1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Module for pnfs flexfile layout driver.
4 *
5 * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
6 *
7 * Tao Peng <bergwolf@primarydata.com>
8 */
9
10 #include <linux/nfs_fs.h>
11 #include <linux/nfs_mount.h>
12 #include <linux/nfs_page.h>
13 #include <linux/module.h>
14 #include <linux/sched/mm.h>
15
16 #include <linux/sunrpc/metrics.h>
17
18 #include "flexfilelayout.h"
19 #include "../nfs4session.h"
20 #include "../nfs4idmap.h"
21 #include "../internal.h"
22 #include "../delegation.h"
23 #include "../nfs4trace.h"
24 #include "../iostat.h"
25 #include "../nfs.h"
26 #include "../nfs42.h"
27
28 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
29
30 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
31 #define FF_LAYOUTRETURN_MAXERR 20
32
33 static unsigned short io_maxretrans;
34
35 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
36 struct nfs_pgio_header *hdr);
37 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
38 struct nfs42_layoutstat_devinfo *devinfo,
39 int dev_limit);
40 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
41 const struct nfs42_layoutstat_devinfo *devinfo,
42 struct nfs4_ff_layout_mirror *mirror);
43
44 static struct pnfs_layout_hdr *
ff_layout_alloc_layout_hdr(struct inode * inode,gfp_t gfp_flags)45 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
46 {
47 struct nfs4_flexfile_layout *ffl;
48
49 ffl = kzalloc(sizeof(*ffl), gfp_flags);
50 if (ffl) {
51 INIT_LIST_HEAD(&ffl->error_list);
52 INIT_LIST_HEAD(&ffl->mirrors);
53 ffl->last_report_time = ktime_get();
54 return &ffl->generic_hdr;
55 } else
56 return NULL;
57 }
58
59 static void
ff_layout_free_layout_hdr(struct pnfs_layout_hdr * lo)60 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
61 {
62 struct nfs4_ff_layout_ds_err *err, *n;
63
64 list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
65 list) {
66 list_del(&err->list);
67 kfree(err);
68 }
69 kfree(FF_LAYOUT_FROM_HDR(lo));
70 }
71
decode_pnfs_stateid(struct xdr_stream * xdr,nfs4_stateid * stateid)72 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
73 {
74 __be32 *p;
75
76 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
77 if (unlikely(p == NULL))
78 return -ENOBUFS;
79 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
80 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
81 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
82 p[0], p[1], p[2], p[3]);
83 return 0;
84 }
85
decode_deviceid(struct xdr_stream * xdr,struct nfs4_deviceid * devid)86 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
87 {
88 __be32 *p;
89
90 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
91 if (unlikely(!p))
92 return -ENOBUFS;
93 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
94 nfs4_print_deviceid(devid);
95 return 0;
96 }
97
decode_nfs_fh(struct xdr_stream * xdr,struct nfs_fh * fh)98 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
99 {
100 __be32 *p;
101
102 p = xdr_inline_decode(xdr, 4);
103 if (unlikely(!p))
104 return -ENOBUFS;
105 fh->size = be32_to_cpup(p++);
106 if (fh->size > NFS_MAXFHSIZE) {
107 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
108 fh->size);
109 return -EOVERFLOW;
110 }
111 /* fh.data */
112 p = xdr_inline_decode(xdr, fh->size);
113 if (unlikely(!p))
114 return -ENOBUFS;
115 memcpy(&fh->data, p, fh->size);
116 dprintk("%s: fh len %d\n", __func__, fh->size);
117
118 return 0;
119 }
120
121 /*
122 * Currently only stringified uids and gids are accepted.
123 * I.e., kerberos is not supported to the DSes, so no pricipals.
124 *
125 * That means that one common function will suffice, but when
126 * principals are added, this should be split to accomodate
127 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
128 */
129 static int
decode_name(struct xdr_stream * xdr,u32 * id)130 decode_name(struct xdr_stream *xdr, u32 *id)
131 {
132 __be32 *p;
133 int len;
134
135 /* opaque_length(4)*/
136 p = xdr_inline_decode(xdr, 4);
137 if (unlikely(!p))
138 return -ENOBUFS;
139 len = be32_to_cpup(p++);
140 if (len < 0)
141 return -EINVAL;
142
143 dprintk("%s: len %u\n", __func__, len);
144
145 /* opaque body */
146 p = xdr_inline_decode(xdr, len);
147 if (unlikely(!p))
148 return -ENOBUFS;
149
150 if (!nfs_map_string_to_numeric((char *)p, len, id))
151 return -EINVAL;
152
153 return 0;
154 }
155
ff_mirror_match_fh(const struct nfs4_ff_layout_mirror * m1,const struct nfs4_ff_layout_mirror * m2)156 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
157 const struct nfs4_ff_layout_mirror *m2)
158 {
159 int i, j;
160
161 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
162 return false;
163 for (i = 0; i < m1->fh_versions_cnt; i++) {
164 bool found_fh = false;
165 for (j = 0; j < m2->fh_versions_cnt; j++) {
166 if (nfs_compare_fh(&m1->fh_versions[i],
167 &m2->fh_versions[j]) == 0) {
168 found_fh = true;
169 break;
170 }
171 }
172 if (!found_fh)
173 return false;
174 }
175 return true;
176 }
177
178 static struct nfs4_ff_layout_mirror *
ff_layout_add_mirror(struct pnfs_layout_hdr * lo,struct nfs4_ff_layout_mirror * mirror)179 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
180 struct nfs4_ff_layout_mirror *mirror)
