]> 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
pNFS: Add a layoutreturn callback to performa layout-private setup
[mirror_ubuntu-jammy-kernel.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>
12
13#include <linux/sunrpc/metrics.h>
d67ae825
TH		 14
15#include "flexfilelayout.h"
16#include "../nfs4session.h"
40c64c26 17#include "../nfs4idmap.h"
d67ae825
TH		 18#include "../internal.h"
19#include "../delegation.h"
20#include "../nfs4trace.h"
21#include "../iostat.h"
22#include "../nfs.h"
ad4dc53e 23#include "../nfs42.h"
d67ae825
TH		 24
25#define NFSDBG_FACILITY NFSDBG_PNFS_LD
26
27#define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
28
90a0be00
JL		 29static struct group_info *ff_zero_group;
30
54e4a0df
TM		 31static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
32 struct nfs_pgio_header *hdr);
33
d67ae825
TH		 34static struct pnfs_layout_hdr *
35ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
36{
37 struct nfs4_flexfile_layout *ffl;
38
39 ffl = kzalloc(sizeof(*ffl), gfp_flags);
40 if (ffl) {
41 INIT_LIST_HEAD(&ffl->error_list);
266d12d4 42 INIT_LIST_HEAD(&ffl->mirrors);
1c8d477a 43 ffl->last_report_time = ktime_get();
d67ae825
TH		 44 return &ffl->generic_hdr;
45 } else
46 return NULL;
47}
48
49static void
50ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
51{
52 struct nfs4_ff_layout_ds_err *err, *n;
53
54 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
55 list) {
56 list_del(&err->list);
57 kfree(err);
58 }
59 kfree(FF_LAYOUT_FROM_HDR(lo));
60}
61
93b717fd 62static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
d67ae825
TH		 63{
64 __be32 *p;
65
66 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
67 if (unlikely(p == NULL))
68 return -ENOBUFS;
93b717fd
TM		 69 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
70 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
d67ae825
TH		 71 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
72 p[0], p[1], p[2], p[3]);
73 return 0;
74}
75
76static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
77{
78 __be32 *p;
79
80 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
81 if (unlikely(!p))
82 return -ENOBUFS;
83 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
84 nfs4_print_deviceid(devid);
85 return 0;
86}
87
88static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
89{
90 __be32 *p;
91
92 p = xdr_inline_decode(xdr, 4);
93 if (unlikely(!p))
94 return -ENOBUFS;
95 fh->size = be32_to_cpup(p++);
96 if (fh->size > sizeof(struct nfs_fh)) {
97 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
98 fh->size);
99 return -EOVERFLOW;
100 }
101 /* fh.data */
102 p = xdr_inline_decode(xdr, fh->size);
103 if (unlikely(!p))
104 return -ENOBUFS;
105 memcpy(&fh->data, p, fh->size);
106 dprintk("%s: fh len %d\n", __func__, fh->size);
107
108 return 0;
109}
110
111/*
112 * Currently only stringified uids and gids are accepted.
113 * I.e., kerberos is not supported to the DSes, so no pricipals.
114 *
115 * That means that one common function will suffice, but when
116 * principals are added, this should be split to accomodate
117 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
118 */
119static int
120decode_name(struct xdr_stream *xdr, u32 *id)
121{
122 __be32 *p;
123 int len;
124
125 /* opaque_length(4)*/
126 p = xdr_inline_decode(xdr, 4);
127 if (unlikely(!p))
128 return -ENOBUFS;
129 len = be32_to_cpup(p++);
130 if (len < 0)
131 return -EINVAL;
132
133 dprintk("%s: len %u\n", __func__, len);
134
135 /* opaque body */
136 p = xdr_inline_decode(xdr, len);
137 if (unlikely(!p))
138 return -ENOBUFS;
139
140 if (!nfs_map_string_to_numeric((char *)p, len, id))
141 return -EINVAL;
142
143 return 0;
144}
145
266d12d4
TM		 146static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
147 const struct nfs4_ff_layout_mirror *m2)
148{
149 int i, j;
150
151 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
152 return false;
153 for (i = 0; i < m1->fh_versions_cnt; i++) {
154 bool found_fh = false;
86fb449b 155 for (j = 0; j < m2->fh_versions_cnt; j++) {
266d12d4
TM		 156 if (nfs_compare_fh(&m1->fh_versions[i],
157 &m2->fh_versions[j]) == 0) {
158 found_fh = true;
159 break;
160 }
161 }
162 if (!found_fh)
163 return false;
164 }
165 return true;
166}
167
168static struct nfs4_ff_layout_mirror *
169ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
170 struct nfs4_ff_layout_mirror *mirror)
171{
172 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
173 struct nfs4_ff_layout_mirror *pos;
174 struct inode *inode = lo->plh_inode;
175
176 spin_lock(&inode->i_lock);
177 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
178 if (mirror->mirror_ds != pos->mirror_ds)
179 continue;
180 if (!ff_mirror_match_fh(mirror, pos))
181 continue;
182 if (atomic_inc_not_zero(&pos->ref)) {
183 spin_unlock(&inode->i_lock);
184 return pos;
185 }
186 }
187 list_add(&mirror->mirrors, &ff_layout->mirrors);
188 mirror->layout = lo;
189 spin_unlock(&inode->i_lock);
190 return mirror;
191}
192
e3b1df2d 193static void
266d12d4
TM		 194ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
195{
196 struct inode *inode;
197 if (mirror->layout == NULL)
198 return;
199 inode = mirror->layout->plh_inode;
200 spin_lock(&inode->i_lock);
201 list_del(&mirror->mirrors);
202 spin_unlock(&inode->i_lock);
203 mirror->layout = NULL;
204}
205
28a0d72c
TM		 206static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
207{
208 struct nfs4_ff_layout_mirror *mirror;
209
210 mirror = kzalloc(sizeof(*mirror), gfp_flags);
211 if (mirror != NULL) {
212 spin_lock_init(&mirror->lock);
213 atomic_set(&mirror->ref, 1);
266d12d4 214 INIT_LIST_HEAD(&mirror->mirrors);
28a0d72c
TM		 215 }
216 return mirror;
217}
218
219static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
220{
3064b686
JL		 221 struct rpc_cred *cred;
222
266d12d4 223 ff_layout_remove_mirror(mirror);
28a0d72c 224 kfree(mirror->fh_versions);
3064b686
JL		 225 cred = rcu_access_pointer(mirror->ro_cred);
226 if (cred)
227 put_rpccred(cred);
228 cred = rcu_access_pointer(mirror->rw_cred);
229 if (cred)
230 put_rpccred(cred);
28a0d72c
TM		 231 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
232 kfree(mirror);
233}
234
235static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
236{
237 if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
238 ff_layout_free_mirror(mirror);
239}
240
d67ae825
TH		 241static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
242{
243 int i;
244
245 if (fls->mirror_array) {
246 for (i = 0; i < fls->mirror_array_cnt; i++) {
247 /* normally mirror_ds is freed in
248 * .free_deviceid_node but we still do it here
249 * for .alloc_lseg error path */
28a0d72c 250 ff_layout_put_mirror(fls->mirror_array[i]);
d67ae825
TH		 251 }
252 kfree(fls->mirror_array);
253 fls->mirror_array = NULL;
254 }
255}
256
257static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
258{
259 int ret = 0;
260
261 dprintk("--> %s\n", __func__);
262
263 /* FIXME: remove this check when layout segment support is added */
264 if (lgr->range.offset != 0 ||
265 lgr->range.length != NFS4_MAX_UINT64) {
266 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
267 __func__);
268 ret = -EINVAL;
269 }
270
271 dprintk("--> %s returns %d\n", __func__, ret);
272 return ret;
273}
274
275static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
276{
277 if (fls) {
278 ff_layout_free_mirror_array(fls);
279 kfree(fls);
280 }
281}
282
0762ed2c
TM		 283static bool
284ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
285 const struct pnfs_layout_range *l2)
286{
287 u64 end1, end2;
288
289 if (l1->iomode != l2->iomode)
290 return l1->iomode != IOMODE_READ;
291 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
292 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
293 if (end1 < l2->offset)
294 return false;
295 if (end2 < l1->offset)
296 return true;
297 return l2->offset <= l1->offset;
298}
299
300static bool
301ff_lseg_merge(struct pnfs_layout_segment *new,
302 struct pnfs_layout_segment *old)
303{
304 u64 new_end, old_end;
305
ee26bdd6
JL		 306 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
307 return false;
0762ed2c
TM		 308 if (new->pls_range.iomode != old->pls_range.iomode)
