!mirror->start_time)
9bbd9bb4 633 mirror->start_time = now;
d0379a5d
TM		 634 if (mirror->report_interval != 0)
635 report_interval = (s64)mirror->report_interval * 1000LL;
636 else if (layoutstats_timer != 0)
bbf58bf3 637 report_interval = (s64)layoutstats_timer * 1000LL;
1c8d477a 638 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
bbf58bf3 639 report_interval) {
1c8d477a 640 ffl->last_report_time = now;
97ba375b
PT		 641 return true;
642 }
643
644 return false;
abcb7bfc
TM		 645}
646
647static void
648nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
649 __u64 requested)
650{
651 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
652
653 iostat->ops_requested++;
654 iostat->bytes_requested += requested;
655}
656
657static void
658nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
659 __u64 requested,
660 __u64 completed,
e76d28dd
TM		 661 ktime_t time_completed,
662 ktime_t time_started)
abcb7bfc
TM		 663{
664 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
e76d28dd 665 ktime_t completion_time = ktime_sub(time_completed, time_started);
abcb7bfc
TM		 666 ktime_t timer;
667
668 iostat->ops_completed++;
669 iostat->bytes_completed += completed;
670 iostat->bytes_not_delivered += requested - completed;
671
e76d28dd 672 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
abcb7bfc
TM		 673 iostat->total_busy_time =
674 ktime_add(iostat->total_busy_time, timer);
675 iostat->aggregate_completion_time =
e76d28dd
TM		 676 ktime_add(iostat->aggregate_completion_time,
677 completion_time);
abcb7bfc
TM		 678}
679
680static void
0b7baf94
TM		 681nfs4_ff_layout_stat_io_start_read(struct inode *inode,
682 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 683 __u64 requested, ktime_t now)
abcb7bfc 684{
97ba375b
PT		 685 bool report;
686
abcb7bfc 687 spin_lock(&mirror->lock);
e76d28dd 688 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
abcb7bfc 689 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
06946c6a 690 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 691 spin_unlock(&mirror->lock);
97ba375b
PT		 692
693 if (report)
0b7baf94 694 pnfs_report_layoutstat(inode, GFP_KERNEL);
abcb7bfc
TM		 695}
696
697static void
698nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
699 struct nfs4_ff_layout_mirror *mirror,
700 __u64 requested,
701 __u64 completed)
702{
703 spin_lock(&mirror->lock);
704 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
705 requested, completed,
e76d28dd 706 ktime_get(), task->tk_start);
06946c6a 707 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 708 spin_unlock(&mirror->lock);
709}
710
711static void
0b7baf94
TM		 712nfs4_ff_layout_stat_io_start_write(struct inode *inode,
713 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 714 __u64 requested, ktime_t now)
abcb7bfc 715{
97ba375b
PT		 716 bool report;
717
abcb7bfc 718 spin_lock(&mirror->lock);
e76d28dd 719 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
abcb7bfc 720 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
06946c6a 721 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 722 spin_unlock(&mirror->lock);
97ba375b
PT		 723
724 if (report)
0b7baf94 725 pnfs_report_layoutstat(inode, GFP_NOIO);
abcb7bfc
TM		 726}
727
728static void
729nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
730 struct nfs4_ff_layout_mirror *mirror,
731 __u64 requested,
732 __u64 completed,
733 enum nfs3_stable_how committed)
734{
735 if (committed == NFS_UNSTABLE)
736 requested = completed = 0;
737
738 spin_lock(&mirror->lock);
739 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
e76d28dd 740 requested, completed, ktime_get(), task->tk_start);
06946c6a 741 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 742 spin_unlock(&mirror->lock);
743}
744
d67ae825
TH		 745static int
746ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
747 struct nfs_commit_info *cinfo,
748 gfp_t gfp_flags)
749{
750 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
751 struct pnfs_commit_bucket *buckets;
752 int size;
753
754 if (cinfo->ds->nbuckets != 0) {
755 /* This assumes there is only one RW lseg per file.
756 * To support multiple lseg per file, we need to
757 * change struct pnfs_commit_bucket to allow dynamic
758 * increasing nbuckets.
759 */
760 return 0;
761 }
762
763 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
764
765 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
766 gfp_flags);
767 if (!buckets)
768 return -ENOMEM;
769 else {
770 int i;
771
fe238e60 772 spin_lock(&cinfo->inode->i_lock);
d67ae825
TH		 773 if (cinfo->ds->nbuckets != 0)
774 kfree(buckets);
775 else {
776 cinfo->ds->buckets = buckets;
777 cinfo->ds->nbuckets = size;
778 for (i = 0; i < size; i++) {
779 INIT_LIST_HEAD(&buckets[i].written);
780 INIT_LIST_HEAD(&buckets[i].committing);
781 /* mark direct verifier as unset */
782 buckets[i].direct_verf.committed =
783 NFS_INVALID_STABLE_HOW;
784 }
785 }
fe238e60 786 spin_unlock(&cinfo->inode->i_lock);
d67ae825
TH		 787 return 0;
788 }
789}
790
76c66905
TM		 791static void
792ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
793{
794 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
795
796 if (devid)
797 nfs4_mark_deviceid_unavailable(devid);
798}
799
800static void
801ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
802{
803 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
804
805 if (devid)
806 nfs4_mark_deviceid_available(devid);
807}
808
d67ae825 809static struct nfs4_pnfs_ds *
76c66905
TM		 810ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
811 int start_idx, int *best_idx,
812 bool check_device)
d67ae825 813{
13544412 814 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
2444ff27 815 struct nfs4_ff_layout_mirror *mirror;
d67ae825 816 struct nfs4_pnfs_ds *ds;
3dc14735 817 bool fail_return = false;
d67ae825
TH		 818 int idx;
819
76c66905 820 /* mirrors are initially sorted by efficiency */
13544412 821 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
3dc14735 822 if (idx+1 == fls->mirror_array_cnt)
76c66905 823 fail_return = !check_device;
2444ff27
TM		 824
825 mirror = FF_LAYOUT_COMP(lseg, idx);
76c66905
TM		 826 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
827 if (!ds)
2444ff27
TM		 828 continue;
829
76c66905
TM		 830 if (check_device &&
831 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
832 continue;
833
834 *best_idx = idx;
835 return ds;
d67ae825
TH		 836 }
837
838 return NULL;
839}
840
76c66905
TM		 841static struct nfs4_pnfs_ds *
842ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
843 int start_idx, int *best_idx)
844{
845 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
846}
847
848static struct nfs4_pnfs_ds *
849ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
850 int start_idx, int *best_idx)
851{
852 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
853}
854
855static struct nfs4_pnfs_ds *
856ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
857 int start_idx, int *best_idx)
858{
859 struct nfs4_pnfs_ds *ds;
860
861 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
862 if (ds)
863 return ds;
864 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
865}
866
c7d73af2
TH		 867static void
868ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
869 struct nfs_page *req,
870 bool strict_iomode)
871{
c7d73af2
TH		 872 pnfs_put_lseg(pgio->pg_lseg);
873 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
874 req->wb_context,
875 0,
876 NFS4_MAX_UINT64,
877 IOMODE_READ,
878 strict_iomode,
879 GFP_KERNEL);
880 if (IS_ERR(pgio->pg_lseg)) {
881 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
882 pgio->pg_lseg = NULL;
883 }
c7d73af2
TH		 884}
885
d67ae825
TH		 886static void
887ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
888 struct nfs_page *req)
889{
890 struct nfs_pgio_mirror *pgm;
891 struct nfs4_ff_layout_mirror *mirror;