181 {
182 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
183 struct nfs4_ff_layout_mirror *pos;
184 struct inode *inode = lo->plh_inode;
185
186 spin_lock(&inode->i_lock);
187 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
188 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
189 continue;
190 if (!ff_mirror_match_fh(mirror, pos))
191 continue;
192 if (refcount_inc_not_zero(&pos->ref)) {
193 spin_unlock(&inode->i_lock);
194 return pos;
195 }
196 }
197 list_add(&mirror->mirrors, &ff_layout->mirrors);
198 mirror->layout = lo;
199 spin_unlock(&inode->i_lock);
200 return mirror;
201 }
202
203 static void
ff_layout_remove_mirror(struct nfs4_ff_layout_mirror * mirror)204 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
205 {
206 struct inode *inode;
207 if (mirror->layout == NULL)
208 return;
209 inode = mirror->layout->plh_inode;
210 spin_lock(&inode->i_lock);
211 list_del(&mirror->mirrors);
212 spin_unlock(&inode->i_lock);
213 mirror->layout = NULL;
214 }
215
ff_layout_alloc_mirror(gfp_t gfp_flags)216 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
217 {
218 struct nfs4_ff_layout_mirror *mirror;
219
220 mirror = kzalloc(sizeof(*mirror), gfp_flags);
221 if (mirror != NULL) {
222 spin_lock_init(&mirror->lock);
223 refcount_set(&mirror->ref, 1);
224 INIT_LIST_HEAD(&mirror->mirrors);
225 }
226 return mirror;
227 }
228
ff_layout_free_mirror(struct nfs4_ff_layout_mirror * mirror)229 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
230 {
231 const struct cred *cred;
232
233 ff_layout_remove_mirror(mirror);
234 kfree(mirror->fh_versions);
235 cred = rcu_access_pointer(mirror->ro_cred);
236 put_cred(cred);
237 cred = rcu_access_pointer(mirror->rw_cred);
238 put_cred(cred);
239 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
240 kfree(mirror);
241 }
242
ff_layout_put_mirror(struct nfs4_ff_layout_mirror * mirror)243 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
244 {
245 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
246 ff_layout_free_mirror(mirror);
247 }
248
ff_layout_free_mirror_array(struct nfs4_ff_layout_segment * fls)249 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
250 {
251 int i;
252
253 if (fls->mirror_array) {
254 for (i = 0; i < fls->mirror_array_cnt; i++) {
255 /* normally mirror_ds is freed in
256 * .free_deviceid_node but we still do it here
257 * for .alloc_lseg error path */
258 ff_layout_put_mirror(fls->mirror_array[i]);
259 }
260 kfree(fls->mirror_array);
261 fls->mirror_array = NULL;
262 }
263 }
264
ff_layout_check_layout(struct nfs4_layoutget_res * lgr)265 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
266 {
267 int ret = 0;
268
269 dprintk("--> %s\n", __func__);
270
271 /* FIXME: remove this check when layout segment support is added */
272 if (lgr->range.offset != 0 ||
273 lgr->range.length != NFS4_MAX_UINT64) {
274 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
275 __func__);
276 ret = -EINVAL;
277 }
278
279 dprintk("--> %s returns %d\n", __func__, ret);
280 return ret;
281 }
282
_ff_layout_free_lseg(struct nfs4_ff_layout_segment * fls)283 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
284 {
285 if (fls) {
286 ff_layout_free_mirror_array(fls);
287 kfree(fls);
288 }
289 }
290
291 static bool
ff_lseg_range_is_after(const struct pnfs_layout_range * l1,const struct pnfs_layout_range * l2)292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
293 const struct pnfs_layout_range *l2)
294 {
295 u64 end1, end2;
296
297 if (l1->iomode != l2->iomode)
298 return l1->iomode != IOMODE_READ;
299 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
300 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
301 if (end1 < l2->offset)
302 return false;
303 if (end2 < l1->offset)
304 return true;
305 return l2->offset <= l1->offset;
306 }
307
308 static bool
ff_lseg_merge(struct pnfs_layout_segment * new,struct pnfs_layout_segment * old)309 ff_lseg_merge(struct pnfs_layout_segment *new,
310 struct pnfs_layout_segment *old)
311 {
312 u64 new_end, old_end;
313
314 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
315 return false;
316 if (new->pls_range.iomode != old->pls_range.iomode)
317 return false;
318 old_end = pnfs_calc_offset_end(old->pls_range.offset,
319 old->pls_range.length);
320 if (old_end < new->pls_range.offset)
321 return false;
322 new_end = pnfs_calc_offset_end(new->pls_range.offset,
323 new->pls_range.length);
324 if (new_end < old->pls_range.offset)
325 return false;
326
327 /* Mergeable: copy info from 'old' to 'new' */
328 if (new_end < old_end)
329 new_end = old_end;
330 if (new->pls_range.offset < old->pls_range.offset)
331 new->pls_range.offset = old->pls_range.offset;
332 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
333 new_end);
334 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
335 set_bit(NFS_LSEG_ROC, &new->pls_flags);
336 return true;
337 }
338
339 static void
ff_layout_add_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg,struct list_head * free_me)340 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
341 struct pnfs_layout_segment *lseg,
342 struct list_head *free_me)
343 {
344 pnfs_generic_layout_insert_lseg(lo, lseg,
345 ff_lseg_range_is_after,