309 return false;
310 old_end = pnfs_calc_offset_end(old->pls_range.offset,
311 old->pls_range.length);
312 if (old_end < new->pls_range.offset)
313 return false;
314 new_end = pnfs_calc_offset_end(new->pls_range.offset,
315 new->pls_range.length);
316 if (new_end < old->pls_range.offset)
317 return false;
318
319 /* Mergeable: copy info from 'old' to 'new' */
320 if (new_end < old_end)
321 new_end = old_end;
322 if (new->pls_range.offset < old->pls_range.offset)
323 new->pls_range.offset = old->pls_range.offset;
324 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
325 new_end);
326 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
327 set_bit(NFS_LSEG_ROC, &new->pls_flags);
0762ed2c
TM		 328 return true;
329}
330
331static void
332ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
333 struct pnfs_layout_segment *lseg,
334 struct list_head *free_me)
335{
336 pnfs_generic_layout_insert_lseg(lo, lseg,
337 ff_lseg_range_is_after,
338 ff_lseg_merge,
339 free_me);
340}
341
d67ae825
TH		 342static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
343{
d67ae825
TH		 344 int i, j;
345
346 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
347 for (j = i + 1; j < fls->mirror_array_cnt; j++)
348 if (fls->mirror_array[i]->efficiency <
455b6ee6
FF		 349 fls->mirror_array[j]->efficiency)
350 swap(fls->mirror_array[i],
351 fls->mirror_array[j]);
d67ae825
TH		 352 }
353}
354
260074cd
TM		 355static void ff_layout_mark_devices_valid(struct nfs4_ff_layout_segment *fls)
356{
357 struct nfs4_deviceid_node *node;
358 int i;
359
360 if (!(fls->flags & FF_FLAGS_NO_IO_THRU_MDS))
361 return;
362 for (i = 0; i < fls->mirror_array_cnt; i++) {
363 node = &fls->mirror_array[i]->mirror_ds->id_node;
364 clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
365 }
366}
367
d67ae825
TH		 368static struct pnfs_layout_segment *
369ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
370 struct nfs4_layoutget_res *lgr,
371 gfp_t gfp_flags)
372{
373 struct pnfs_layout_segment *ret;
374 struct nfs4_ff_layout_segment *fls = NULL;
375 struct xdr_stream stream;
376 struct xdr_buf buf;
377 struct page *scratch;
378 u64 stripe_unit;
379 u32 mirror_array_cnt;
380 __be32 *p;
381 int i, rc;
382
383 dprintk("--> %s\n", __func__);
384 scratch = alloc_page(gfp_flags);
385 if (!scratch)
386 return ERR_PTR(-ENOMEM);
387
388 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
389 lgr->layoutp->len);
390 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
391
392 /* stripe unit and mirror_array_cnt */
393 rc = -EIO;
394 p = xdr_inline_decode(&stream, 8 + 4);
395 if (!p)
396 goto out_err_free;
397
398 p = xdr_decode_hyper(p, &stripe_unit);
399 mirror_array_cnt = be32_to_cpup(p++);
400 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
401 stripe_unit, mirror_array_cnt);
402
403 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
404 mirror_array_cnt == 0)
405 goto out_err_free;
406
407 rc = -ENOMEM;
408 fls = kzalloc(sizeof(*fls), gfp_flags);
409 if (!fls)
410 goto out_err_free;
411
412 fls->mirror_array_cnt = mirror_array_cnt;
413 fls->stripe_unit = stripe_unit;
414 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
415 sizeof(fls->mirror_array[0]), gfp_flags);
416 if (fls->mirror_array == NULL)
417 goto out_err_free;
418
419 for (i = 0; i < fls->mirror_array_cnt; i++) {
266d12d4 420 struct nfs4_ff_layout_mirror *mirror;
d67ae825
TH		 421 struct nfs4_deviceid devid;
422 struct nfs4_deviceid_node *idnode;
90a0be00 423 struct auth_cred acred = { .group_info = ff_zero_group };
3064b686 424 struct rpc_cred __rcu *cred;
90a0be00 425 u32 ds_count, fh_count, id;
d67ae825
TH		 426 int j;
427
428 rc = -EIO;
429 p = xdr_inline_decode(&stream, 4);
430 if (!p)
431 goto out_err_free;
432 ds_count = be32_to_cpup(p);
433
434 /* FIXME: allow for striping? */
435 if (ds_count != 1)
436 goto out_err_free;
437
28a0d72c 438 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
d67ae825
TH		 439 if (fls->mirror_array[i] == NULL) {
440 rc = -ENOMEM;
441 goto out_err_free;
442 }
443
d67ae825
TH		 444 fls->mirror_array[i]->ds_count = ds_count;
445
446 /* deviceid */
447 rc = decode_deviceid(&stream, &devid);
448 if (rc)
449 goto out_err_free;
450
451 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
452 &devid, lh->plh_lc_cred,
453 gfp_flags);
454 /*
455 * upon success, mirror_ds is allocated by previous
456 * getdeviceinfo, or newly by .alloc_deviceid_node
457 * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
458 */
459 if (idnode)
460 fls->mirror_array[i]->mirror_ds =
461 FF_LAYOUT_MIRROR_DS(idnode);
462 else
463 goto out_err_free;
464
465 /* efficiency */
466 rc = -EIO;
467 p = xdr_inline_decode(&stream, 4);
468 if (!p)
469 goto out_err_free;
470 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
471
472 /* stateid */
93b717fd 473 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
d67ae825
TH		 474 if (rc)
475 goto out_err_free;
476
477 /* fh */
478 p = xdr_inline_decode(&stream, 4);
479 if (!p)
480 goto out_err_free;
481 fh_count = be32_to_cpup(p);
482
483 fls->mirror_array[i]->fh_versions =
484 kzalloc(fh_count * sizeof(struct nfs_fh),
485 gfp_flags);
486 if (fls->mirror_array[i]->fh_versions == NULL) {
487 rc = -ENOMEM;
488 goto out_err_free;
489 }
490
491 for (j = 0; j < fh_count; j++) {
492 rc = decode_nfs_fh(&stream,
493 &fls->mirror_array[i]->fh_versions[j]);
494 if (rc)
495 goto out_err_free;
496 }
497
498 fls->mirror_array[i]->fh_versions_cnt = fh_count;
499
500 /* user */
90a0be00 501 rc = decode_name(&stream, &id);
d67ae825
TH		 502 if (rc)
503 goto out_err_free;
504
90a0be00
JL		 505 acred.uid = make_kuid(&init_user_ns, id);
506
d67ae825 507 /* group */
90a0be00 508 rc = decode_name(&stream, &id);
d67ae825
TH		 509 if (rc)
510 goto out_err_free;
511
90a0be00
JL		 512 acred.gid = make_kgid(&init_user_ns, id);
513
514 /* find the cred for it */
3064b686 515 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags));
90a0be00
JL		 516 if (IS_ERR(cred)) {
517 rc = PTR_ERR(cred);
518 goto out_err_free;
519 }
520
3064b686
JL		 521 if (lgr->range.iomode == IOMODE_READ)
522 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
523 else
524 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
90a0be00 525
266d12d4
TM		 526 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
527 if (mirror != fls->mirror_array[i]) {
90a0be00 528 /* swap cred ptrs so free_mirror will clean up old */
3064b686
JL		 529 if (lgr->range.iomode == IOMODE_READ) {
530 cred = xchg(&mirror->ro_cred, cred);
531 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
532 } else {
533 cred = xchg(&mirror->rw_cred, cred);
534 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
535 }
266d12d4
TM		 536 ff_layout_free_mirror(fls->mirror_array[i]);
537 fls->mirror_array[i] = mirror;
538 }
539
3064b686
JL		 540 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
541 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
90a0be00
JL		 542 from_kuid(&init_user_ns, acred.uid),
543 from_kgid(&init_user_ns, acred.gid));
d67ae825
TH		 544 }
545
c0f5f505 546 p = xdr_inline_decode(&stream, 4);
d0379a5d
TM		 547 if (!p)
548 goto out_sort_mirrors;
549 fls->flags = be32_to_cpup(p);
550
551 p = xdr_inline_decode(&stream, 4);
552 if (!p)
553 goto out_sort_mirrors;
554 for (i=0; i < fls->mirror_array_cnt; i++)
555 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
c0f5f505 556
d0379a5d 557out_sort_mirrors:
d67ae825
TH		 558 ff_layout_sort_mirrors(fls);
559 rc = ff_layout_check_layout(lgr);
560 if (rc)
561 goto out_err_free;
260074cd 562 ff_layout_mark_devices_valid(fls);
d67ae825
TH		 563
564 ret = &fls->generic_hdr;
565 dprintk("<-- %s (success)\n", __func__);
566out_free_page:
567 __free_page(scratch);
568 return ret;
569out_err_free:
570 _ff_layout_free_lseg(fls);
571 ret = ERR_PTR(rc);
572 dprintk("<-- %s (%d)\n", __func__, rc);
573 goto out_free_page;
574}
575
576static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
577{
578 struct pnfs_layout_segment *lseg;
579
580 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
581 if (lseg->pls_range.iomode == IOMODE_RW)
582 return true;
583
584 return false;
585}
586
587static void
588ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