892 struct nfs4_pnfs_ds *ds;
893 int ds_idx;
894
3dc14735 895retry:
b3230e80 896 pnfs_generic_pg_check_layout(pgio);
d67ae825 897 /* Use full layout for now */
1c1aeaf1 898 if (!pgio->pg_lseg) {
c7d73af2 899 ff_layout_pg_get_read(pgio, req, false);
1c1aeaf1
TM		 900 if (!pgio->pg_lseg)
901 goto out_nolseg;
902 }
903 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
c7d73af2 904 ff_layout_pg_get_read(pgio, req, true);
1c1aeaf1
TM		 905 if (!pgio->pg_lseg)
906 goto out_nolseg;
907 }
d67ae825 908
13544412 909 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
446ca219 910 if (!ds) {
3dc14735 911 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 912 goto out_mds;
3dc14735
TM		 913 pnfs_put_lseg(pgio->pg_lseg);
914 pgio->pg_lseg = NULL;
915 /* Sleep for 1 second before retrying */
916 ssleep(1);
917 goto retry;
446ca219
TH		 918 }
919
d67ae825
TH		 920 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
921
922 pgio->pg_mirror_idx = ds_idx;
923
924 /* read always uses only one mirror - idx 0 for pgio layer */
925 pgm = &pgio->pg_mirrors[0];
926 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
927
928 return;
1c1aeaf1
TM		 929out_nolseg:
930 if (pgio->pg_error < 0)
931 return;
d67ae825
TH		 932out_mds:
933 pnfs_put_lseg(pgio->pg_lseg);
934 pgio->pg_lseg = NULL;
935 nfs_pageio_reset_read_mds(pgio);
936}
937
938static void
939ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
940 struct nfs_page *req)
941{
942 struct nfs4_ff_layout_mirror *mirror;
943 struct nfs_pgio_mirror *pgm;
944 struct nfs_commit_info cinfo;
945 struct nfs4_pnfs_ds *ds;
946 int i;
947 int status;
948
3dc14735 949retry:
b3230e80 950 pnfs_generic_pg_check_layout(pgio);
d600ad1f 951 if (!pgio->pg_lseg) {
d67ae825
TH		 952 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
953 req->wb_context,
954 0,
955 NFS4_MAX_UINT64,
956 IOMODE_RW,
c7d73af2 957 false,
d67ae825 958 GFP_NOFS);
d600ad1f
PT		 959 if (IS_ERR(pgio->pg_lseg)) {
960 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
961 pgio->pg_lseg = NULL;
962 return;
963 }
964 }
d67ae825
TH		 965 /* If no lseg, fall back to write through mds */
966 if (pgio->pg_lseg == NULL)
967 goto out_mds;
968
969 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
970 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
971 if (status < 0)
972 goto out_mds;
973
974 /* Use a direct mapping of ds_idx to pgio mirror_idx */
975 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
976 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
977 goto out_mds;
978
979 for (i = 0; i < pgio->pg_mirror_count; i++) {
2444ff27 980 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
2444ff27 981 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
446ca219 982 if (!ds) {
3dc14735 983 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 984 goto out_mds;
3dc14735
TM		 985 pnfs_put_lseg(pgio->pg_lseg);
986 pgio->pg_lseg = NULL;
987 /* Sleep for 1 second before retrying */
988 ssleep(1);
989 goto retry;
446ca219 990 }
d67ae825 991 pgm = &pgio->pg_mirrors[i];
d67ae825
TH		 992 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
993 }
994
995 return;
996
997out_mds:
998 pnfs_put_lseg(pgio->pg_lseg);
999 pgio->pg_lseg = NULL;
1000 nfs_pageio_reset_write_mds(pgio);
1001}
1002
1003static unsigned int
1004ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1005 struct nfs_page *req)
1006{
d600ad1f 1007 if (!pgio->pg_lseg) {
d67ae825
TH		 1008 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1009 req->wb_context,
1010 0,
1011 NFS4_MAX_UINT64,
1012 IOMODE_RW,
c7d73af2 1013 false,
d67ae825 1014 GFP_NOFS);
d600ad1f
PT		 1015 if (IS_ERR(pgio->pg_lseg)) {
1016 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1017 pgio->pg_lseg = NULL;
1018 goto out;
1019 }
1020 }
d67ae825
TH		 1021 if (pgio->pg_lseg)
1022 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1023
1024 /* no lseg means that pnfs is not in use, so no mirroring here */
d67ae825 1025 nfs_pageio_reset_write_mds(pgio);
d600ad1f 1026out:
d67ae825
TH		 1027 return 1;
1028}
1029
1030static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1031 .pg_init = ff_layout_pg_init_read,
1032 .pg_test = pnfs_generic_pg_test,
1033 .pg_doio = pnfs_generic_pg_readpages,
1034 .pg_cleanup = pnfs_generic_pg_cleanup,
1035};
1036
1037static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1038 .pg_init = ff_layout_pg_init_write,
1039 .pg_test = pnfs_generic_pg_test,
1040 .pg_doio = pnfs_generic_pg_writepages,
1041 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1042 .pg_cleanup = pnfs_generic_pg_cleanup,
1043};
1044
1045static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1046{
1047 struct rpc_task *task = &hdr->task;
1048
1049 pnfs_layoutcommit_inode(hdr->inode, false);
1050
1051 if (retry_pnfs) {
1052 dprintk("%s Reset task %5u for i/o through pNFS "
1053 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1054 hdr->task.tk_pid,
1055 hdr->inode->i_sb->s_id,
1056 (unsigned long long)NFS_FILEID(hdr->inode),
1057 hdr->args.count,
1058 (unsigned long long)hdr->args.offset);
1059
dc602dd7 1060 hdr->completion_ops->reschedule_io(hdr);
d67ae825
TH		 1061 return;
1062 }
1063
1064 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1065 dprintk("%s Reset task %5u for i/o through MDS "
1066 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1067 hdr->task.tk_pid,
1068 hdr->inode->i_sb->s_id,
1069 (unsigned long long)NFS_FILEID(hdr->inode),
1070 hdr->args.count,
1071 (unsigned long long)hdr->args.offset);
1072
1073 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1074 }
1075}
1076
1077static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1078{
1079 struct rpc_task *task = &hdr->task;
1080
1081 pnfs_layoutcommit_inode(hdr->inode, false);
1082
1083 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1084 dprintk("%s Reset task %5u for i/o through MDS "
1085 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1086 hdr->task.tk_pid,
1087 hdr->inode->i_sb->s_id,
1088 (unsigned long long)NFS_FILEID(hdr->inode),
1089 hdr->args.count,
1090 (unsigned long long)hdr->args.offset);
1091
1092 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1093 }
1094}
1095
1096static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1097 struct nfs4_state *state,
1098 struct nfs_client *clp,
1099 struct pnfs_layout_segment *lseg,
1100 int idx)
1101{
1102 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1103 struct inode *inode = lo->plh_inode;
d67ae825 1104 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
d67ae825
TH		 1105 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1106
d67ae825 1107 switch (task->tk_status) {
d67ae825
TH		 1108 case -NFS4ERR_BADSESSION:
1109 case -NFS4ERR_BADSLOT:
1110 case -NFS4ERR_BAD_HIGH_SLOT:
1111 case -NFS4ERR_DEADSESSION:
1112 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1113 case -NFS4ERR_SEQ_FALSE_RETRY:
1114 case -NFS4ERR_SEQ_MISORDERED:
1115 dprintk("%s ERROR %d, Reset session. Exchangeid "
1116 "flags 0x%x\n", __func__, task->tk_status,
1117 clp->cl_exchange_flags);
1118 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1119 break;
1120 case -NFS4ERR_DELAY:
1121 case -NFS4ERR_GRACE:
1122 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1123 break;
1124 case -NFS4ERR_RETRY_UNCACHED_REP:
1125 break;
a79f194a
TM		 1126 case -EAGAIN:
1127 return -NFS4ERR_RESET_TO_PNFS;
d67ae825
TH		 1128 /* Invalidate Layout errors */
1129 case -NFS4ERR_PNFS_NO_LAYOUT:
1130 case -ESTALE: /* mapped NFS4ERR_STALE */
1131 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1132 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1133 case -NFS4ERR_FHEXPIRED:
1134 case -NFS4ERR_WRONG_TYPE:
1135 dprintk("%s Invalid layout error %d\n", __func__,
1136 task->tk_status);
1137 /*
1138 * Destroy layout so new i/o will get a new layout.