346 ff_lseg_merge,
347 free_me);
348 }
349
ff_layout_sort_mirrors(struct nfs4_ff_layout_segment * fls)350 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
351 {
352 int i, j;
353
354 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
355 for (j = i + 1; j < fls->mirror_array_cnt; j++)
356 if (fls->mirror_array[i]->efficiency <
357 fls->mirror_array[j]->efficiency)
358 swap(fls->mirror_array[i],
359 fls->mirror_array[j]);
360 }
361 }
362
363 static struct pnfs_layout_segment *
ff_layout_alloc_lseg(struct pnfs_layout_hdr * lh,struct nfs4_layoutget_res * lgr,gfp_t gfp_flags)364 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
365 struct nfs4_layoutget_res *lgr,
366 gfp_t gfp_flags)
367 {
368 struct pnfs_layout_segment *ret;
369 struct nfs4_ff_layout_segment *fls = NULL;
370 struct xdr_stream stream;
371 struct xdr_buf buf;
372 struct page *scratch;
373 u64 stripe_unit;
374 u32 mirror_array_cnt;
375 __be32 *p;
376 int i, rc;
377
378 dprintk("--> %s\n", __func__);
379 scratch = alloc_page(gfp_flags);
380 if (!scratch)
381 return ERR_PTR(-ENOMEM);
382
383 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
384 lgr->layoutp->len);
385 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
386
387 /* stripe unit and mirror_array_cnt */
388 rc = -EIO;
389 p = xdr_inline_decode(&stream, 8 + 4);
390 if (!p)
391 goto out_err_free;
392
393 p = xdr_decode_hyper(p, &stripe_unit);
394 mirror_array_cnt = be32_to_cpup(p++);
395 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
396 stripe_unit, mirror_array_cnt);
397
398 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
399 mirror_array_cnt == 0)
400 goto out_err_free;
401
402 rc = -ENOMEM;
403 fls = kzalloc(sizeof(*fls), gfp_flags);
404 if (!fls)
405 goto out_err_free;
406
407 fls->mirror_array_cnt = mirror_array_cnt;
408 fls->stripe_unit = stripe_unit;
409 fls->mirror_array = kcalloc(fls->mirror_array_cnt,
410 sizeof(fls->mirror_array[0]), gfp_flags);
411 if (fls->mirror_array == NULL)
412 goto out_err_free;
413
414 for (i = 0; i < fls->mirror_array_cnt; i++) {
415 struct nfs4_ff_layout_mirror *mirror;
416 struct cred *kcred;
417 const struct cred __rcu *cred;
418 kuid_t uid;
419 kgid_t gid;
420 u32 ds_count, fh_count, id;
421 int j;
422
423 rc = -EIO;
424 p = xdr_inline_decode(&stream, 4);
425 if (!p)
426 goto out_err_free;
427 ds_count = be32_to_cpup(p);
428
429 /* FIXME: allow for striping? */
430 if (ds_count != 1)
431 goto out_err_free;
432
433 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
434 if (fls->mirror_array[i] == NULL) {
435 rc = -ENOMEM;
436 goto out_err_free;
437 }
438
439 fls->mirror_array[i]->ds_count = ds_count;
440
441 /* deviceid */
442 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
443 if (rc)
444 goto out_err_free;
445
446 /* efficiency */
447 rc = -EIO;
448 p = xdr_inline_decode(&stream, 4);
449 if (!p)
450 goto out_err_free;
451 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
452
453 /* stateid */
454 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
455 if (rc)
456 goto out_err_free;
457
458 /* fh */
459 rc = -EIO;
460 p = xdr_inline_decode(&stream, 4);
461 if (!p)
462 goto out_err_free;
463 fh_count = be32_to_cpup(p);
464
465 fls->mirror_array[i]->fh_versions =
466 kcalloc(fh_count, sizeof(struct nfs_fh),
467 gfp_flags);
468 if (fls->mirror_array[i]->fh_versions == NULL) {
469 rc = -ENOMEM;
470 goto out_err_free;
471 }
472
473 for (j = 0; j < fh_count; j++) {
474 rc = decode_nfs_fh(&stream,
475 &fls->mirror_array[i]->fh_versions[j]);
476 if (rc)
477 goto out_err_free;
478 }
479
480 fls->mirror_array[i]->fh_versions_cnt = fh_count;
481
482 /* user */
483 rc = decode_name(&stream, &id);
484 if (rc)
485 goto out_err_free;
486
487 uid = make_kuid(&init_user_ns, id);
488
489 /* group */
490 rc = decode_name(&stream, &id);
491 if (rc)
492 goto out_err_free;
493
494 gid = make_kgid(&init_user_ns, id);
495
496 if (gfp_flags & __GFP_FS)
497 kcred = prepare_kernel_cred(NULL);
498 else {
499 unsigned int nofs_flags = memalloc_nofs_save();
500 kcred = prepare_kernel_cred(NULL);
501 memalloc_nofs_restore(nofs_flags);
502 }
503 rc = -ENOMEM;
504 if (!kcred)
505 goto out_err_free;
506 kcred->fsuid = uid;
507 kcred->fsgid = gid;
508 cred = RCU_INITIALIZER(kcred);
509
510 if (lgr->range.iomode == IOMODE_READ)
511 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
512 else
513 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
514
515 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
516 if (mirror != fls->mirror_array[i]) {
517 /* swap cred ptrs so free_mirror will clean up old */
518 if (lgr->range.iomode == IOMODE_READ) {
519 cred = xchg(&mirror->ro_cred, cred);
520 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
521 } else {
522 cred = xchg(&mirror->rw_cred, cred);
523 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
524 }
525 ff_layout_free_mirror(fls->mirror_array[i]);
526 fls->mirror_array[i] = mirror;
527 }
528
529 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
530 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
531 from_kuid(&init_user_ns, uid),
532 from_kgid(&init_user_ns, gid));
533 }
534
535 p = xdr_inline_decode(&stream, 4);