589{
590 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825
TH		 591
592 dprintk("--> %s\n", __func__);
593
d67ae825
TH		 594 if (lseg->pls_range.iomode == IOMODE_RW) {
595 struct nfs4_flexfile_layout *ffl;
596 struct inode *inode;
597
598 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
599 inode = ffl->generic_hdr.plh_inode;
600 spin_lock(&inode->i_lock);
601 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
602 ffl->commit_info.nbuckets = 0;
603 kfree(ffl->commit_info.buckets);
604 ffl->commit_info.buckets = NULL;
605 }
606 spin_unlock(&inode->i_lock);
607 }
608 _ff_layout_free_lseg(fls);
609}
610
611/* Return 1 until we have multiple lsegs support */
612static int
613ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
614{
615 return 1;
616}
617
abcb7bfc 618static void
e76d28dd 619nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 620{
9bbd9bb4
PT		 621 /* first IO request? */
622 if (atomic_inc_return(&timer->n_ops) == 1) {
e76d28dd 623 timer->start_time = now;
abcb7bfc
TM		 624 }
625}
626
627static ktime_t
e76d28dd 628nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
abcb7bfc 629{
e76d28dd 630 ktime_t start;
abcb7bfc 631
9bbd9bb4
PT		 632 if (atomic_dec_return(&timer->n_ops) < 0)
633 WARN_ON_ONCE(1);
634
9bbd9bb4
PT		 635 start = timer->start_time;
636 timer->start_time = now;
abcb7bfc
TM		 637 return ktime_sub(now, start);
638}
639
97ba375b 640static bool
d983803d 641nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
e76d28dd
TM		 642 struct nfs4_ff_layoutstat *layoutstat,
643 ktime_t now)
abcb7bfc 644{
d983803d 645 static const ktime_t notime = {0};
bbf58bf3 646 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
1c8d477a 647 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
d983803d 648
e76d28dd 649 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
9bbd9bb4
PT		 650 if (ktime_equal(mirror->start_time, notime))
651 mirror->start_time = now;
d0379a5d
TM		 652 if (mirror->report_interval != 0)
653 report_interval = (s64)mirror->report_interval * 1000LL;
654 else if (layoutstats_timer != 0)
bbf58bf3 655 report_interval = (s64)layoutstats_timer * 1000LL;
1c8d477a 656 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
bbf58bf3 657 report_interval) {
1c8d477a 658 ffl->last_report_time = now;
97ba375b
PT		 659 return true;
660 }
661
662 return false;
abcb7bfc
TM		 663}
664
665static void
666nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
667 __u64 requested)
668{
669 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
670
671 iostat->ops_requested++;
672 iostat->bytes_requested += requested;
673}
674
675static void
676nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
677 __u64 requested,
678 __u64 completed,
e76d28dd
TM		 679 ktime_t time_completed,
680 ktime_t time_started)
abcb7bfc
TM		 681{
682 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
e76d28dd 683 ktime_t completion_time = ktime_sub(time_completed, time_started);
abcb7bfc
TM		 684 ktime_t timer;
685
686 iostat->ops_completed++;
687 iostat->bytes_completed += completed;
688 iostat->bytes_not_delivered += requested - completed;
689
e76d28dd 690 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
abcb7bfc
TM		 691 iostat->total_busy_time =
692 ktime_add(iostat->total_busy_time, timer);
693 iostat->aggregate_completion_time =
e76d28dd
TM		 694 ktime_add(iostat->aggregate_completion_time,
695 completion_time);
abcb7bfc
TM		 696}
697
698static void
0b7baf94
TM		 699nfs4_ff_layout_stat_io_start_read(struct inode *inode,
700 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 701 __u64 requested, ktime_t now)
abcb7bfc 702{
97ba375b
PT		 703 bool report;
704
abcb7bfc 705 spin_lock(&mirror->lock);
e76d28dd 706 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
abcb7bfc 707 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
06946c6a 708 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 709 spin_unlock(&mirror->lock);
97ba375b
PT		 710
711 if (report)
0b7baf94 712 pnfs_report_layoutstat(inode, GFP_KERNEL);
abcb7bfc
TM		 713}
714
715static void
716nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
717 struct nfs4_ff_layout_mirror *mirror,
718 __u64 requested,
719 __u64 completed)
720{
721 spin_lock(&mirror->lock);
722 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
723 requested, completed,
e76d28dd 724 ktime_get(), task->tk_start);
06946c6a 725 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 726 spin_unlock(&mirror->lock);
727}
728
729static void
0b7baf94
TM		 730nfs4_ff_layout_stat_io_start_write(struct inode *inode,
731 struct nfs4_ff_layout_mirror *mirror,
e76d28dd 732 __u64 requested, ktime_t now)
abcb7bfc 733{
97ba375b
PT		 734 bool report;
735
abcb7bfc 736 spin_lock(&mirror->lock);
e76d28dd 737 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
abcb7bfc 738 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
06946c6a 739 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc 740 spin_unlock(&mirror->lock);
97ba375b
PT		 741
742 if (report)
0b7baf94 743 pnfs_report_layoutstat(inode, GFP_NOIO);
abcb7bfc
TM		 744}
745
746static void
747nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
748 struct nfs4_ff_layout_mirror *mirror,
749 __u64 requested,
750 __u64 completed,
751 enum nfs3_stable_how committed)
752{
753 if (committed == NFS_UNSTABLE)
754 requested = completed = 0;
755
756 spin_lock(&mirror->lock);
757 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
e76d28dd 758 requested, completed, ktime_get(), task->tk_start);
06946c6a 759 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
abcb7bfc
TM		 760 spin_unlock(&mirror->lock);
761}
762
d67ae825
TH		 763static int
764ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
765 struct nfs_commit_info *cinfo,
766 gfp_t gfp_flags)
767{
768 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
769 struct pnfs_commit_bucket *buckets;
770 int size;
771
772 if (cinfo->ds->nbuckets != 0) {
773 /* This assumes there is only one RW lseg per file.
774 * To support multiple lseg per file, we need to
775 * change struct pnfs_commit_bucket to allow dynamic
776 * increasing nbuckets.
777 */
778 return 0;
779 }
780
781 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
782
783 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
784 gfp_flags);
785 if (!buckets)
786 return -ENOMEM;
787 else {
788 int i;
789
fe238e60 790 spin_lock(&cinfo->inode->i_lock);
d67ae825
TH		 791 if (cinfo->ds->nbuckets != 0)
792 kfree(buckets);
793 else {
794 cinfo->ds->buckets = buckets;
795 cinfo->ds->nbuckets = size;
796 for (i = 0; i < size; i++) {
797 INIT_LIST_HEAD(&buckets[i].written);
798 INIT_LIST_HEAD(&buckets[i].committing);
799 /* mark direct verifier as unset */
800 buckets[i].direct_verf.committed =
801 NFS_INVALID_STABLE_HOW;
802 }
803 }
fe238e60 804 spin_unlock(&cinfo->inode->i_lock);
d67ae825
TH		 805 return 0;
806 }
807}
808
809static struct nfs4_pnfs_ds *
13544412
TM		 810ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
811 int start_idx,
d67ae825
TH		 812 int *best_idx)
813{
13544412 814 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
d67ae825 815 struct nfs4_pnfs_ds *ds;
3dc14735 816 bool fail_return = false;
d67ae825
TH		 817 int idx;
818
d67ae825 819 /* mirrors are sorted by efficiency */
13544412 820 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
3dc14735
TM		 821 if (idx+1 == fls->mirror_array_cnt)
822 fail_return = true;
823 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return);
d67ae825
TH		 824 if (ds) {
825 *best_idx = idx;
826 return ds;
827 }
828 }
829
830 return NULL;
831}
832
c7d73af2
TH		 833static void
834ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
835 struct nfs_page *req,
836 bool strict_iomode)
837{
838retry_strict:
839 pnfs_put_lseg(pgio->pg_lseg);
840 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
841 req->wb_context,
842 0,
843 NFS4_MAX_UINT64,
844 IOMODE_READ,
845 strict_iomode,
846 GFP_KERNEL);
847 if (IS_ERR(pgio->pg_lseg)) {
848 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
849 pgio->pg_lseg = NULL;
850 }
851
852 /* If we don't have checking, do get a IOMODE_RW
853 * segment, and the server wants to avoid READs
854 * there, then retry!