1139 * Layout will not be destroyed until all current lseg
1140 * references are put. Mark layout as invalid to resend failed
1141 * i/o and all i/o waiting on the slot table to the MDS until
1142 * layout is destroyed and a new valid layout is obtained.
1143 */
1144 pnfs_destroy_layout(NFS_I(inode));
1145 rpc_wake_up(&tbl->slot_tbl_waitq);
1146 goto reset;
1147 /* RPC connection errors */
1148 case -ECONNREFUSED:
1149 case -EHOSTDOWN:
1150 case -EHOSTUNREACH:
1151 case -ENETUNREACH:
1152 case -EIO:
1153 case -ETIMEDOUT:
1154 case -EPIPE:
1155 dprintk("%s DS connection error %d\n", __func__,
1156 task->tk_status);
1c48cee8
WAA		 1157 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1158 &devid->deviceid);
d67ae825
TH		 1159 rpc_wake_up(&tbl->slot_tbl_waitq);
1160 /* fall through */
1161 default:
3b13b4b3 1162 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1163 return -NFS4ERR_RESET_TO_PNFS;
1164reset:
1165 dprintk("%s Retry through MDS. Error %d\n", __func__,
1166 task->tk_status);
1167 return -NFS4ERR_RESET_TO_MDS;
1168 }
d67ae825
TH		 1169 task->tk_status = 0;
1170 return -EAGAIN;
d67ae825
TH		 1171}
1172
1173/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1174static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1175 struct pnfs_layout_segment *lseg,
1176 int idx)
1177{
1178 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1179
90816d1d
TM		 1180 switch (task->tk_status) {
1181 /* File access problems. Don't mark the device as unavailable */
1182 case -EACCES:
1183 case -ESTALE:
1184 case -EISDIR:
1185 case -EBADHANDLE:
1186 case -ELOOP:
1187 case -ENOSPC:
a79f194a 1188 case -EAGAIN:
90816d1d
TM		 1189 break;
1190 case -EJUKEBOX:
1191 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1192 goto out_retry;
1193 default:
d67ae825
TH		 1194 dprintk("%s DS connection error %d\n", __func__,
1195 task->tk_status);
1c48cee8
WAA		 1196 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1197 &devid->deviceid);
d67ae825 1198 }
90816d1d
TM		 1199 /* FIXME: Need to prevent infinite looping here. */
1200 return -NFS4ERR_RESET_TO_PNFS;
1201out_retry:
d67ae825 1202 task->tk_status = 0;
7eeea167 1203 rpc_restart_call_prepare(task);
d67ae825
TH		 1204 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1205 return -EAGAIN;
1206}
1207
1208static int ff_layout_async_handle_error(struct rpc_task *task,
1209 struct nfs4_state *state,
1210 struct nfs_client *clp,
1211 struct pnfs_layout_segment *lseg,
1212 int idx)
1213{
1214 int vers = clp->cl_nfs_mod->rpc_vers->number;
1215
76c66905
TM		 1216 if (task->tk_status >= 0) {
1217 ff_layout_mark_ds_reachable(lseg, idx);
df3ab232 1218 return 0;
76c66905 1219 }
df3ab232
TM		 1220
1221 /* Handle the case of an invalid layout segment */
1222 if (!pnfs_is_valid_lseg(lseg))
1223 return -NFS4ERR_RESET_TO_PNFS;
1224
d67ae825
TH		 1225 switch (vers) {
1226 case 3:
1227 return ff_layout_async_handle_error_v3(task, lseg, idx);
1228 case 4:
1229 return ff_layout_async_handle_error_v4(task, state, clp,
1230 lseg, idx);
1231 default:
1232 /* should never happen */
1233 WARN_ON_ONCE(1);
1234 return 0;
1235 }
1236}
1237
1238static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1239 int idx, u64 offset, u64 length,
dd52128a 1240 u32 status, int opnum, int error)
d67ae825
TH		 1241{
1242 struct nfs4_ff_layout_mirror *mirror;
1243 int err;
1244
dd52128a
TM		 1245 if (status == 0) {
1246 switch (error) {
1247 case -ETIMEDOUT:
1248 case -EPFNOSUPPORT:
1249 case -EPROTONOSUPPORT:
1250 case -EOPNOTSUPP:
1251 case -ECONNREFUSED:
1252 case -ECONNRESET:
1253 case -EHOSTDOWN:
1254 case -EHOSTUNREACH:
1255 case -ENETUNREACH:
1256 case -EADDRINUSE:
1257 case -ENOBUFS:
1258 case -EPIPE:
1259 case -EPERM:
1260 status = NFS4ERR_NXIO;
1261 break;
1262 case -EACCES:
1263 status = NFS4ERR_ACCESS;
1264 break;
1265 default:
1266 return;
1267 }
1268 }
1269
c22eeb86
PT		 1270 switch (status) {
1271 case NFS4ERR_DELAY:
1272 case NFS4ERR_GRACE:
1273 return;
1274 default:
1275 break;
1276 }
1277
d67ae825
TH		 1278 mirror = FF_LAYOUT_COMP(lseg, idx);
1279 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1280 mirror, offset, length, status, opnum,
1281 GFP_NOIO);
76c66905
TM		 1282 if (status == NFS4ERR_NXIO)
1283 ff_layout_mark_ds_unreachable(lseg, idx);
f95c03b2 1284 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
d67ae825
TH		 1285 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1286}
1287
1288/* NFS_PROTO call done callback routines */
d67ae825
TH		 1289static int ff_layout_read_done_cb(struct rpc_task *task,
1290 struct nfs_pgio_header *hdr)
1291{
166bd5b8 1292 int new_idx = hdr->pgio_mirror_idx;
d67ae825
TH		 1293 int err;
1294
1295 trace_nfs4_pnfs_read(hdr, task->tk_status);
dd52128a 1296 if (task->tk_status < 0)
d67ae825
TH		 1297 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1298 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1299 hdr->res.op_status, OP_READ,
1300 task->tk_status);
d67ae825
TH		 1301 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1302 hdr->ds_clp, hdr->lseg,
1303 hdr->pgio_mirror_idx);
1304
42f86b44
TM		 1305 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1306 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1307 switch (err) {
1308 case -NFS4ERR_RESET_TO_PNFS:
13544412
TM		 1309 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1310 hdr->pgio_mirror_idx + 1,
166bd5b8 1311 &new_idx))
f0922a6c 1312 goto out_layouterror;
42f86b44 1313 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
d67ae825
TH		 1314 return task->tk_status;
1315 case -NFS4ERR_RESET_TO_MDS:
42f86b44 1316 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1317 return task->tk_status;
1318 case -EAGAIN:
13544412 1319 goto out_eagain;
d67ae825
TH		 1320 }
1321
1322 return 0;
f0922a6c 1323out_layouterror:
166bd5b8 1324 ff_layout_read_record_layoutstats_done(task, hdr);
f0922a6c 1325 ff_layout_send_layouterror(hdr->lseg);
166bd5b8 1326 hdr->pgio_mirror_idx = new_idx;
13544412
TM		 1327out_eagain:
1328 rpc_restart_call_prepare(task);
1329 return -EAGAIN;
d67ae825
TH		 1330}
1331
c0f5f505
TM		 1332static bool
1333ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1334{
1335 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1336}
1337
d67ae825
TH		 1338/*
1339 * We reference the rpc_cred of the first WRITE that triggers the need for
1340 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1341 * rfc5661 is not clear about which credential should be used.
1342 *
1343 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1344 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1345 * we always send layoutcommit after DS writes.