536 if (!p)
537 goto out_sort_mirrors;
538 fls->flags = be32_to_cpup(p);
539
540 p = xdr_inline_decode(&stream, 4);
541 if (!p)
542 goto out_sort_mirrors;
543 for (i=0; i < fls->mirror_array_cnt; i++)
544 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
545
546 out_sort_mirrors:
547 ff_layout_sort_mirrors(fls);
548 rc = ff_layout_check_layout(lgr);
549 if (rc)
550 goto out_err_free;
551 ret = &fls->generic_hdr;
552 dprintk("<-- %s (success)\n", __func__);
553 out_free_page:
554 __free_page(scratch);
555 return ret;
556 out_err_free:
557 _ff_layout_free_lseg(fls);
558 ret = ERR_PTR(rc);
559 dprintk("<-- %s (%d)\n", __func__, rc);
560 goto out_free_page;
561 }
562
ff_layout_has_rw_segments(struct pnfs_layout_hdr * layout)563 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
564 {
565 struct pnfs_layout_segment *lseg;
566
567 list_for_each_entry(lseg, &layout->plh_segs, pls_list)
568 if (lseg->pls_range.iomode == IOMODE_RW)
569 return true;
570
571 return false;
572 }
573
574 static void
ff_layout_free_lseg(struct pnfs_layout_segment * lseg)575 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
576 {
577 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
578
579 dprintk("--> %s\n", __func__);
580
581 if (lseg->pls_range.iomode == IOMODE_RW) {
582 struct nfs4_flexfile_layout *ffl;
583 struct inode *inode;
584
585 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
586 inode = ffl->generic_hdr.plh_inode;
587 spin_lock(&inode->i_lock);
588 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
589 ffl->commit_info.nbuckets = 0;
590 kfree(ffl->commit_info.buckets);
591 ffl->commit_info.buckets = NULL;
592 }
593 spin_unlock(&inode->i_lock);
594 }
595 _ff_layout_free_lseg(fls);
596 }
597
598 /* Return 1 until we have multiple lsegs support */
599 static int
ff_layout_get_lseg_count(struct nfs4_ff_layout_segment * fls)600 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
601 {
602 return 1;
603 }
604
605 static void
nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer * timer,ktime_t now)606 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
607 {
608 /* first IO request? */
609 if (atomic_inc_return(&timer->n_ops) == 1) {
610 timer->start_time = now;
611 }
612 }
613
614 static ktime_t
nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer * timer,ktime_t now)615 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
616 {
617 ktime_t start;
618
619 if (atomic_dec_return(&timer->n_ops) < 0)
620 WARN_ON_ONCE(1);
621
622 start = timer->start_time;
623 timer->start_time = now;
624 return ktime_sub(now, start);
625 }
626
627 static bool
nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror * mirror,struct nfs4_ff_layoutstat * layoutstat,ktime_t now)628 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
629 struct nfs4_ff_layoutstat *layoutstat,
630 ktime_t now)
631 {
632 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
633 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
634
635 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
636 if (!mirror->start_time)
637 mirror->start_time = now;
638 if (mirror->report_interval != 0)
639 report_interval = (s64)mirror->report_interval * 1000LL;
640 else if (layoutstats_timer != 0)
641 report_interval = (s64)layoutstats_timer * 1000LL;
642 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
643 report_interval) {
644 ffl->last_report_time = now;
645 return true;
646 }
647
648 return false;
649 }
650
651 static void
nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat * layoutstat,__u64 requested)652 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
653 __u64 requested)
654 {
655 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
656
657 iostat->ops_requested++;
658 iostat->bytes_requested += requested;
659 }
660
661 static void
nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat * layoutstat,__u64 requested,__u64 completed,ktime_t time_completed,ktime_t time_started)662 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
663 __u64 requested,
664 __u64 completed,
665 ktime_t time_completed,
666 ktime_t time_started)
667 {
668 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
669 ktime_t completion_time = ktime_sub(time_completed, time_started);
670 ktime_t timer;
671
672 iostat->ops_completed++;
673 iostat->bytes_completed += completed;
674 iostat->bytes_not_delivered += requested - completed;
675
676 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
677 iostat->total_busy_time =
678 ktime_add(iostat->total_busy_time, timer);
679 iostat->aggregate_completion_time =
680 ktime_add(iostat->aggregate_completion_time,
681 completion_time);
682 }
683
684 static void
nfs4_ff_layout_stat_io_start_read(struct inode * inode,struct nfs4_ff_layout_mirror * mirror,__u64 requested,ktime_t now)685 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
686 struct nfs4_ff_layout_mirror *mirror,
687 __u64 requested, ktime_t now)
688 {
689 bool report;
690
691 spin_lock(&mirror->lock);
692 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
693 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
694 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
695 spin_unlock(&mirror->lock);
696
697 if (report)