855 */
856 if (pgio->pg_lseg && !strict_iomode &&
857 ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
858 strict_iomode = true;
859 goto retry_strict;
860 }
861}
862
d67ae825
TH		 863static void
864ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
865 struct nfs_page *req)
866{
867 struct nfs_pgio_mirror *pgm;
868 struct nfs4_ff_layout_mirror *mirror;
869 struct nfs4_pnfs_ds *ds;
870 int ds_idx;
871
3dc14735 872retry:
d67ae825 873 /* Use full layout for now */
c7d73af2
TH		 874 if (!pgio->pg_lseg)
875 ff_layout_pg_get_read(pgio, req, false);
876 else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg))
877 ff_layout_pg_get_read(pgio, req, true);
602c4cd4 878
d67ae825
TH		 879 /* If no lseg, fall back to read through mds */
880 if (pgio->pg_lseg == NULL)
881 goto out_mds;
882
13544412 883 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
446ca219 884 if (!ds) {
3dc14735 885 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 886 goto out_mds;
3dc14735
TM		 887 pnfs_put_lseg(pgio->pg_lseg);
888 pgio->pg_lseg = NULL;
889 /* Sleep for 1 second before retrying */
890 ssleep(1);
891 goto retry;
446ca219
TH		 892 }
893
d67ae825
TH		 894 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
895
896 pgio->pg_mirror_idx = ds_idx;
897
898 /* read always uses only one mirror - idx 0 for pgio layer */
899 pgm = &pgio->pg_mirrors[0];
900 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
901
902 return;
903out_mds:
904 pnfs_put_lseg(pgio->pg_lseg);
905 pgio->pg_lseg = NULL;
906 nfs_pageio_reset_read_mds(pgio);
907}
908
909static void
910ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
911 struct nfs_page *req)
912{
913 struct nfs4_ff_layout_mirror *mirror;
914 struct nfs_pgio_mirror *pgm;
915 struct nfs_commit_info cinfo;
916 struct nfs4_pnfs_ds *ds;
917 int i;
918 int status;
919
3dc14735 920retry:
d600ad1f 921 if (!pgio->pg_lseg) {
d67ae825
TH		 922 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
923 req->wb_context,
924 0,
925 NFS4_MAX_UINT64,
926 IOMODE_RW,
c7d73af2 927 false,
d67ae825 928 GFP_NOFS);
d600ad1f
PT		 929 if (IS_ERR(pgio->pg_lseg)) {
930 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
931 pgio->pg_lseg = NULL;
932 return;
933 }
934 }
d67ae825
TH		 935 /* If no lseg, fall back to write through mds */
936 if (pgio->pg_lseg == NULL)
937 goto out_mds;
938
939 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
940 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
941 if (status < 0)
942 goto out_mds;
943
944 /* Use a direct mapping of ds_idx to pgio mirror_idx */
945 if (WARN_ON_ONCE(pgio->pg_mirror_count !=
946 FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
947 goto out_mds;
948
949 for (i = 0; i < pgio->pg_mirror_count; i++) {
950 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
446ca219 951 if (!ds) {
3dc14735 952 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
446ca219 953 goto out_mds;
3dc14735
TM		 954 pnfs_put_lseg(pgio->pg_lseg);
955 pgio->pg_lseg = NULL;
956 /* Sleep for 1 second before retrying */
957 ssleep(1);
958 goto retry;
446ca219 959 }
d67ae825
TH		 960 pgm = &pgio->pg_mirrors[i];
961 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
962 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
963 }
964
965 return;
966
967out_mds:
968 pnfs_put_lseg(pgio->pg_lseg);
969 pgio->pg_lseg = NULL;
970 nfs_pageio_reset_write_mds(pgio);
971}
972
973static unsigned int
974ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
975 struct nfs_page *req)
976{
d600ad1f 977 if (!pgio->pg_lseg) {
d67ae825
TH		 978 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
979 req->wb_context,
980 0,
981 NFS4_MAX_UINT64,
982 IOMODE_RW,
c7d73af2 983 false,
d67ae825 984 GFP_NOFS);
d600ad1f
PT		 985 if (IS_ERR(pgio->pg_lseg)) {
986 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
987 pgio->pg_lseg = NULL;
988 goto out;
989 }
990 }
d67ae825
TH		 991 if (pgio->pg_lseg)
992 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
993
994 /* no lseg means that pnfs is not in use, so no mirroring here */
d67ae825 995 nfs_pageio_reset_write_mds(pgio);
d600ad1f 996out:
d67ae825
TH		 997 return 1;
998}
999
1000static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1001 .pg_init = ff_layout_pg_init_read,
1002 .pg_test = pnfs_generic_pg_test,
1003 .pg_doio = pnfs_generic_pg_readpages,
1004 .pg_cleanup = pnfs_generic_pg_cleanup,
1005};
1006
1007static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1008 .pg_init = ff_layout_pg_init_write,
1009 .pg_test = pnfs_generic_pg_test,
1010 .pg_doio = pnfs_generic_pg_writepages,
1011 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1012 .pg_cleanup = pnfs_generic_pg_cleanup,
1013};
1014
1015static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1016{
1017 struct rpc_task *task = &hdr->task;
1018
1019 pnfs_layoutcommit_inode(hdr->inode, false);
1020
1021 if (retry_pnfs) {
1022 dprintk("%s Reset task %5u for i/o through pNFS "
1023 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1024 hdr->task.tk_pid,
1025 hdr->inode->i_sb->s_id,
1026 (unsigned long long)NFS_FILEID(hdr->inode),
1027 hdr->args.count,
1028 (unsigned long long)hdr->args.offset);
1029
dc602dd7 1030 hdr->completion_ops->reschedule_io(hdr);
d67ae825
TH		 1031 return;
1032 }
1033
1034 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1035 dprintk("%s Reset task %5u for i/o through MDS "
1036 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1037 hdr->task.tk_pid,
1038 hdr->inode->i_sb->s_id,
1039 (unsigned long long)NFS_FILEID(hdr->inode),
1040 hdr->args.count,
1041 (unsigned long long)hdr->args.offset);
1042
1043 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1044 }
1045}
1046
1047static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1048{
1049 struct rpc_task *task = &hdr->task;
1050
1051 pnfs_layoutcommit_inode(hdr->inode, false);
1052
1053 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1054 dprintk("%s Reset task %5u for i/o through MDS "
1055 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1056 hdr->task.tk_pid,
1057 hdr->inode->i_sb->s_id,
1058 (unsigned long long)NFS_FILEID(hdr->inode),
1059 hdr->args.count,
1060 (unsigned long long)hdr->args.offset);
1061
1062 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1063 }
1064}
1065
1066static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1067 struct nfs4_state *state,
1068 struct nfs_client *clp,
1069 struct pnfs_layout_segment *lseg,
1070 int idx)
1071{
1072 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1073 struct inode *inode = lo->plh_inode;
1074 struct nfs_server *mds_server = NFS_SERVER(inode);
1075
1076 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1077 struct nfs_client *mds_client = mds_server->nfs_client;
1078 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1079
1080 if (task->tk_status >= 0)
1081 return 0;
1082
1083 switch (task->tk_status) {
1084 /* MDS state errors */
1085 case -NFS4ERR_DELEG_REVOKED:
1086 case -NFS4ERR_ADMIN_REVOKED:
1087 case -NFS4ERR_BAD_STATEID:
1088 if (state == NULL)
1089 break;
41020b67 1090 nfs_remove_bad_delegation(state->inode, NULL);
d67ae825
TH		 1091 case -NFS4ERR_OPENMODE:
1092 if (state == NULL)
1093 break;
1094 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1095 goto out_bad_stateid;
1096 goto wait_on_recovery;
1097 case -NFS4ERR_EXPIRED:
1098 if (state != NULL) {
1099 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1100 goto out_bad_stateid;
1101 }
1102 nfs4_schedule_lease_recovery(mds_client);
1103 goto wait_on_recovery;
1104 /* DS session errors */
1105 case -NFS4ERR_BADSESSION:
1106 case -NFS4ERR_BADSLOT:
1107 case -NFS4ERR_BAD_HIGH_SLOT:
1108 case -NFS4ERR_DEADSESSION:
1109 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1110 case -NFS4ERR_SEQ_FALSE_RETRY:
1111 case -NFS4ERR_SEQ_MISORDERED:
1112 dprintk("%s ERROR %d, Reset session. Exchangeid "
1113 "flags 0x%x\n", __func__, task->tk_status,
1114 clp->cl_exchange_flags);
1115 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1116 break;
1117 case -NFS4ERR_DELAY:
1118 case -NFS4ERR_GRACE:
1119 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1120 break;
1121 case -NFS4ERR_RETRY_UNCACHED_REP:
1122 break;
1123 /* Invalidate Layout errors */
1124 case -NFS4ERR_PNFS_NO_LAYOUT:
1125 case -ESTALE: /* mapped NFS4ERR_STALE */
1126 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1127 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1128 case -NFS4ERR_FHEXPIRED:
1129 case -NFS4ERR_WRONG_TYPE:
1130 dprintk("%s Invalid layout error %d\n", __func__,
1131 task->tk_status);
1132 /*
1133 * Destroy layout so new i/o will get a new layout.
1134 * Layout will not be destroyed until all current lseg
1135 * references are put. Mark layout as invalid to resend failed
1136 * i/o and all i/o waiting on the slot table to the MDS until
1137 * layout is destroyed and a new valid layout is obtained.