1346 */
1347static void
bc28e1c2
TM		 1348ff_layout_set_layoutcommit(struct inode *inode,
1349 struct pnfs_layout_segment *lseg,
1350 loff_t end_offset)
d67ae825 1351{
bc28e1c2 1352 if (!ff_layout_need_layoutcommit(lseg))
c0f5f505
TM		 1353 return;
1354
bc28e1c2
TM		 1355 pnfs_set_layoutcommit(inode, lseg, end_offset);
1356 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1357 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
d67ae825
TH		 1358}
1359
37e9ed22
TM		 1360static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1361 struct nfs_pgio_header *hdr)
d67ae825 1362{
37e9ed22
TM		 1363 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1364 return;
0b7baf94 1365 nfs4_ff_layout_stat_io_start_read(hdr->inode,
abcb7bfc 1366 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1367 hdr->args.count,
1368 task->tk_start);
37e9ed22
TM		 1369}
1370
1371static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1372 struct nfs_pgio_header *hdr)
1373{
1374 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1375 return;
1376 nfs4_ff_layout_stat_io_end_read(task,
1377 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1378 hdr->args.count,
1379 hdr->res.count);
056f9ad6 1380 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1381}
abcb7bfc 1382
37e9ed22
TM		 1383static int ff_layout_read_prepare_common(struct rpc_task *task,
1384 struct nfs_pgio_header *hdr)
1385{
d67ae825
TH		 1386 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1387 rpc_exit(task, -EIO);
1388 return -EIO;
1389 }
d67ae825 1390
37e9ed22 1391 ff_layout_read_record_layoutstats_start(task, hdr);
d67ae825
TH		 1392 return 0;
1393}
1394
1395/*
1396 * Call ops for the async read/write cases
1397 * In the case of dense layouts, the offset needs to be reset to its
1398 * original value.
1399 */
1400static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1401{
1402 struct nfs_pgio_header *hdr = data;
1403
1404 if (ff_layout_read_prepare_common(task, hdr))
1405 return;
1406
1407 rpc_call_start(task);
1408}
1409
d67ae825
TH		 1410static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1411{
1412 struct nfs_pgio_header *hdr = data;
1413
6de7e12f
AS		 1414 if (nfs4_setup_sequence(hdr->ds_clp,
1415 &hdr->args.seq_args,
1416 &hdr->res.seq_res,
1417 task))
d67ae825
TH		 1418 return;
1419
bb21ce0a 1420 ff_layout_read_prepare_common(task, hdr);
d67ae825
TH		 1421}
1422
a79f194a
TM		 1423static void
1424ff_layout_io_prepare_transmit(struct rpc_task *task,
1425 void *data)
1426{
1427 struct nfs_pgio_header *hdr = data;
1428
1429 if (!pnfs_is_valid_lseg(hdr->lseg))
1430 rpc_exit(task, -EAGAIN);
1431}
1432
d67ae825
TH		 1433static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1434{
1435 struct nfs_pgio_header *hdr = data;
1436
1437 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1438
1439 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1440 task->tk_status == 0) {
1441 nfs4_sequence_done(task, &hdr->res.seq_res);
1442 return;
1443 }
1444
1445 /* Note this may cause RPC to be resent */
1446 hdr->mds_ops->rpc_call_done(task, hdr);
1447}
1448
1449static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1450{
1451 struct nfs_pgio_header *hdr = data;
1452
37e9ed22 1453 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1454 rpc_count_iostats_metrics(task,
1455 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1456}
1457
4d0ac221
TM		 1458static void ff_layout_read_release(void *data)
1459{
1460 struct nfs_pgio_header *hdr = data;
1461
1462 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1463 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1464 ff_layout_send_layouterror(hdr->lseg);
42f86b44 1465 pnfs_read_resend_pnfs(hdr);
f0922a6c 1466 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
42f86b44 1467 ff_layout_reset_read(hdr);
4d0ac221
TM		 1468 pnfs_generic_rw_release(data);
1469}
1470
1471
d67ae825
TH		 1472static int ff_layout_write_done_cb(struct rpc_task *task,
1473 struct nfs_pgio_header *hdr)
1474{
2e18d4d8 1475 loff_t end_offs = 0;
d67ae825
TH		 1476 int err;
1477
1478 trace_nfs4_pnfs_write(hdr, task->tk_status);
dd52128a 1479 if (task->tk_status < 0)
d67ae825
TH		 1480 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1481 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1482 hdr->res.op_status, OP_WRITE,
1483 task->tk_status);
d67ae825
TH		 1484 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1485 hdr->ds_clp, hdr->lseg,
1486 hdr->pgio_mirror_idx);
1487
7b0df92a
TM		 1488 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1489 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1490 switch (err) {
1491 case -NFS4ERR_RESET_TO_PNFS:
7b0df92a 1492 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
7cc8c5cd 1493 return task->tk_status;
d67ae825 1494 case -NFS4ERR_RESET_TO_MDS:
7b0df92a 1495 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
d67ae825
TH		 1496 return task->tk_status;
1497 case -EAGAIN:
d67ae825
TH		 1498 return -EAGAIN;
1499 }
1500
1501 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1502 hdr->res.verf->committed == NFS_DATA_SYNC)
2e18d4d8
TM		 1503 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1504
1505 /* Note: if the write is unstable, don't set end_offs until commit */
1506 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
d67ae825 1507
54204010
PT		 1508 /* zero out fattr since we don't care DS attr at all */
1509 hdr->fattr.valid = 0;
69f230d9
PT		 1510 if (task->tk_status >= 0)
1511 nfs_writeback_update_inode(hdr);
1512
d67ae825
TH		 1513 return 0;
1514}
1515
1516static int ff_layout_commit_done_cb(struct rpc_task *task,
1517 struct nfs_commit_data *data)
1518{
d67ae825
TH		 1519 int err;
1520
1521 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
dd52128a 1522 if (task->tk_status < 0)
d67ae825
TH		 1523 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1524 data->args.offset, data->args.count,
dd52128a
TM		 1525 data->res.op_status, OP_COMMIT,
1526 task->tk_status);
d67ae825
TH		 1527 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1528 data->lseg, data->ds_commit_index);
1529
1530 switch (err) {
1531 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1532 pnfs_generic_prepare_to_resend_writes(data);
1533 return -EAGAIN;
d67ae825 1534 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1535 pnfs_generic_prepare_to_resend_writes(data);
1536 return -EAGAIN;
1537 case -EAGAIN:
1538 rpc_restart_call_prepare(task);
1539 return -EAGAIN;
1540 }
1541
bc28e1c2 1542 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
d67ae825
TH		 1543
1544 return 0;
1545}
1546
37e9ed22
TM		 1547static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1548 struct nfs_pgio_header *hdr)
d67ae825 1549{
37e9ed22
TM		 1550 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1551 return;
0b7baf94 1552 nfs4_ff_layout_stat_io_start_write(hdr->inode,
abcb7bfc 1553 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1554 hdr->args.count,
1555 task->tk_start);
37e9ed22
TM		 1556}
1557
1558static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1559 struct nfs_pgio_header *hdr)
1560{
1561 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1562 return;
1563 nfs4_ff_layout_stat_io_end_write(task,
1564 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1565 hdr->args.count, hdr->res.count,
1566 hdr->res.verf->committed);
056f9ad6 1567 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
37e9ed22 1568}
abcb7bfc 1569
37e9ed22
TM		 1570static int ff_layout_write_prepare_common(struct rpc_task *task,
1571 struct nfs_pgio_header *hdr)
1572{
d67ae825
TH		 1573 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1574 rpc_exit(task, -EIO);
1575 return -EIO;
1576 }
1577
37e9ed22 1578 ff_layout_write_record_layoutstats_start(task, hdr);
d67ae825
TH		 1579 return 0;
1580}
1581
1582static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1583{
1584 struct nfs_pgio_header *hdr = data;
1585
1586 if (ff_layout_write_prepare_common(task, hdr))