698 pnfs_report_layoutstat(inode, GFP_KERNEL);
699 }
700
701 static void
nfs4_ff_layout_stat_io_end_read(struct rpc_task * task,struct nfs4_ff_layout_mirror * mirror,__u64 requested,__u64 completed)702 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
703 struct nfs4_ff_layout_mirror *mirror,
704 __u64 requested,
705 __u64 completed)
706 {
707 spin_lock(&mirror->lock);
708 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
709 requested, completed,
710 ktime_get(), task->tk_start);
711 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
712 spin_unlock(&mirror->lock);
713 }
714
715 static void
nfs4_ff_layout_stat_io_start_write(struct inode * inode,struct nfs4_ff_layout_mirror * mirror,__u64 requested,ktime_t now)716 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
717 struct nfs4_ff_layout_mirror *mirror,
718 __u64 requested, ktime_t now)
719 {
720 bool report;
721
722 spin_lock(&mirror->lock);
723 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
724 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
725 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
726 spin_unlock(&mirror->lock);
727
728 if (report)
729 pnfs_report_layoutstat(inode, GFP_NOIO);
730 }
731
732 static void
nfs4_ff_layout_stat_io_end_write(struct rpc_task * task,struct nfs4_ff_layout_mirror * mirror,__u64 requested,__u64 completed,enum nfs3_stable_how committed)733 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
734 struct nfs4_ff_layout_mirror *mirror,
735 __u64 requested,
736 __u64 completed,
737 enum nfs3_stable_how committed)
738 {
739 if (committed == NFS_UNSTABLE)
740 requested = completed = 0;
741
742 spin_lock(&mirror->lock);
743 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
744 requested, completed, ktime_get(), task->tk_start);
745 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
746 spin_unlock(&mirror->lock);
747 }
748
749 static int
ff_layout_alloc_commit_info(struct pnfs_layout_segment * lseg,struct nfs_commit_info * cinfo,gfp_t gfp_flags)750 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
751 struct nfs_commit_info *cinfo,
752 gfp_t gfp_flags)
753 {
754 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
755 struct pnfs_commit_bucket *buckets;
756 int size;
757
758 if (cinfo->ds->nbuckets != 0) {
759 /* This assumes there is only one RW lseg per file.
760 * To support multiple lseg per file, we need to
761 * change struct pnfs_commit_bucket to allow dynamic
762 * increasing nbuckets.
763 */
764 return 0;
765 }
766
767 size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
768
769 buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
770 gfp_flags);
771 if (!buckets)
772 return -ENOMEM;
773 else {
774 int i;
775
776 spin_lock(&cinfo->inode->i_lock);
777 if (cinfo->ds->nbuckets != 0)
778 kfree(buckets);
779 else {
780 cinfo->ds->buckets = buckets;
781 cinfo->ds->nbuckets = size;
782 for (i = 0; i < size; i++) {
783 INIT_LIST_HEAD(&buckets[i].written);
784 INIT_LIST_HEAD(&buckets[i].committing);
785 /* mark direct verifier as unset */
786 buckets[i].direct_verf.committed =
787 NFS_INVALID_STABLE_HOW;
788 }
789 }
790 spin_unlock(&cinfo->inode->i_lock);
791 return 0;
792 }
793 }
794
795 static void
ff_layout_mark_ds_unreachable(struct pnfs_layout_segment * lseg,int idx)796 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, int idx)
797 {
798 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
799
800 if (devid)
801 nfs4_mark_deviceid_unavailable(devid);
802 }
803
804 static void
ff_layout_mark_ds_reachable(struct pnfs_layout_segment * lseg,int idx)805 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, int idx)
806 {
807 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
808
809 if (devid)
810 nfs4_mark_deviceid_available(devid);
811 }
812
813 static struct nfs4_pnfs_ds *
ff_layout_choose_ds_for_read(struct pnfs_layout_segment * lseg,int start_idx,int * best_idx,bool check_device)814 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
815 int start_idx, int *best_idx,
816 bool check_device)
817 {
818 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
819 struct nfs4_ff_layout_mirror *mirror;
820 struct nfs4_pnfs_ds *ds;
821 bool fail_return = false;
822 int idx;
823
824 /* mirrors are initially sorted by efficiency */
825 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
826 if (idx+1 == fls->mirror_array_cnt)
827 fail_return = !check_device;
828
829 mirror = FF_LAYOUT_COMP(lseg, idx);
830 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
831 if (!ds)
832 continue;
833
834 if (check_device &&
835 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
836 continue;
837
838 *best_idx = idx;
839 return ds;
840 }
841
842 return NULL;
843 }
844
845 static struct nfs4_pnfs_ds *
ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment * lseg,int start_idx,int * best_idx)846 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
847 int start_idx, int *best_idx)
848 {
849 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
850 }
851
852 static struct nfs4_pnfs_ds *
ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment * lseg,int start_idx,int * best_idx)853 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