1138 */
1139 pnfs_destroy_layout(NFS_I(inode));
1140 rpc_wake_up(&tbl->slot_tbl_waitq);
1141 goto reset;
1142 /* RPC connection errors */
1143 case -ECONNREFUSED:
1144 case -EHOSTDOWN:
1145 case -EHOSTUNREACH:
1146 case -ENETUNREACH:
1147 case -EIO:
1148 case -ETIMEDOUT:
1149 case -EPIPE:
1150 dprintk("%s DS connection error %d\n", __func__,
1151 task->tk_status);
1152 nfs4_mark_deviceid_unavailable(devid);
1153 rpc_wake_up(&tbl->slot_tbl_waitq);
1154 /* fall through */
1155 default:
3b13b4b3 1156 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1157 return -NFS4ERR_RESET_TO_PNFS;
1158reset:
1159 dprintk("%s Retry through MDS. Error %d\n", __func__,
1160 task->tk_status);
1161 return -NFS4ERR_RESET_TO_MDS;
1162 }
1163out:
1164 task->tk_status = 0;
1165 return -EAGAIN;
1166out_bad_stateid:
1167 task->tk_status = -EIO;
1168 return 0;
1169wait_on_recovery:
1170 rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
1171 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
1172 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
1173 goto out;
1174}
1175
1176/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1177static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1178 struct pnfs_layout_segment *lseg,
1179 int idx)
1180{
1181 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1182
1183 if (task->tk_status >= 0)
1184 return 0;
1185
90816d1d
TM		 1186 switch (task->tk_status) {
1187 /* File access problems. Don't mark the device as unavailable */
1188 case -EACCES:
1189 case -ESTALE:
1190 case -EISDIR:
1191 case -EBADHANDLE:
1192 case -ELOOP:
1193 case -ENOSPC:
1194 break;
1195 case -EJUKEBOX:
1196 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1197 goto out_retry;
1198 default:
d67ae825
TH		 1199 dprintk("%s DS connection error %d\n", __func__,
1200 task->tk_status);
1201 nfs4_mark_deviceid_unavailable(devid);
d67ae825 1202 }
90816d1d
TM		 1203 /* FIXME: Need to prevent infinite looping here. */
1204 return -NFS4ERR_RESET_TO_PNFS;
1205out_retry:
d67ae825 1206 task->tk_status = 0;
7eeea167 1207 rpc_restart_call_prepare(task);
d67ae825
TH		 1208 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1209 return -EAGAIN;
1210}
1211
1212static int ff_layout_async_handle_error(struct rpc_task *task,
1213 struct nfs4_state *state,
1214 struct nfs_client *clp,
1215 struct pnfs_layout_segment *lseg,
1216 int idx)
1217{
1218 int vers = clp->cl_nfs_mod->rpc_vers->number;
1219
1220 switch (vers) {
1221 case 3:
1222 return ff_layout_async_handle_error_v3(task, lseg, idx);
1223 case 4:
1224 return ff_layout_async_handle_error_v4(task, state, clp,
1225 lseg, idx);
1226 default:
1227 /* should never happen */
1228 WARN_ON_ONCE(1);
1229 return 0;
1230 }
1231}
1232
1233static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1234 int idx, u64 offset, u64 length,
dd52128a 1235 u32 status, int opnum, int error)
d67ae825
TH		 1236{
1237 struct nfs4_ff_layout_mirror *mirror;
1238 int err;
1239
dd52128a
TM		 1240 if (status == 0) {
1241 switch (error) {
1242 case -ETIMEDOUT:
1243 case -EPFNOSUPPORT:
1244 case -EPROTONOSUPPORT:
1245 case -EOPNOTSUPP:
1246 case -ECONNREFUSED:
1247 case -ECONNRESET:
1248 case -EHOSTDOWN:
1249 case -EHOSTUNREACH:
1250 case -ENETUNREACH:
1251 case -EADDRINUSE:
1252 case -ENOBUFS:
1253 case -EPIPE:
1254 case -EPERM:
1255 status = NFS4ERR_NXIO;
1256 break;
1257 case -EACCES:
1258 status = NFS4ERR_ACCESS;
1259 break;
1260 default:
1261 return;
1262 }
1263 }
1264
c22eeb86
PT		 1265 switch (status) {
1266 case NFS4ERR_DELAY:
1267 case NFS4ERR_GRACE:
1268 return;
1269 default:
1270 break;
1271 }
1272
d67ae825
TH		 1273 mirror = FF_LAYOUT_COMP(lseg, idx);
1274 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1275 mirror, offset, length, status, opnum,
1276 GFP_NOIO);
f95c03b2 1277 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
d67ae825
TH		 1278 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1279}
1280
1281/* NFS_PROTO call done callback routines */
d67ae825
TH		 1282static int ff_layout_read_done_cb(struct rpc_task *task,
1283 struct nfs_pgio_header *hdr)
1284{
d67ae825
TH		 1285 int err;
1286
1287 trace_nfs4_pnfs_read(hdr, task->tk_status);
dd52128a 1288 if (task->tk_status < 0)
d67ae825
TH		 1289 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1290 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1291 hdr->res.op_status, OP_READ,
1292 task->tk_status);
d67ae825
TH		 1293 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1294 hdr->ds_clp, hdr->lseg,
1295 hdr->pgio_mirror_idx);
1296
1297 switch (err) {
1298 case -NFS4ERR_RESET_TO_PNFS:
13544412
TM		 1299 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1300 hdr->pgio_mirror_idx + 1,
1301 &hdr->pgio_mirror_idx))
1302 goto out_eagain;
54e4a0df 1303 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1304 pnfs_read_resend_pnfs(hdr);
1305 return task->tk_status;
1306 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1307 ff_layout_reset_read(hdr);
1308 return task->tk_status;
1309 case -EAGAIN:
13544412 1310 goto out_eagain;
d67ae825
TH		 1311 }
1312
1313 return 0;
13544412
TM		 1314out_eagain:
1315 rpc_restart_call_prepare(task);
1316 return -EAGAIN;
d67ae825
TH		 1317}
1318
c0f5f505
TM		 1319static bool
1320ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1321{
1322 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1323}
1324
d67ae825
TH		 1325/*
1326 * We reference the rpc_cred of the first WRITE that triggers the need for
1327 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1328 * rfc5661 is not clear about which credential should be used.
1329 *
1330 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1331 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1332 * we always send layoutcommit after DS writes.
1333 */
1334static void
bc28e1c2
TM		 1335ff_layout_set_layoutcommit(struct inode *inode,
1336 struct pnfs_layout_segment *lseg,
1337 loff_t end_offset)
d67ae825 1338{
bc28e1c2 1339 if (!ff_layout_need_layoutcommit(lseg))
c0f5f505
TM		 1340 return;
1341
bc28e1c2
TM		 1342 pnfs_set_layoutcommit(inode, lseg, end_offset);
1343 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1344 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
d67ae825
TH		 1345}
1346
1347static bool
75bf47eb 1348ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx)
d67ae825
TH		 1349{
1350 /* No mirroring for now */
1351 struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1352
1353 return ff_layout_test_devid_unavailable(node);
1354}
1355
37e9ed22
TM		 1356static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1357 struct nfs_pgio_header *hdr)
d67ae825 1358{
37e9ed22
TM		 1359 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1360 return;
0b7baf94 1361 nfs4_ff_layout_stat_io_start_read(hdr->inode,
abcb7bfc 1362 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1363 hdr->args.count,
1364 task->tk_start);
37e9ed22
TM		 1365}
1366
1367static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1368 struct nfs_pgio_header *hdr)
1369{
1370 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1371 return;
1372 nfs4_ff_layout_stat_io_end_read(task,
1373 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1374 hdr->args.count,
1375 hdr->res.count);
1376}
abcb7bfc 1377
37e9ed22
TM		 1378static int ff_layout_read_prepare_common(struct rpc_task *task,
1379 struct nfs_pgio_header *hdr)
1380{
d67ae825
TH		 1381 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1382 rpc_exit(task, -EIO);
1383 return -EIO;
1384 }
75bf47eb
TM		 1385 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1386 rpc_exit(task, -EHOSTDOWN);
d67ae825
TH		 1387 return -EAGAIN;
1388 }
d67ae825 1389
37e9ed22 1390 ff_layout_read_record_layoutstats_start(task, hdr);
d67ae825
TH		 1391 return 0;
1392}
1393
1394/*
1395 * Call ops for the async read/write cases
1396 * In the case of dense layouts, the offset needs to be reset to its
1397 * original value.
1398 */
1399static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1400{
1401 struct nfs_pgio_header *hdr = data;
1402
1403 if (ff_layout_read_prepare_common(task, hdr))
1404 return;
1405
1406 rpc_call_start(task);
1407}
1408
1409static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1410 struct nfs4_sequence_args *args,
1411 struct nfs4_sequence_res *res,
1412 struct rpc_task *task)
1413{
1414 if (ds_clp->cl_session)
1415 return nfs41_setup_sequence(ds_clp->cl_session,
1416 args,
1417 res,
1418 task);
1419 return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1420 args,
1421 res,
1422 task);
1423}
1424
1425static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1426{
1427 struct nfs_pgio_header *hdr = data;
1428
d67ae825
TH		 1429 if (ff_layout_setup_sequence(hdr->ds_clp,
1430 &hdr->args.seq_args,
1431 &hdr->res.seq_res,
1432 task))
1433 return;
1434
abcb7bfc
TM		 1435 if (ff_layout_read_prepare_common(task, hdr))
1436 return;
1437
d67ae825
TH		 1438 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1439 hdr->args.lock_context, FMODE_READ) == -EIO)
1440 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1441}
1442
1443static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1444{
1445 struct nfs_pgio_header *hdr = data;
1446
1447 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1448
1449 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1450 task->tk_status == 0) {
1451 nfs4_sequence_done(task, &hdr->res.seq_res);
1452 return;
1453 }
1454
1455 /* Note this may cause RPC to be resent */
1456 hdr->mds_ops->rpc_call_done(task, hdr);
1457}
1458
1459static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1460{
1461 struct nfs_pgio_header *hdr = data;
1462
37e9ed22 1463 ff_layout_read_record_layoutstats_done(task, hdr);
d67ae825
TH		 1464 rpc_count_iostats_metrics(task,
1465 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1466}
1467
4d0ac221
TM		 1468static void ff_layout_read_release(void *data)
1469{
1470 struct nfs_pgio_header *hdr = data;
1471
1472 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1473 pnfs_generic_rw_release(data);
1474}
1475
1476
d67ae825
TH		 1477static int ff_layout_write_done_cb(struct rpc_task *task,
1478 struct nfs_pgio_header *hdr)
1479{
2e18d4d8 1480 loff_t end_offs = 0;
d67ae825
TH		 1481 int err;
1482
1483 trace_nfs4_pnfs_write(hdr, task->tk_status);
dd52128a 1484 if (task->tk_status < 0)
d67ae825
TH		 1485 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1486 hdr->args.offset, hdr->args.count,
dd52128a
TM		 1487 hdr->res.op_status, OP_WRITE,
1488 task->tk_status);
d67ae825
TH		 1489 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1490 hdr->ds_clp, hdr->lseg,