1587 return;
1588
1589 rpc_call_start(task);
1590}
1591
1592static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1593{
1594 struct nfs_pgio_header *hdr = data;
1595
6de7e12f
AS		 1596 if (nfs4_setup_sequence(hdr->ds_clp,
1597 &hdr->args.seq_args,
1598 &hdr->res.seq_res,
1599 task))
d67ae825
TH		 1600 return;
1601
bb21ce0a 1602 ff_layout_write_prepare_common(task, hdr);
d67ae825
TH		 1603}
1604
1605static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1606{
1607 struct nfs_pgio_header *hdr = data;
1608
1609 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1610 task->tk_status == 0) {
1611 nfs4_sequence_done(task, &hdr->res.seq_res);
1612 return;
1613 }
1614
1615 /* Note this may cause RPC to be resent */
1616 hdr->mds_ops->rpc_call_done(task, hdr);
1617}
1618
1619static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1620{
1621 struct nfs_pgio_header *hdr = data;
1622
37e9ed22 1623 ff_layout_write_record_layoutstats_done(task, hdr);
d67ae825
TH		 1624 rpc_count_iostats_metrics(task,
1625 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1626}
1627
4d0ac221
TM		 1628static void ff_layout_write_release(void *data)
1629{
1630 struct nfs_pgio_header *hdr = data;
1631
1632 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
f0922a6c
TM		 1633 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1634 ff_layout_send_layouterror(hdr->lseg);
7b0df92a 1635 ff_layout_reset_write(hdr, true);
f0922a6c 1636 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
7b0df92a 1637 ff_layout_reset_write(hdr, false);
4d0ac221
TM		 1638 pnfs_generic_rw_release(data);
1639}
1640
37e9ed22 1641static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
abcb7bfc
TM		 1642 struct nfs_commit_data *cdata)
1643{
37e9ed22
TM		 1644 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1645 return;
0b7baf94 1646 nfs4_ff_layout_stat_io_start_write(cdata->inode,
abcb7bfc 1647 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
e76d28dd 1648 0, task->tk_start);
abcb7bfc
TM		 1649}
1650
37e9ed22
TM		 1651static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1652 struct nfs_commit_data *cdata)
1653{
1654 struct nfs_page *req;
1655 __u64 count = 0;
1656
1657 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1658 return;
1659
1660 if (task->tk_status == 0) {
1661 list_for_each_entry(req, &cdata->pages, wb_list)
1662 count += req->wb_bytes;
1663 }
1664 nfs4_ff_layout_stat_io_end_write(task,
1665 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1666 count, count, NFS_FILE_SYNC);
056f9ad6 1667 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
37e9ed22
TM		 1668}
1669
1670static void ff_layout_commit_prepare_common(struct rpc_task *task,
1671 struct nfs_commit_data *cdata)
1672{
1673 ff_layout_commit_record_layoutstats_start(task, cdata);
1674}
1675
d67ae825
TH		 1676static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1677{
abcb7bfc 1678 ff_layout_commit_prepare_common(task, data);
d67ae825
TH		 1679 rpc_call_start(task);
1680}
1681
1682static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1683{
1684 struct nfs_commit_data *wdata = data;
1685
6de7e12f
AS		 1686 if (nfs4_setup_sequence(wdata->ds_clp,
1687 &wdata->args.seq_args,
1688 &wdata->res.seq_res,
1689 task))
abcb7bfc
TM		 1690 return;
1691 ff_layout_commit_prepare_common(task, data);
1692}
1693
1694static void ff_layout_commit_done(struct rpc_task *task, void *data)
141b9b59
PT		 1695{
1696 pnfs_generic_write_commit_done(task, data);
1697}
1698
1699static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
abcb7bfc
TM		 1700{
1701 struct nfs_commit_data *cdata = data;
abcb7bfc 1702
37e9ed22 1703 ff_layout_commit_record_layoutstats_done(task, cdata);
d67ae825
TH		 1704 rpc_count_iostats_metrics(task,
1705 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1706}
1707
4d0ac221
TM		 1708static void ff_layout_commit_release(void *data)
1709{
1710 struct nfs_commit_data *cdata = data;
1711
1712 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1713 pnfs_generic_commit_release(data);
1714}
1715
d67ae825
TH		 1716static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1717 .rpc_call_prepare = ff_layout_read_prepare_v3,
a79f194a 1718 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
d67ae825
TH		 1719 .rpc_call_done = ff_layout_read_call_done,
1720 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1721 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1722};
1723
1724static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1725 .rpc_call_prepare = ff_layout_read_prepare_v4,
a79f194a 1726 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
d67ae825
TH		 1727 .rpc_call_done = ff_layout_read_call_done,
1728 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1729 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1730};
1731
1732static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1733 .rpc_call_prepare = ff_layout_write_prepare_v3,
a79f194a 1734 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
d67ae825
TH		 1735 .rpc_call_done = ff_layout_write_call_done,
1736 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1737 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1738};
1739
1740static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1741 .rpc_call_prepare = ff_layout_write_prepare_v4,
a79f194a 1742 .rpc_call_prepare_transmit = ff_layout_io_prepare_transmit,
d67ae825
TH		 1743 .rpc_call_done = ff_layout_write_call_done,
1744 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1745 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1746};
1747
1748static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1749 .rpc_call_prepare = ff_layout_commit_prepare_v3,
abcb7bfc 1750 .rpc_call_done = ff_layout_commit_done,
d67ae825 1751 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1752 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1753};
1754
1755static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1756 .rpc_call_prepare = ff_layout_commit_prepare_v4,
abcb7bfc 1757 .rpc_call_done = ff_layout_commit_done,
d67ae825 1758 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1759 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1760};
1761
1762static enum pnfs_try_status
1763ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1764{
1765 struct pnfs_layout_segment *lseg = hdr->lseg;
1766 struct nfs4_pnfs_ds *ds;
1767 struct rpc_clnt *ds_clnt;
2444ff27 1768 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1769 const struct cred *ds_cred;
d67ae825
TH		 1770 loff_t offset = hdr->args.offset;
1771 u32 idx = hdr->pgio_mirror_idx;
1772 int vers;
1773 struct nfs_fh *fh;
1774
5b5e0928 1775 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
d67ae825
TH		 1776 __func__, hdr->inode->i_ino,
1777 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1778
2444ff27
TM		 1779 mirror = FF_LAYOUT_COMP(lseg, idx);
1780 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
d67ae825
TH		 1781 if (!ds)
1782 goto out_failed;
1783
561d6f8a 1784 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1785 hdr->inode);
1786 if (IS_ERR(ds_clnt))
1787 goto out_failed;
1788
312cd4cb 1789 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1790 if (!ds_cred)
d67ae825
TH		 1791 goto out_failed;
1792
626d48b1 1793 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1794
1795 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
212bf41d 1796 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
d67ae825 1797
75bf47eb 1798 hdr->pgio_done_cb = ff_layout_read_done_cb;
212bf41d 1799 refcount_inc(&ds->ds_clp->cl_count);
d67ae825 1800 hdr->ds_clp = ds->ds_clp;
749da527 1801 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1802 if (fh)
1803 hdr->args.fh = fh;
bb21ce0a 1804
4cbc8a57 1805 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1806
d67ae825
TH		 1807 /*
1808 * Note that if we ever decide to split across DSes,
1809 * then we may need to handle dense-like offsets.