854 int start_idx, int *best_idx)
855 {
856 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
857 }
858
859 static struct nfs4_pnfs_ds *
ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment * lseg,int start_idx,int * best_idx)860 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
861 int start_idx, int *best_idx)
862 {
863 struct nfs4_pnfs_ds *ds;
864
865 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
866 if (ds)
867 return ds;
868 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
869 }
870
871 static void
ff_layout_pg_get_read(struct nfs_pageio_descriptor * pgio,struct nfs_page * req,bool strict_iomode)872 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
873 struct nfs_page *req,
874 bool strict_iomode)
875 {
876 pnfs_put_lseg(pgio->pg_lseg);
877 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
878 nfs_req_openctx(req),
879 0,
880 NFS4_MAX_UINT64,
881 IOMODE_READ,
882 strict_iomode,
883 GFP_KERNEL);
884 if (IS_ERR(pgio->pg_lseg)) {
885 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
886 pgio->pg_lseg = NULL;
887 }
888 }
889
890 static void
ff_layout_pg_init_read(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)891 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
892 struct nfs_page *req)
893 {
894 struct nfs_pgio_mirror *pgm;
895 struct nfs4_ff_layout_mirror *mirror;
896 struct nfs4_pnfs_ds *ds;
897 int ds_idx;
898
899 retry:
900 pnfs_generic_pg_check_layout(pgio);
901 /* Use full layout for now */
902 if (!pgio->pg_lseg) {
903 ff_layout_pg_get_read(pgio, req, false);
904 if (!pgio->pg_lseg)
905 goto out_nolseg;
906 }
907 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
908 ff_layout_pg_get_read(pgio, req, true);
909 if (!pgio->pg_lseg)
910 goto out_nolseg;
911 }
912
913 ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
914 if (!ds) {
915 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
916 goto out_mds;
917 pnfs_put_lseg(pgio->pg_lseg);
918 pgio->pg_lseg = NULL;
919 /* Sleep for 1 second before retrying */
920 ssleep(1);
921 goto retry;
922 }
923
924 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
925
926 pgio->pg_mirror_idx = ds_idx;
927
928 /* read always uses only one mirror - idx 0 for pgio layer */
929 pgm = &pgio->pg_mirrors[0];
930 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
931
932 if (NFS_SERVER(pgio->pg_inode)->flags &
933 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
934 pgio->pg_maxretrans = io_maxretrans;
935 return;
936 out_nolseg:
937 if (pgio->pg_error < 0)
938 return;
939 out_mds:
940 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
941 0, NFS4_MAX_UINT64, IOMODE_READ,
942 NFS_I(pgio->pg_inode)->layout,
943 pgio->pg_lseg);
944 pnfs_put_lseg(pgio->pg_lseg);
945 pgio->pg_lseg = NULL;
946 pgio->pg_maxretrans = 0;
947 nfs_pageio_reset_read_mds(pgio);
948 }
949
950 static void
ff_layout_pg_init_write(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)951 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
952 struct nfs_page *req)
953 {
954 struct nfs4_ff_layout_mirror *mirror;
955 struct nfs_pgio_mirror *pgm;
956 struct nfs_commit_info cinfo;
957 struct nfs4_pnfs_ds *ds;
958 int i;
959 int status;
960
961 retry:
962 pnfs_generic_pg_check_layout(pgio);
963 if (!pgio->pg_lseg) {
964 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
965 nfs_req_openctx(req),
966 0,
967 NFS4_MAX_UINT64,
968 IOMODE_RW,
969 false,
970 GFP_NOFS);
971 if (IS_ERR(pgio->pg_lseg)) {
972 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
973 pgio->pg_lseg = NULL;
974 return;
975 }
976 }
977 /* If no lseg, fall back to write through mds */
978 if (pgio->pg_lseg == NULL)
979 goto out_mds;
980
981 nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
982 status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
983 if (status < 0)
984 goto out_mds;
985
986 /* Use a direct mapping of ds_idx to pgio mirror_idx */
987 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
988 goto out_eagain;
989
990 for (i = 0; i < pgio->pg_mirror_count; i++) {
991 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
992 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
993 if (!ds) {
994 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
995 goto out_mds;
996 pnfs_put_lseg(pgio->pg_lseg);
997 pgio->pg_lseg = NULL;
998 /* Sleep for 1 second before retrying */
999 ssleep(1);
1000 goto retry;
1001 }
1002 pgm = &pgio->pg_mirrors[i];
1003 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
1004 }
1005
1006 if (NFS_SERVER(pgio->pg_inode)->flags &
1007 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
1008 pgio->pg_maxretrans = io_maxretrans;
1009 return;
1010 out_eagain:
1011 pnfs_generic_pg_cleanup(pgio);
1012 pgio->pg_error = -EAGAIN;
1013 return;
1014 out_mds:
1015 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
1016 0, NFS4_MAX_UINT64, IOMODE_RW,
1017 NFS_I(pgio->pg_inode)->layout,
1018 pgio->pg_lseg);
1019 pnfs_put_lseg(pgio->pg_lseg);
1020 pgio->pg_lseg = NULL;
1021 pgio->pg_maxretrans = 0;
1022 nfs_pageio_reset_write_mds(pgio);
1023 pgio->pg_error = -EAGAIN;
1024 }
1025
1026 static unsigned int
ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)1027 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