1491 hdr->pgio_mirror_idx);
1492
1493 switch (err) {
1494 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1495 ff_layout_reset_write(hdr, true);
1496 return task->tk_status;
d67ae825 1497 case -NFS4ERR_RESET_TO_MDS:
7cc8c5cd 1498 ff_layout_reset_write(hdr, false);
d67ae825
TH		 1499 return task->tk_status;
1500 case -EAGAIN:
d67ae825
TH		 1501 return -EAGAIN;
1502 }
1503
1504 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1505 hdr->res.verf->committed == NFS_DATA_SYNC)
2e18d4d8
TM		 1506 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1507
1508 /* Note: if the write is unstable, don't set end_offs until commit */
1509 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
d67ae825 1510
54204010
PT		 1511 /* zero out fattr since we don't care DS attr at all */
1512 hdr->fattr.valid = 0;
69f230d9
PT		 1513 if (task->tk_status >= 0)
1514 nfs_writeback_update_inode(hdr);
1515
d67ae825
TH		 1516 return 0;
1517}
1518
1519static int ff_layout_commit_done_cb(struct rpc_task *task,
1520 struct nfs_commit_data *data)
1521{
d67ae825
TH		 1522 int err;
1523
1524 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
dd52128a 1525 if (task->tk_status < 0)
d67ae825
TH		 1526 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1527 data->args.offset, data->args.count,
dd52128a
TM		 1528 data->res.op_status, OP_COMMIT,
1529 task->tk_status);
d67ae825
TH		 1530 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1531 data->lseg, data->ds_commit_index);
1532
1533 switch (err) {
1534 case -NFS4ERR_RESET_TO_PNFS:
7cc8c5cd
TM		 1535 pnfs_generic_prepare_to_resend_writes(data);
1536 return -EAGAIN;
d67ae825 1537 case -NFS4ERR_RESET_TO_MDS:
d67ae825
TH		 1538 pnfs_generic_prepare_to_resend_writes(data);
1539 return -EAGAIN;
1540 case -EAGAIN:
1541 rpc_restart_call_prepare(task);
1542 return -EAGAIN;
1543 }
1544
bc28e1c2 1545 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
d67ae825
TH		 1546
1547 return 0;
1548}
1549
37e9ed22
TM		 1550static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1551 struct nfs_pgio_header *hdr)
d67ae825 1552{
37e9ed22
TM		 1553 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1554 return;
0b7baf94 1555 nfs4_ff_layout_stat_io_start_write(hdr->inode,
abcb7bfc 1556 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
e76d28dd
TM		 1557 hdr->args.count,
1558 task->tk_start);
37e9ed22
TM		 1559}
1560
1561static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1562 struct nfs_pgio_header *hdr)
1563{
1564 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1565 return;
1566 nfs4_ff_layout_stat_io_end_write(task,
1567 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1568 hdr->args.count, hdr->res.count,
1569 hdr->res.verf->committed);
1570}
abcb7bfc 1571
37e9ed22
TM		 1572static int ff_layout_write_prepare_common(struct rpc_task *task,
1573 struct nfs_pgio_header *hdr)
1574{
d67ae825
TH		 1575 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1576 rpc_exit(task, -EIO);
1577 return -EIO;
1578 }
1579
75bf47eb
TM		 1580 if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1581 rpc_exit(task, -EHOSTDOWN);
d67ae825
TH		 1582 return -EAGAIN;
1583 }
1584
37e9ed22 1585 ff_layout_write_record_layoutstats_start(task, hdr);
d67ae825
TH		 1586 return 0;
1587}
1588
1589static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1590{
1591 struct nfs_pgio_header *hdr = data;
1592
1593 if (ff_layout_write_prepare_common(task, hdr))
1594 return;
1595
1596 rpc_call_start(task);
1597}
1598
1599static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1600{
1601 struct nfs_pgio_header *hdr = data;
1602
d67ae825
TH		 1603 if (ff_layout_setup_sequence(hdr->ds_clp,
1604 &hdr->args.seq_args,
1605 &hdr->res.seq_res,
1606 task))
1607 return;
1608
abcb7bfc
TM		 1609 if (ff_layout_write_prepare_common(task, hdr))
1610 return;
1611
d67ae825
TH		 1612 if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1613 hdr->args.lock_context, FMODE_WRITE) == -EIO)
1614 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1615}
1616
1617static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1618{
1619 struct nfs_pgio_header *hdr = data;
1620
1621 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1622 task->tk_status == 0) {
1623 nfs4_sequence_done(task, &hdr->res.seq_res);
1624 return;
1625 }
1626
1627 /* Note this may cause RPC to be resent */
1628 hdr->mds_ops->rpc_call_done(task, hdr);
1629}
1630
1631static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1632{
1633 struct nfs_pgio_header *hdr = data;
1634
37e9ed22 1635 ff_layout_write_record_layoutstats_done(task, hdr);
d67ae825
TH		 1636 rpc_count_iostats_metrics(task,
1637 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1638}
1639
4d0ac221
TM		 1640static void ff_layout_write_release(void *data)
1641{
1642 struct nfs_pgio_header *hdr = data;
1643
1644 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1645 pnfs_generic_rw_release(data);
1646}
1647
37e9ed22 1648static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
abcb7bfc
TM		 1649 struct nfs_commit_data *cdata)
1650{
37e9ed22
TM		 1651 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1652 return;
0b7baf94 1653 nfs4_ff_layout_stat_io_start_write(cdata->inode,
abcb7bfc 1654 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
e76d28dd 1655 0, task->tk_start);
abcb7bfc
TM		 1656}
1657
37e9ed22
TM		 1658static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1659 struct nfs_commit_data *cdata)
1660{
1661 struct nfs_page *req;
1662 __u64 count = 0;
1663
1664 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1665 return;
1666
1667 if (task->tk_status == 0) {
1668 list_for_each_entry(req, &cdata->pages, wb_list)
1669 count += req->wb_bytes;
1670 }
1671 nfs4_ff_layout_stat_io_end_write(task,
1672 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1673 count, count, NFS_FILE_SYNC);
1674}
1675
1676static void ff_layout_commit_prepare_common(struct rpc_task *task,
1677 struct nfs_commit_data *cdata)
1678{
1679 ff_layout_commit_record_layoutstats_start(task, cdata);
1680}
1681
d67ae825
TH		 1682static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1683{
abcb7bfc 1684 ff_layout_commit_prepare_common(task, data);
d67ae825
TH		 1685 rpc_call_start(task);
1686}
1687
1688static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1689{
1690 struct nfs_commit_data *wdata = data;
1691
abcb7bfc 1692 if (ff_layout_setup_sequence(wdata->ds_clp,
d67ae825
TH		 1693 &wdata->args.seq_args,
1694 &wdata->res.seq_res,
abcb7bfc
TM		 1695 task))
1696 return;
1697 ff_layout_commit_prepare_common(task, data);
1698}
1699
1700static void ff_layout_commit_done(struct rpc_task *task, void *data)
141b9b59
PT		 1701{
1702 pnfs_generic_write_commit_done(task, data);
1703}
1704
1705static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
abcb7bfc
TM		 1706{
1707 struct nfs_commit_data *cdata = data;
abcb7bfc 1708
37e9ed22 1709 ff_layout_commit_record_layoutstats_done(task, cdata);
d67ae825
TH		 1710 rpc_count_iostats_metrics(task,
1711 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1712}
1713
4d0ac221
TM		 1714static void ff_layout_commit_release(void *data)
1715{
1716 struct nfs_commit_data *cdata = data;
1717
1718 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1719 pnfs_generic_commit_release(data);
1720}
1721
d67ae825
TH		 1722static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1723 .rpc_call_prepare = ff_layout_read_prepare_v3,
1724 .rpc_call_done = ff_layout_read_call_done,
1725 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1726 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1727};
1728
1729static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1730 .rpc_call_prepare = ff_layout_read_prepare_v4,
1731 .rpc_call_done = ff_layout_read_call_done,
1732 .rpc_count_stats = ff_layout_read_count_stats,
4d0ac221 1733 .rpc_release = ff_layout_read_release,
d67ae825
TH		 1734};
1735
1736static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1737 .rpc_call_prepare = ff_layout_write_prepare_v3,
1738 .rpc_call_done = ff_layout_write_call_done,
1739 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1740 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1741};
1742
1743static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1744 .rpc_call_prepare = ff_layout_write_prepare_v4,
1745 .rpc_call_done = ff_layout_write_call_done,
1746 .rpc_count_stats = ff_layout_write_count_stats,
4d0ac221 1747 .rpc_release = ff_layout_write_release,
d67ae825
TH		 1748};
1749
1750static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1751 .rpc_call_prepare = ff_layout_commit_prepare_v3,
abcb7bfc 1752 .rpc_call_done = ff_layout_commit_done,
d67ae825 1753 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1754 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1755};
1756
1757static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1758 .rpc_call_prepare = ff_layout_commit_prepare_v4,
abcb7bfc 1759 .rpc_call_done = ff_layout_commit_done,
d67ae825 1760 .rpc_count_stats = ff_layout_commit_count_stats,
4d0ac221 1761 .rpc_release = ff_layout_commit_release,
d67ae825
TH		 1762};
1763
1764static enum pnfs_try_status
1765ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1766{
1767 struct pnfs_layout_segment *lseg = hdr->lseg;
1768 struct nfs4_pnfs_ds *ds;
1769 struct rpc_clnt *ds_clnt;
1770 struct rpc_cred *ds_cred;
1771 loff_t offset = hdr->args.offset;
1772 u32 idx = hdr->pgio_mirror_idx;
1773 int vers;
1774 struct nfs_fh *fh;
1775
1776 dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1777 __func__, hdr->inode->i_ino,
1778 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1779
1780 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1781 if (!ds)
1782 goto out_failed;
1783
1784 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1785 hdr->inode);
1786 if (IS_ERR(ds_clnt))
1787 goto out_failed;
1788
1789 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
57f3f4c0 1790 if (!ds_cred)
d67ae825
TH		 1791 goto out_failed;
1792
1793 vers = nfs4_ff_layout_ds_version(lseg, idx);
1794
1795 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1796 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1797
75bf47eb 1798 hdr->pgio_done_cb = ff_layout_read_done_cb;
d67ae825
TH		 1799 atomic_inc(&ds->ds_clp->cl_count);
1800 hdr->ds_clp = ds->ds_clp;
1801 fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1802 if (fh)
1803 hdr->args.fh = fh;
d67ae825
TH		 1804 /*
1805 * Note that if we ever decide to split across DSes,
1806 * then we may need to handle dense-like offsets.