1810 */
1811 hdr->args.offset = offset;
1812 hdr->mds_offset = offset;
1813
1814 /* Perform an asynchronous read to ds */
1815 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1816 vers == 3 ? &ff_layout_read_call_ops_v3 :
1817 &ff_layout_read_call_ops_v4,
1818 0, RPC_TASK_SOFTCONN);
a52458b4 1819 put_cred(ds_cred);
d67ae825
TH		 1820 return PNFS_ATTEMPTED;
1821
1822out_failed:
3b13b4b3 1823 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1824 return PNFS_TRY_AGAIN;
1825 return PNFS_NOT_ATTEMPTED;
1826}
1827
1828/* Perform async writes. */
1829static enum pnfs_try_status
1830ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1831{
1832 struct pnfs_layout_segment *lseg = hdr->lseg;
1833 struct nfs4_pnfs_ds *ds;
1834 struct rpc_clnt *ds_clnt;
2444ff27 1835 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1836 const struct cred *ds_cred;
d67ae825
TH		 1837 loff_t offset = hdr->args.offset;
1838 int vers;
1839 struct nfs_fh *fh;
1840 int idx = hdr->pgio_mirror_idx;
1841
2444ff27
TM		 1842 mirror = FF_LAYOUT_COMP(lseg, idx);
1843 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825 1844 if (!ds)
37f8aa16 1845 goto out_failed;
d67ae825 1846
561d6f8a 1847 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1848 hdr->inode);
1849 if (IS_ERR(ds_clnt))
37f8aa16 1850 goto out_failed;
d67ae825 1851
312cd4cb 1852 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
57f3f4c0 1853 if (!ds_cred)
37f8aa16 1854 goto out_failed;
d67ae825 1855
626d48b1 1856 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825 1857
5b5e0928 1858 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
d67ae825 1859 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
212bf41d 1860 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1861 vers);
1862
1863 hdr->pgio_done_cb = ff_layout_write_done_cb;
212bf41d 1864 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1865 hdr->ds_clp = ds->ds_clp;
1866 hdr->ds_commit_idx = idx;
749da527 1867 fh = nfs4_ff_layout_select_ds_fh(mirror);
d67ae825
TH		 1868 if (fh)
1869 hdr->args.fh = fh;
1870
4cbc8a57 1871 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
bb21ce0a 1872
d67ae825
TH		 1873 /*
1874 * Note that if we ever decide to split across DSes,
1875 * then we may need to handle dense-like offsets.
1876 */
1877 hdr->args.offset = offset;
1878
1879 /* Perform an asynchronous write */
1880 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1881 vers == 3 ? &ff_layout_write_call_ops_v3 :
1882 &ff_layout_write_call_ops_v4,
1883 sync, RPC_TASK_SOFTCONN);
a52458b4 1884 put_cred(ds_cred);
d67ae825 1885 return PNFS_ATTEMPTED;
37f8aa16
TM		 1886
1887out_failed:
1888 if (ff_layout_avoid_mds_available_ds(lseg))
1889 return PNFS_TRY_AGAIN;
1890 return PNFS_NOT_ATTEMPTED;
d67ae825
TH		 1891}
1892
d67ae825
TH		 1893static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1894{
1895 return i;
1896}
1897
1898static struct nfs_fh *
1899select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1900{
1901 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1902
1903 /* FIXME: Assume that there is only one NFS version available
1904 * for the DS.
1905 */
1906 return &flseg->mirror_array[i]->fh_versions[0];
1907}
1908
1909static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1910{
1911 struct pnfs_layout_segment *lseg = data->lseg;
1912 struct nfs4_pnfs_ds *ds;
1913 struct rpc_clnt *ds_clnt;
2444ff27 1914 struct nfs4_ff_layout_mirror *mirror;
a52458b4 1915 const struct cred *ds_cred;
d67ae825 1916 u32 idx;
57f3f4c0 1917 int vers, ret;
d67ae825
TH		 1918 struct nfs_fh *fh;
1919
4b75053e
TM		 1920 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1921 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1922 goto out_err;
1923
d67ae825 1924 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
2444ff27
TM		 1925 mirror = FF_LAYOUT_COMP(lseg, idx);
1926 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
d67ae825
TH		 1927 if (!ds)
1928 goto out_err;
1929
561d6f8a 1930 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
d67ae825
TH		 1931 data->inode);
1932 if (IS_ERR(ds_clnt))
1933 goto out_err;
1934
312cd4cb 1935 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
57f3f4c0 1936 if (!ds_cred)
d67ae825
TH		 1937 goto out_err;
1938
626d48b1 1939 vers = nfs4_ff_layout_ds_version(mirror);
d67ae825
TH		 1940
1941 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
212bf41d 1942 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
d67ae825
TH		 1943 vers);
1944 data->commit_done_cb = ff_layout_commit_done_cb;
1945 data->cred = ds_cred;
212bf41d 1946 refcount_inc(&ds->ds_clp->cl_count);
d67ae825
TH		 1947 data->ds_clp = ds->ds_clp;
1948 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1949 if (fh)
1950 data->args.fh = fh;
abcb7bfc 1951
57f3f4c0 1952 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
d67ae825
TH		 1953 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1954 &ff_layout_commit_call_ops_v4,
1955 how, RPC_TASK_SOFTCONN);
a52458b4 1956 put_cred(ds_cred);
57f3f4c0 1957 return ret;
d67ae825
TH		 1958out_err:
1959 pnfs_generic_prepare_to_resend_writes(data);
1960 pnfs_generic_commit_release(data);
1961 return -EAGAIN;
1962}
1963
1964static int
1965ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1966 int how, struct nfs_commit_info *cinfo)
1967{
1968 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1969 ff_layout_initiate_commit);
1970}
1971
1972static struct pnfs_ds_commit_info *
1973ff_layout_get_ds_info(struct inode *inode)
1974{
1975 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1976
1977 if (layout == NULL)
1978 return NULL;
1979
1980 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1981}
1982
1983static void
fc87701b 1984ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
d67ae825
TH		 1985{
1986 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1987 id_node));
1988}
1989
5ba6a09e 1990static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
5b9b3c85
TM		 1991 const struct nfs4_layoutreturn_args *args,
1992 const struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825 1993{
d67ae825 1994 __be32 *start;
d67ae825
TH		 1995
1996 start = xdr_reserve_space(xdr, 4);
1997 if (unlikely(!start))
1998 return -E2BIG;
1999
5b9b3c85 2000 *start = cpu_to_be32(ff_args->num_errors);
d67ae825 2001 /* This assume we always return _ALL_ layouts */
5b9b3c85 2002 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
d67ae825
TH		 2003}
2004
230bc962
TM		 2005static void
2006encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2007{
ab6e9aaf 2008 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
230bc962
TM		 2009}
2010
2011static void
2012ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2013 const nfs4_stateid *stateid,
2014 const struct nfs42_layoutstat_devinfo *devinfo)
2015{
2016 __be32 *p;
2017
2018 p = xdr_reserve_space(xdr, 8 + 8);
2019 p = xdr_encode_hyper(p, devinfo->offset);
2020 p = xdr_encode_hyper(p, devinfo->length);