1028 struct nfs_page *req)
1029 {
1030 if (!pgio->pg_lseg) {
1031 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
1032 nfs_req_openctx(req),
1033 0,
1034 NFS4_MAX_UINT64,
1035 IOMODE_RW,
1036 false,
1037 GFP_NOFS);
1038 if (IS_ERR(pgio->pg_lseg)) {
1039 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
1040 pgio->pg_lseg = NULL;
1041 goto out;
1042 }
1043 }
1044 if (pgio->pg_lseg)
1045 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1046
1047 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
1048 0, NFS4_MAX_UINT64, IOMODE_RW,
1049 NFS_I(pgio->pg_inode)->layout,
1050 pgio->pg_lseg);
1051 /* no lseg means that pnfs is not in use, so no mirroring here */
1052 nfs_pageio_reset_write_mds(pgio);
1053 out:
1054 return 1;
1055 }
1056
1057 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1058 .pg_init = ff_layout_pg_init_read,
1059 .pg_test = pnfs_generic_pg_test,
1060 .pg_doio = pnfs_generic_pg_readpages,
1061 .pg_cleanup = pnfs_generic_pg_cleanup,
1062 };
1063
1064 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1065 .pg_init = ff_layout_pg_init_write,
1066 .pg_test = pnfs_generic_pg_test,
1067 .pg_doio = pnfs_generic_pg_writepages,
1068 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1069 .pg_cleanup = pnfs_generic_pg_cleanup,
1070 };
1071
ff_layout_reset_write(struct nfs_pgio_header * hdr,bool retry_pnfs)1072 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1073 {
1074 struct rpc_task *task = &hdr->task;
1075
1076 pnfs_layoutcommit_inode(hdr->inode, false);
1077
1078 if (retry_pnfs) {
1079 dprintk("%s Reset task %5u for i/o through pNFS "
1080 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1081 hdr->task.tk_pid,
1082 hdr->inode->i_sb->s_id,
1083 (unsigned long long)NFS_FILEID(hdr->inode),
1084 hdr->args.count,
1085 (unsigned long long)hdr->args.offset);
1086
1087 hdr->completion_ops->reschedule_io(hdr);
1088 return;
1089 }
1090
1091 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1092 dprintk("%s Reset task %5u for i/o through MDS "
1093 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1094 hdr->task.tk_pid,
1095 hdr->inode->i_sb->s_id,
1096 (unsigned long long)NFS_FILEID(hdr->inode),
1097 hdr->args.count,
1098 (unsigned long long)hdr->args.offset);
1099
1100 trace_pnfs_mds_fallback_write_done(hdr->inode,
1101 hdr->args.offset, hdr->args.count,
1102 IOMODE_RW, NFS_I(hdr->inode)->layout,
1103 hdr->lseg);
1104 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1105 }
1106 }
1107
ff_layout_reset_read(struct nfs_pgio_header * hdr)1108 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1109 {
1110 struct rpc_task *task = &hdr->task;
1111
1112 pnfs_layoutcommit_inode(hdr->inode, false);
1113
1114 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1115 dprintk("%s Reset task %5u for i/o through MDS "
1116 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1117 hdr->task.tk_pid,
1118 hdr->inode->i_sb->s_id,
1119 (unsigned long long)NFS_FILEID(hdr->inode),
1120 hdr->args.count,
1121 (unsigned long long)hdr->args.offset);
1122
1123 trace_pnfs_mds_fallback_read_done(hdr->inode,
1124 hdr->args.offset, hdr->args.count,
1125 IOMODE_READ, NFS_I(hdr->inode)->layout,
1126 hdr->lseg);
1127 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1128 }
1129 }
1130
ff_layout_async_handle_error_v4(struct rpc_task * task,struct nfs4_state * state,struct nfs_client * clp,struct pnfs_layout_segment * lseg,int idx)1131 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1132 struct nfs4_state *state,
1133 struct nfs_client *clp,
1134 struct pnfs_layout_segment *lseg,
1135 int idx)
1136 {
1137 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1138 struct inode *inode = lo->plh_inode;
1139 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1140 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1141
1142 switch (task->tk_status) {
1143 case -NFS4ERR_BADSESSION:
1144 case -NFS4ERR_BADSLOT:
1145 case -NFS4ERR_BAD_HIGH_SLOT:
1146 case -NFS4ERR_DEADSESSION:
1147 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1148 case -NFS4ERR_SEQ_FALSE_RETRY:
1149 case -NFS4ERR_SEQ_MISORDERED:
1150 dprintk("%s ERROR %d, Reset session. Exchangeid "
1151 "flags 0x%x\n", __func__, task->tk_status,
1152 clp->cl_exchange_flags);
1153 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1154 break;
1155 case -NFS4ERR_DELAY:
1156 case -NFS4ERR_GRACE:
1157 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1158 break;
1159 case -NFS4ERR_RETRY_UNCACHED_REP:
1160 break;
1161 /* Invalidate Layout errors */
1162 case -NFS4ERR_PNFS_NO_LAYOUT:
1163 case -ESTALE: /* mapped NFS4ERR_STALE */
1164 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1165 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1166 case -NFS4ERR_FHEXPIRED:
1167 case -NFS4ERR_WRONG_TYPE:
1168 dprintk("%s Invalid layout error %d\n", __func__,
1169 task->tk_status);
1170 /*
1171 * Destroy layout so new i/o will get a new layout.
1172 * Layout will not be destroyed until all current lseg
1173 * references are put. Mark layout as invalid to resend failed
1174 * i/o and all i/o waiting on the slot table to the MDS until
1175 * layout is destroyed and a new valid layout is obtained.