1807 */
1808 hdr->args.offset = offset;
1809 hdr->mds_offset = offset;
1810
1811 /* Perform an asynchronous read to ds */
1812 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1813 vers == 3 ? &ff_layout_read_call_ops_v3 :
1814 &ff_layout_read_call_ops_v4,
1815 0, RPC_TASK_SOFTCONN);
57f3f4c0 1816 put_rpccred(ds_cred);
d67ae825
TH		 1817 return PNFS_ATTEMPTED;
1818
1819out_failed:
3b13b4b3 1820 if (ff_layout_avoid_mds_available_ds(lseg))
d67ae825
TH		 1821 return PNFS_TRY_AGAIN;
1822 return PNFS_NOT_ATTEMPTED;
1823}
1824
1825/* Perform async writes. */
1826static enum pnfs_try_status
1827ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1828{
1829 struct pnfs_layout_segment *lseg = hdr->lseg;
1830 struct nfs4_pnfs_ds *ds;
1831 struct rpc_clnt *ds_clnt;
1832 struct rpc_cred *ds_cred;
1833 loff_t offset = hdr->args.offset;
1834 int vers;
1835 struct nfs_fh *fh;
1836 int idx = hdr->pgio_mirror_idx;
1837
1838 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1839 if (!ds)
1840 return PNFS_NOT_ATTEMPTED;
1841
1842 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1843 hdr->inode);
1844 if (IS_ERR(ds_clnt))
1845 return PNFS_NOT_ATTEMPTED;
1846
1847 ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
57f3f4c0 1848 if (!ds_cred)
d67ae825
TH		 1849 return PNFS_NOT_ATTEMPTED;
1850
1851 vers = nfs4_ff_layout_ds_version(lseg, idx);
1852
1853 dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1854 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1855 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1856 vers);
1857
1858 hdr->pgio_done_cb = ff_layout_write_done_cb;
1859 atomic_inc(&ds->ds_clp->cl_count);
1860 hdr->ds_clp = ds->ds_clp;
1861 hdr->ds_commit_idx = idx;
1862 fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1863 if (fh)
1864 hdr->args.fh = fh;
1865
1866 /*
1867 * Note that if we ever decide to split across DSes,
1868 * then we may need to handle dense-like offsets.
1869 */
1870 hdr->args.offset = offset;
1871
1872 /* Perform an asynchronous write */
1873 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1874 vers == 3 ? &ff_layout_write_call_ops_v3 :
1875 &ff_layout_write_call_ops_v4,
1876 sync, RPC_TASK_SOFTCONN);
57f3f4c0 1877 put_rpccred(ds_cred);
d67ae825
TH		 1878 return PNFS_ATTEMPTED;
1879}
1880
d67ae825
TH		 1881static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1882{
1883 return i;
1884}
1885
1886static struct nfs_fh *
1887select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1888{
1889 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1890
1891 /* FIXME: Assume that there is only one NFS version available
1892 * for the DS.
1893 */
1894 return &flseg->mirror_array[i]->fh_versions[0];
1895}
1896
1897static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1898{
1899 struct pnfs_layout_segment *lseg = data->lseg;
1900 struct nfs4_pnfs_ds *ds;
1901 struct rpc_clnt *ds_clnt;
1902 struct rpc_cred *ds_cred;
1903 u32 idx;
57f3f4c0 1904 int vers, ret;
d67ae825
TH		 1905 struct nfs_fh *fh;
1906
1907 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1908 ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1909 if (!ds)
1910 goto out_err;
1911
1912 ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1913 data->inode);
1914 if (IS_ERR(ds_clnt))
1915 goto out_err;
1916
1917 ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
57f3f4c0 1918 if (!ds_cred)
d67ae825
TH		 1919 goto out_err;
1920
1921 vers = nfs4_ff_layout_ds_version(lseg, idx);
1922
1923 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1924 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1925 vers);
1926 data->commit_done_cb = ff_layout_commit_done_cb;
1927 data->cred = ds_cred;
1928 atomic_inc(&ds->ds_clp->cl_count);
1929 data->ds_clp = ds->ds_clp;
1930 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1931 if (fh)
1932 data->args.fh = fh;
abcb7bfc 1933
57f3f4c0 1934 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
d67ae825
TH		 1935 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1936 &ff_layout_commit_call_ops_v4,
1937 how, RPC_TASK_SOFTCONN);
57f3f4c0
JL		 1938 put_rpccred(ds_cred);
1939 return ret;
d67ae825
TH		 1940out_err:
1941 pnfs_generic_prepare_to_resend_writes(data);
1942 pnfs_generic_commit_release(data);
1943 return -EAGAIN;
1944}
1945
1946static int
1947ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1948 int how, struct nfs_commit_info *cinfo)
1949{
1950 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1951 ff_layout_initiate_commit);
1952}
1953
1954static struct pnfs_ds_commit_info *
1955ff_layout_get_ds_info(struct inode *inode)
1956{
1957 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1958
1959 if (layout == NULL)
1960 return NULL;
1961
1962 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1963}
1964
1965static void
fc87701b 1966ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
d67ae825
TH		 1967{
1968 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1969 id_node));
1970}
1971
1972static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1973 struct xdr_stream *xdr,
1974 const struct nfs4_layoutreturn_args *args)
1975{
1976 struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1977 __be32 *start;
1978 int count = 0, ret = 0;
1979
1980 start = xdr_reserve_space(xdr, 4);
1981 if (unlikely(!start))
1982 return -E2BIG;
1983
1984 /* This assume we always return _ALL_ layouts */
1985 spin_lock(&hdr->plh_inode->i_lock);
1986 ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1987 spin_unlock(&hdr->plh_inode->i_lock);
1988
1989 *start = cpu_to_be32(count);
1990
1991 return ret;
1992}
1993
1994/* report nothing for now */
1995static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1996 struct xdr_stream *xdr,
1997 const struct nfs4_layoutreturn_args *args)
1998{
1999 __be32 *p;
2000
2001 p = xdr_reserve_space(xdr, 4);
2002 if (likely(p))
2003 *p = cpu_to_be32(0);
2004}
2005
2006static struct nfs4_deviceid_node *
2007ff_layout_alloc_deviceid_node(struct nfs_server *server,
2008 struct pnfs_device *pdev, gfp_t gfp_flags)
2009{
2010 struct nfs4_ff_layout_ds *dsaddr;
2011
2012 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2013 if (!dsaddr)
2014 return NULL;
2015 return &dsaddr->id_node;
2016}
2017
2018static void
94e5c571 2019ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
d67ae825
TH		 2020 const struct nfs4_layoutreturn_args *args)
2021{
94e5c571 2022 struct pnfs_layout_hdr *lo = args->layout;
d67ae825
TH		 2023 struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
2024 __be32 *start;
2025
2026 dprintk("%s: Begin\n", __func__);
2027 start = xdr_reserve_space(xdr, 4);
2028 BUG_ON(!start);
2029
082fa37d 2030 ff_layout_encode_ioerr(flo, xdr, args);
d67ae825 2031 ff_layout_encode_iostats(flo, xdr, args);
082fa37d 2032
d67ae825
TH		 2033 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2034 dprintk("%s: Return\n", __func__);
2035}
2036
27c43064
PT		 2037static int
2038ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2039{
2040 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2041
2042 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2043}
2044
2045static size_t
2046ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2047 const int buflen)
2048{
2049 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2050 const struct in6_addr *addr = &sin6->sin6_addr;
2051
2052 /*
2053 * RFC 4291, Section 2.2.2
2054 *
2055 * Shorthanded ANY address
2056 */
2057 if (ipv6_addr_any(addr))
2058 return snprintf(buf, buflen, "::");
2059
2060 /*
2061 * RFC 4291, Section 2.2.2
2062 *
2063 * Shorthanded loopback address
2064 */
2065 if (ipv6_addr_loopback(addr))
2066 return snprintf(buf, buflen, "::1");
2067
2068 /*
2069 * RFC 4291, Section 2.2.3
2070 *
2071 * Special presentation address format for mapped v4
2072 * addresses.