2021 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2022 p = xdr_reserve_space(xdr, 4*8);
2023 p = xdr_encode_hyper(p, devinfo->read_count);
2024 p = xdr_encode_hyper(p, devinfo->read_bytes);
2025 p = xdr_encode_hyper(p, devinfo->write_count);
2026 p = xdr_encode_hyper(p, devinfo->write_bytes);
2027 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2028}
2029
2030static void
2031ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2032 const nfs4_stateid *stateid,
2033 const struct nfs42_layoutstat_devinfo *devinfo)
2034{
2035 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2036 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2037 devinfo->ld_private.data);
2038}
2039
d67ae825 2040/* report nothing for now */
230bc962
TM		 2041static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2042 const struct nfs4_layoutreturn_args *args,
2043 struct nfs4_flexfile_layoutreturn_args *ff_args)
d67ae825
TH		 2044{
2045 __be32 *p;
230bc962 2046 int i;
d67ae825
TH		 2047
2048 p = xdr_reserve_space(xdr, 4);
230bc962
TM		 2049 *p = cpu_to_be32(ff_args->num_dev);
2050 for (i = 0; i < ff_args->num_dev; i++)
2051 ff_layout_encode_ff_iostat(xdr,
2052 &args->layout->plh_stateid,
2053 &ff_args->devinfo[i]);
2054}
2055
2056static void
2057ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2058 unsigned int num_entries)
2059{
2060 unsigned int i;
2061
2062 for (i = 0; i < num_entries; i++) {
2063 if (!devinfo[i].ld_private.ops)
2064 continue;
2065 if (!devinfo[i].ld_private.ops->free)
2066 continue;
2067 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2068 }
d67ae825
TH		 2069}
2070
2071static struct nfs4_deviceid_node *
2072ff_layout_alloc_deviceid_node(struct nfs_server *server,
2073 struct pnfs_device *pdev, gfp_t gfp_flags)
2074{
2075 struct nfs4_ff_layout_ds *dsaddr;
2076
2077 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2078 if (!dsaddr)
2079 return NULL;
2080 return &dsaddr->id_node;
2081}
2082
2083static void
94e5c571 2084ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
5b9b3c85
TM		 2085 const void *voidargs,
2086 const struct nfs4_xdr_opaque_data *ff_opaque)
d67ae825 2087{
5b9b3c85 2088 const struct nfs4_layoutreturn_args *args = voidargs;
5ba6a09e 2089 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
d9152114
TM		 2090 struct xdr_buf tmp_buf = {
2091 .head = {
2092 [0] = {
2093 .iov_base = page_address(ff_args->pages[0]),
2094 },
2095 },
2096 .buflen = PAGE_SIZE,
2097 };
2098 struct xdr_stream tmp_xdr;
d67ae825
TH		 2099 __be32 *start;
2100
2101 dprintk("%s: Begin\n", __func__);
d67ae825 2102
0ccc61b1 2103 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
d9152114
TM		 2104
2105 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2106 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2107
2108 start = xdr_reserve_space(xdr, 4);
2109 *start = cpu_to_be32(tmp_buf.len);
2110 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
082fa37d 2111
d67ae825
TH		 2112 dprintk("%s: Return\n", __func__);
2113}
2114
5b9b3c85
TM		 2115static void
2116ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2117{
2118 struct nfs4_flexfile_layoutreturn_args *ff_args;
2119
2120 if (!args->data)
2121 return;
2122 ff_args = args->data;
2123 args->data = NULL;
2124
2125 ff_layout_free_ds_ioerr(&ff_args->errors);
230bc962 2126 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
5b9b3c85 2127
d9152114 2128 put_page(ff_args->pages[0]);
5b9b3c85
TM		 2129 kfree(ff_args);
2130}
2131
2e54b9b1 2132static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
5b9b3c85
TM		 2133 .encode = ff_layout_encode_layoutreturn,
2134 .free = ff_layout_free_layoutreturn,
2135};
2136
2137static int
2138ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2139{
2140 struct nfs4_flexfile_layoutreturn_args *ff_args;
230bc962 2141 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
5b9b3c85
TM		 2142
2143 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2144 if (!ff_args)
d9152114
TM		 2145 goto out_nomem;
2146 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2147 if (!ff_args->pages[0])
2148 goto out_nomem_free;
5b9b3c85
TM		 2149
2150 INIT_LIST_HEAD(&ff_args->errors);
2151 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2152 &args->range, &ff_args->errors,
2153 FF_LAYOUTRETURN_MAXERR);
2154
230bc962
TM		 2155 spin_lock(&args->inode->i_lock);
2156 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2157 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2158 spin_unlock(&args->inode->i_lock);
2159
5b9b3c85
TM		 2160 args->ld_private->ops = &layoutreturn_ops;
2161 args->ld_private->data = ff_args;
2162 return 0;
d9152114
TM		 2163out_nomem_free:
2164 kfree(ff_args);
2165out_nomem:
2166 return -ENOMEM;
5b9b3c85
TM		 2167}
2168
f0922a6c
TM		 2169#ifdef CONFIG_NFS_V4_2
2170void
2171ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2172{
2173 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2174 struct nfs42_layout_error *errors;
2175 LIST_HEAD(head);
2176
2177 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2178 return;
2179 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2180 if (list_empty(&head))
2181 return;
2182
2183 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2184 sizeof(*errors), GFP_NOFS);
2185 if (errors != NULL) {
2186 const struct nfs4_ff_layout_ds_err *pos;
2187 size_t n = 0;
2188
2189 list_for_each_entry(pos, &head, list) {
2190 errors[n].offset = pos->offset;
2191 errors[n].length = pos->length;
2192 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2193 errors[n].errors[0].dev_id = pos->deviceid;
2194 errors[n].errors[0].status = pos->status;
2195 errors[n].errors[0].opnum = pos->opnum;
2196 n++;
2197 if (!list_is_last(&pos->list, &head) &&
2198 n < NFS42_LAYOUTERROR_MAX)
2199 continue;
2200 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2201 break;
2202 n = 0;
2203 }
2204 kfree(errors);
2205 }
2206 ff_layout_free_ds_ioerr(&head);
2207}
2208#else
2209void
2210ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2211{
2212}
2213#endif
2214
27c43064
PT		 2215static int
2216ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2217{
2218 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2219
2220 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2221}
2222
2223static size_t
2224ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2225 const int buflen)
2226{
2227 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2228 const struct in6_addr *addr = &sin6->sin6_addr;
2229
2230 /*
2231 * RFC 4291, Section 2.2.2
2232 *
2233 * Shorthanded ANY address
2234 */
2235 if (ipv6_addr_any(addr))
2236 return snprintf(buf, buflen, "::");
2237
2238 /*
2239 * RFC 4291, Section 2.2.2
2240 *
2241 * Shorthanded loopback address
2242 */
2243 if (ipv6_addr_loopback(addr))
2244 return snprintf(buf, buflen, "::1");
2245
2246 /*
2247 * RFC 4291, Section 2.2.3
2248 *
2249 * Special presentation address format for mapped v4
2250 * addresses.
2251 */
2252 if (ipv6_addr_v4mapped(addr))
2253 return snprintf(buf, buflen, "::ffff:%pI4",
2254 &addr->s6_addr32[3]);
2255
2256 /*
2257 * RFC 4291, Section 2.2.1
2258 */
2259 return snprintf(buf, buflen, "%pI6c", addr);
2260}
2261
2262/* Derived from rpc_sockaddr2uaddr */
2263static void
2264ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2265{
2266 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2267 char portbuf[RPCBIND_MAXUADDRPLEN];
2268 char addrbuf[RPCBIND_MAXUADDRLEN];
2269 char *netid;
2270 unsigned short port;
2271 int len, netid_len;
2272 __be32 *p;
2273
2274 switch (sap->sa_family) {
2275 case AF_INET:
2276 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2277 return;
2278 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2279 netid = "tcp";
2280 netid_len = 3;
2281 break;
2282 case AF_INET6:
2283 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2284 return;
2285 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2286 netid = "tcp6";
2287 netid_len = 4;
2288 break;
2289 default:
2290 /* we only support tcp and tcp6 */
2291 WARN_ON_ONCE(1);
2292 return;
2293 }
2294
2295 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2296 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2297
2298 p = xdr_reserve_space(xdr, 4 + netid_len);
2299 xdr_encode_opaque(p, netid, netid_len);
2300
2301 p = xdr_reserve_space(xdr, 4 + len);
2302 xdr_encode_opaque(p, addrbuf, len);
2303}
2304
2305static void
2306ff_layout_encode_nfstime(struct xdr_stream *xdr,
2307 ktime_t t)
2308{
2309 struct timespec64 ts;
2310 __be32 *p;
2311
2312 p = xdr_reserve_space(xdr, 12);
2313 ts = ktime_to_timespec64(t);
2314 p = xdr_encode_hyper(p, ts.tv_sec);
2315 *p++ = cpu_to_be32(ts.tv_nsec);
2316}
2317
2318static void
2319ff_layout_encode_io_latency(struct xdr_stream *xdr,
2320 struct nfs4_ff_io_stat *stat)
2321{
2322 __be32 *p;
2323
2324 p = xdr_reserve_space(xdr, 5 * 8);
2325 p = xdr_encode_hyper(p, stat->ops_requested);
2326 p = xdr_encode_hyper(p, stat->bytes_requested);
2327 p = xdr_encode_hyper(p, stat->ops_completed);
2328 p = xdr_encode_hyper(p, stat->bytes_completed);
2329 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2330 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2331 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2332}
2333
2334static void
422c93c8
TM		 2335ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2336 const struct nfs42_layoutstat_devinfo *devinfo,
2337 struct nfs4_ff_layout_mirror *mirror)
27c43064 2338{
27c43064
PT		 2339 struct nfs4_pnfs_ds_addr *da;
2340 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2341 struct nfs_fh *fh = &mirror->fh_versions[0];
422c93c8 2342 __be32 *p;
27c43064
PT		 2343
2344 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2345 dprintk("%s: DS %s: encoding address %s\n",
2346 __func__, ds->ds_remotestr, da->da_remotestr);
27c43064
PT		 2347 /* netaddr4 */
2348 ff_layout_encode_netaddr(xdr, da);
2349 /* nfs_fh4 */
2350 p = xdr_reserve_space(xdr, 4 + fh->size);
2351 xdr_encode_opaque(p, fh->data, fh->size);
2352 /* ff_io_latency4 read */
2353 spin_lock(&mirror->lock);
2354 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2355 /* ff_io_latency4 write */
2356 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2357 spin_unlock(&mirror->lock);
2358 /* nfstime4 */
2359 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2360 /* bool */
2361 p = xdr_reserve_space(xdr, 4);
2362 *p = cpu_to_be32(false);
422c93c8
TM		 2363}
2364
2365static void
2366ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2367 const struct nfs4_xdr_opaque_data *opaque)
2368{
2369 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2370 struct nfs42_layoutstat_devinfo, ld_private);
2371 __be32 *start;
2372
2373 /* layoutupdate length */
2374 start = xdr_reserve_space(xdr, 4);
2375 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
27c43064
PT		 2376
2377 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2378}
2379
422c93c8
TM		 2380static void
2381ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2382{
2383 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2384
2385 ff_layout_put_mirror(mirror);
2386}
2387
2388static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2389 .encode = ff_layout_encode_layoutstats,
2390 .free = ff_layout_free_layoutstats,
2391};
2392
266d12d4 2393static int
08e2e5bc
TM		 2394ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2395 struct nfs42_layoutstat_devinfo *devinfo,
266d12d4 2396 int dev_limit)
ad4dc53e 2397{
266d12d4 2398 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
ad4dc53e
PT		 2399 struct nfs4_ff_layout_mirror *mirror;
2400 struct nfs4_deviceid_node *dev;
266d12d4 2401 int i = 0;
ad4dc53e 2402
266d12d4
TM		 2403 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2404 if (i >= dev_limit)
ad4dc53e 2405 break;
65990d1a 2406 if (IS_ERR_OR_NULL(mirror->mirror_ds))
27c43064 2407 continue;
06946c6a
TM		 2408 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2409 continue;
266d12d4 2410 /* mirror refcount put in cleanup_layoutstats */
81a090b9 2411 if (!refcount_inc_not_zero(&mirror->ref))
266d12d4
TM		 2412 continue;
2413 dev = &mirror->mirror_ds->id_node;
ad4dc53e 2414 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
266d12d4
TM		 2415 devinfo->offset = 0;
2416 devinfo->length = NFS4_MAX_UINT64;
2f8220c1 2417 spin_lock(&mirror->lock);
d099d7b8 2418 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
ad4dc53e 2419 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
d099d7b8 2420 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
ad4dc53e 2421 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2f8220c1 2422 spin_unlock(&mirror->lock);
ad4dc53e 2423 devinfo->layout_type = LAYOUT_FLEX_FILES;
422c93c8
TM		 2424 devinfo->ld_private.ops = &layoutstat_ops;
2425 devinfo->ld_private.data = mirror;
ad4dc53e 2426
08e2e5bc 2427 devinfo++;
266d12d4 2428 i++;
ad4dc53e 2429 }
266d12d4 2430 return i;
ad4dc53e
PT		 2431}
2432
2433static int
2434ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2435{
266d12d4 2436 struct nfs4_flexfile_layout *ff_layout;
08e2e5bc 2437 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
ad4dc53e 2438
ad4dc53e 2439 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
266d12d4 2440 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
ad4dc53e
PT		 2441 if (!args->devinfo)
2442 return -ENOMEM;
2443
ad4dc53e 2444 spin_lock(&args->inode->i_lock);
08e2e5bc
TM		 2445 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2446 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2447 &args->devinfo[0], dev_count);
ad4dc53e 2448 spin_unlock(&args->inode->i_lock);
46c98c6d
TM		 2449 if (!args->num_dev) {
2450 kfree(args->devinfo);
2451 args->devinfo = NULL;
2452 return -ENOENT;
2453 }
ad4dc53e
PT		 2454
2455 return 0;
2456}
2457
76b2a303
TM		 2458static int
2459ff_layout_set_layoutdriver(struct nfs_server *server,
2460 const struct nfs_fh *dummy)
2461{
2462#if IS_ENABLED(CONFIG_NFS_V4_2)
2463 server->caps |= NFS_CAP_LAYOUTSTATS;
2464#endif
2465 return 0;
2466}
2467
d67ae825
TH		 2468static struct pnfs_layoutdriver_type flexfilelayout_type = {
2469 .id = LAYOUT_FLEX_FILES,
2470 .name = "LAYOUT_FLEX_FILES",
2471 .owner = THIS_MODULE,
f86c3ac5 2472 .flags = PNFS_LAYOUTGET_ON_OPEN,
28ced9a8 2473 .max_layoutget_response = 4096, /* 1 page or so... */
76b2a303 2474 .set_layoutdriver = ff_layout_set_layoutdriver,
d67ae825
TH		 2475 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2476 .free_layout_hdr = ff_layout_free_layout_hdr,
2477 .alloc_lseg = ff_layout_alloc_lseg,
2478 .free_lseg = ff_layout_free_lseg,
0762ed2c 2479 .add_lseg = ff_layout_add_lseg,
d67ae825
TH		 2480 .pg_read_ops = &ff_layout_pg_read_ops,
2481 .pg_write_ops = &ff_layout_pg_write_ops,
2482 .get_ds_info = ff_layout_get_ds_info,
fc87701b 2483 .free_deviceid_node = ff_layout_free_deviceid_node,
338d00cf 2484 .mark_request_commit = pnfs_layout_mark_request_commit,
d67ae825
TH		 2485 .clear_request_commit = pnfs_generic_clear_request_commit,
2486 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2487 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2488 .commit_pagelist = ff_layout_commit_pagelist,
2489 .read_pagelist = ff_layout_read_pagelist,
2490 .write_pagelist = ff_layout_write_pagelist,
2491 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
5b9b3c85 2492 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
5bb89b47 2493 .sync = pnfs_nfs_generic_sync,
ad4dc53e 2494 .prepare_layoutstats = ff_layout_prepare_layoutstats,
d67ae825
TH		 2495};
2496
2497static int __init nfs4flexfilelayout_init(void)
2498{
2499 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2500 __func__);
2501 return pnfs_register_layoutdriver(&flexfilelayout_type);
2502}
2503
2504static void __exit nfs4flexfilelayout_exit(void)
2505{
2506 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2507 __func__);
2508 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2509}
2510
2511MODULE_ALIAS("nfs-layouttype4-4");
2512
2513MODULE_LICENSE("GPL");
2514MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2515
2516module_init(nfs4flexfilelayout_init);
2517module_exit(nfs4flexfilelayout_exit);
Linux 6.x block layer and io_uring tree(s)