2073 */
2074 if (ipv6_addr_v4mapped(addr))
2075 return snprintf(buf, buflen, "::ffff:%pI4",
2076 &addr->s6_addr32[3]);
2077
2078 /*
2079 * RFC 4291, Section 2.2.1
2080 */
2081 return snprintf(buf, buflen, "%pI6c", addr);
2082}
2083
2084/* Derived from rpc_sockaddr2uaddr */
2085static void
2086ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2087{
2088 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2089 char portbuf[RPCBIND_MAXUADDRPLEN];
2090 char addrbuf[RPCBIND_MAXUADDRLEN];
2091 char *netid;
2092 unsigned short port;
2093 int len, netid_len;
2094 __be32 *p;
2095
2096 switch (sap->sa_family) {
2097 case AF_INET:
2098 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2099 return;
2100 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2101 netid = "tcp";
2102 netid_len = 3;
2103 break;
2104 case AF_INET6:
2105 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2106 return;
2107 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2108 netid = "tcp6";
2109 netid_len = 4;
2110 break;
2111 default:
2112 /* we only support tcp and tcp6 */
2113 WARN_ON_ONCE(1);
2114 return;
2115 }
2116
2117 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2118 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2119
2120 p = xdr_reserve_space(xdr, 4 + netid_len);
2121 xdr_encode_opaque(p, netid, netid_len);
2122
2123 p = xdr_reserve_space(xdr, 4 + len);
2124 xdr_encode_opaque(p, addrbuf, len);
2125}
2126
2127static void
2128ff_layout_encode_nfstime(struct xdr_stream *xdr,
2129 ktime_t t)
2130{
2131 struct timespec64 ts;
2132 __be32 *p;
2133
2134 p = xdr_reserve_space(xdr, 12);
2135 ts = ktime_to_timespec64(t);
2136 p = xdr_encode_hyper(p, ts.tv_sec);
2137 *p++ = cpu_to_be32(ts.tv_nsec);
2138}
2139
2140static void
2141ff_layout_encode_io_latency(struct xdr_stream *xdr,
2142 struct nfs4_ff_io_stat *stat)
2143{
2144 __be32 *p;
2145
2146 p = xdr_reserve_space(xdr, 5 * 8);
2147 p = xdr_encode_hyper(p, stat->ops_requested);
2148 p = xdr_encode_hyper(p, stat->bytes_requested);
2149 p = xdr_encode_hyper(p, stat->ops_completed);
2150 p = xdr_encode_hyper(p, stat->bytes_completed);
2151 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2152 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2153 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2154}
2155
2156static void
2157ff_layout_encode_layoutstats(struct xdr_stream *xdr,
2158 struct nfs42_layoutstat_args *args,
2159 struct nfs42_layoutstat_devinfo *devinfo)
2160{
2161 struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
2162 struct nfs4_pnfs_ds_addr *da;
2163 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2164 struct nfs_fh *fh = &mirror->fh_versions[0];
2165 __be32 *p, *start;
2166
2167 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2168 dprintk("%s: DS %s: encoding address %s\n",
2169 __func__, ds->ds_remotestr, da->da_remotestr);
2170 /* layoutupdate length */
2171 start = xdr_reserve_space(xdr, 4);
2172 /* netaddr4 */
2173 ff_layout_encode_netaddr(xdr, da);
2174 /* nfs_fh4 */
2175 p = xdr_reserve_space(xdr, 4 + fh->size);
2176 xdr_encode_opaque(p, fh->data, fh->size);
2177 /* ff_io_latency4 read */
2178 spin_lock(&mirror->lock);
2179 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2180 /* ff_io_latency4 write */
2181 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2182 spin_unlock(&mirror->lock);
2183 /* nfstime4 */
2184 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2185 /* bool */
2186 p = xdr_reserve_space(xdr, 4);
2187 *p = cpu_to_be32(false);
2188
2189 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2190}
2191
266d12d4 2192static int
ad4dc53e 2193ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
266d12d4
TM		 2194 struct pnfs_layout_hdr *lo,
2195 int dev_limit)
ad4dc53e 2196{
266d12d4 2197 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
ad4dc53e
PT		 2198 struct nfs4_ff_layout_mirror *mirror;
2199 struct nfs4_deviceid_node *dev;
2200 struct nfs42_layoutstat_devinfo *devinfo;
266d12d4 2201 int i = 0;
ad4dc53e 2202
266d12d4
TM		 2203 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2204 if (i >= dev_limit)
ad4dc53e 2205 break;
266d12d4 2206 if (!mirror->mirror_ds)
27c43064 2207 continue;
06946c6a
TM		 2208 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2209 continue;
266d12d4
TM		 2210 /* mirror refcount put in cleanup_layoutstats */
2211 if (!atomic_inc_not_zero(&mirror->ref))
2212 continue;
2213 dev = &mirror->mirror_ds->id_node;
2214 devinfo = &args->devinfo[i];
ad4dc53e 2215 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
266d12d4
TM		 2216 devinfo->offset = 0;
2217 devinfo->length = NFS4_MAX_UINT64;
d099d7b8 2218 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
ad4dc53e 2219 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
d099d7b8 2220 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
ad4dc53e
PT		 2221 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2222 devinfo->layout_type = LAYOUT_FLEX_FILES;
27c43064
PT		 2223 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
2224 devinfo->layout_private = mirror;
ad4dc53e 2225
266d12d4 2226 i++;
ad4dc53e 2227 }
266d12d4 2228 return i;
ad4dc53e
PT		 2229}
2230
2231static int
2232ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2233{
266d12d4
TM		 2234 struct nfs4_flexfile_layout *ff_layout;
2235 struct nfs4_ff_layout_mirror *mirror;
ad4dc53e
PT		 2236 int dev_count = 0;
2237
2238 spin_lock(&args->inode->i_lock);
266d12d4
TM		 2239 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2240 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2241 if (atomic_read(&mirror->ref) != 0)
2242 dev_count ++;
ad4dc53e
PT		 2243 }
2244 spin_unlock(&args->inode->i_lock);
2245 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2246 if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
2247 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
2248 __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
2249 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2250 }
266d12d4 2251 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
ad4dc53e
PT		 2252 if (!args->devinfo)
2253 return -ENOMEM;
2254
ad4dc53e 2255 spin_lock(&args->inode->i_lock);
266d12d4
TM		 2256 args->num_dev = ff_layout_mirror_prepare_stats(args,
2257 &ff_layout->generic_hdr, dev_count);
ad4dc53e 2258 spin_unlock(&args->inode->i_lock);
46c98c6d
TM		 2259 if (!args->num_dev) {
2260 kfree(args->devinfo);
2261 args->devinfo = NULL;
2262 return -ENOENT;
2263 }
ad4dc53e
PT		 2264
2265 return 0;
2266}
2267
27c43064
PT		 2268static void
2269ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
2270{
2271 struct nfs4_ff_layout_mirror *mirror;
2272 int i;
2273
2274 for (i = 0; i < data->args.num_dev; i++) {
2275 mirror = data->args.devinfo[i].layout_private;
2276 data->args.devinfo[i].layout_private = NULL;
0b7baf94 2277 ff_layout_put_mirror(mirror);
27c43064
PT		 2278 }
2279}
2280
d67ae825
TH		 2281static struct pnfs_layoutdriver_type flexfilelayout_type = {
2282 .id = LAYOUT_FLEX_FILES,
2283 .name = "LAYOUT_FLEX_FILES",
2284 .owner = THIS_MODULE,
2285 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2286 .free_layout_hdr = ff_layout_free_layout_hdr,
2287 .alloc_lseg = ff_layout_alloc_lseg,
2288 .free_lseg = ff_layout_free_lseg,
0762ed2c 2289 .add_lseg = ff_layout_add_lseg,
d67ae825
TH		 2290 .pg_read_ops = &ff_layout_pg_read_ops,
2291 .pg_write_ops = &ff_layout_pg_write_ops,
2292 .get_ds_info = ff_layout_get_ds_info,
fc87701b 2293 .free_deviceid_node = ff_layout_free_deviceid_node,
338d00cf 2294 .mark_request_commit = pnfs_layout_mark_request_commit,
d67ae825
TH		 2295 .clear_request_commit = pnfs_generic_clear_request_commit,
2296 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2297 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2298 .commit_pagelist = ff_layout_commit_pagelist,
2299 .read_pagelist = ff_layout_read_pagelist,
2300 .write_pagelist = ff_layout_write_pagelist,
2301 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2302 .encode_layoutreturn = ff_layout_encode_layoutreturn,
5bb89b47 2303 .sync = pnfs_nfs_generic_sync,
ad4dc53e 2304 .prepare_layoutstats = ff_layout_prepare_layoutstats,
27c43064 2305 .cleanup_layoutstats = ff_layout_cleanup_layoutstats,
d67ae825
TH		 2306};
2307
2308static int __init nfs4flexfilelayout_init(void)
2309{
2310 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2311 __func__);
90a0be00
JL		 2312 if (!ff_zero_group) {
2313 ff_zero_group = groups_alloc(0);
2314 if (!ff_zero_group)
2315 return -ENOMEM;
2316 }
d67ae825
TH		 2317 return pnfs_register_layoutdriver(&flexfilelayout_type);
2318}
2319
2320static void __exit nfs4flexfilelayout_exit(void)
2321{
2322 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2323 __func__);
2324 pnfs_unregister_layoutdriver(&flexfilelayout_type);
90a0be00
JL		 2325 if (ff_zero_group) {
2326 put_group_info(ff_zero_group);
2327 ff_zero_group = NULL;
2328 }
d67ae825
TH		 2329}
2330
2331MODULE_ALIAS("nfs-layouttype4-4");
2332
2333MODULE_LICENSE("GPL");
2334MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2335
2336module_init(nfs4flexfilelayout_init);
2337module_exit(nfs4flexfilelayout_exit);
Ubuntu Jammy Linux kernel mirror