1176 */
1177 pnfs_destroy_layout(NFS_I(inode));
1178 rpc_wake_up(&tbl->slot_tbl_waitq);
1179 goto reset;
1180 /* RPC connection errors */
1181 case -ECONNREFUSED:
1182 case -EHOSTDOWN:
1183 case -EHOSTUNREACH:
1184 case -ENETUNREACH:
1185 case -EIO:
1186 case -ETIMEDOUT:
1187 case -EPIPE:
1188 dprintk("%s DS connection error %d\n", __func__,
1189 task->tk_status);
1190 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1191 &devid->deviceid);
1192 rpc_wake_up(&tbl->slot_tbl_waitq);
1193 /* fall through */
1194 default:
1195 if (ff_layout_avoid_mds_available_ds(lseg))
1196 return -NFS4ERR_RESET_TO_PNFS;
1197 reset:
1198 dprintk("%s Retry through MDS. Error %d\n", __func__,
1199 task->tk_status);
1200 return -NFS4ERR_RESET_TO_MDS;
1201 }
1202 task->tk_status = 0;
1203 return -EAGAIN;
1204 }
1205
1206 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
ff_layout_async_handle_error_v3(struct rpc_task * task,struct pnfs_layout_segment * lseg,int idx)1207 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1208 struct pnfs_layout_segment *lseg,
1209 int idx)
1210 {
1211 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1212
1213 switch (task->tk_status) {
1214 /* File access problems. Don't mark the device as unavailable */
1215 case -EACCES:
1216 case -ESTALE:
1217 case -EISDIR:
1218 case -EBADHANDLE:
1219 case -ELOOP:
1220 case -ENOSPC:
1221 break;
1222 case -EJUKEBOX:
1223 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1224 goto out_retry;
1225 default:
1226 dprintk("%s DS connection error %d\n", __func__,
1227 task->tk_status);
1228 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1229 &devid->deviceid);
1230 }
1231 /* FIXME: Need to prevent infinite looping here. */
1232 return -NFS4ERR_RESET_TO_PNFS;
1233 out_retry:
1234 task->tk_status = 0;
1235 rpc_restart_call_prepare(task);
1236 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1237 return -EAGAIN;
1238 }
1239
ff_layout_async_handle_error(struct rpc_task * task,struct nfs4_state * state,struct nfs_client * clp,struct pnfs_layout_segment * lseg,int idx)1240 static int ff_layout_async_handle_error(struct rpc_task *task,
1241 struct nfs4_state *state,
1242 struct nfs_client *clp,
1243 struct pnfs_layout_segment *lseg,
1244 int idx)
1245 {
1246 int vers = clp->cl_nfs_mod->rpc_vers->number;
1247
1248 if (task->tk_status >= 0) {
1249 ff_layout_mark_ds_reachable(lseg, idx);
1250 return 0;
1251 }
1252
1253 /* Handle the case of an invalid layout segment */
1254 if (!pnfs_is_valid_lseg(lseg))
1255 return -NFS4ERR_RESET_TO_PNFS;
1256
1257 switch (vers) {
1258 case 3:
1259 return ff_layout_async_handle_error_v3(task, lseg, idx);
1260 case 4:
1261 return ff_layout_async_handle_error_v4(task, state, clp,
1262 lseg, idx);
1263 default:
1264 /* should never happen */
1265 WARN_ON_ONCE(1);
1266 return 0;
1267 }
1268 }
1269
ff_layout_io_track_ds_error(struct pnfs_layout_segment * lseg,int idx,u64 offset,u64 length,u32 status,int opnum,int error)1270 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1271 int idx, u64 offset, u64 length,
1272 u32 status, int opnum, int error)
1273 {
1274 struct nfs4_ff_layout_mirror *mirror;
1275 int err;
1276
1277 if (status == 0) {
1278 switch (error) {
1279 case -ETIMEDOUT:
1280 case -EPFNOSUPPORT:
1281 case -EPROTONOSUPPORT:
1282 case -EOPNOTSUPP:
1283 case -EINVAL:
1284 case -ECONNREFUSED:
1285 case -ECONNRESET:
1286 case -EHOSTDOWN:
1287 case -EHOSTUNREACH:
1288 case -ENETUNREACH:
1289 case -EADDRINUSE:
1290 case -ENOBUFS:
1291 case -EPIPE:
1292 case -EPERM:
1293 status = NFS4ERR_NXIO;
1294 break;
1295 case -EACCES:
1296 status = NFS4ERR_ACCESS;
1297 break;
1298 default:
1299 return;
1300 }
1301 }
1302
1303 switch (status) {
1304 case NFS4ERR_DELAY:
1305 case NFS4ERR_GRACE:
1306 return;
1307 default:
1308 break;
1309 }
1310
1311 mirror = FF_LAYOUT_COMP(lseg, idx);
1312 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1313 mirror, offset, length, status, opnum,
1314 GFP_NOIO);
1315 if (status == NFS4ERR_NXIO)
1316 ff_layout_mark_ds_unreachable(lseg, idx);
1317 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1318 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1319 }
1320
1321 /* NFS_PROTO call done callback routines */
ff_layout_read_done_cb(struct rpc_task * task,struct nfs_pgio_header * hdr)1322 static int ff_layout_read_done_cb(struct rpc_task *task,
1323 struct nfs_pgio_header *hdr)
1324 {
1325 int new_idx = hdr->pgio_mirror_idx;
1326 int err;
1327
1328 trace_nfs4_pnfs_read(hdr, task->tk_status);
1329 if (task->tk_status < 0)
1330 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1331 hdr->args.offset, hdr->args.count,
1332 hdr->res.op_status, OP_READ,
1333 task->tk_status);
1334 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1335 hdr->ds_clp, hdr->lseg,
1336 hdr->pgio_mirror_idx);
1337
1338 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1339 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1340 switch (err) {
1341 case -NFS4ERR_RESET_TO_PNFS:
1342 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1343 hdr->pgio_mirror_idx + 1,
1344 &new_idx))
1345 goto out_layouterror;
1346 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1347 return task->tk_status;
1348 case -NFS4ERR_RESET_TO_MDS:
1349 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1350 return task->tk_status;
1351 case -EAGAIN:
1352 goto out_eagain;
1353 }
1354
1355 return 0;
1356 out_layouterror:
1357 ff_layout_read_record_layoutstats_done(task, hdr);
1358 ff_layout_send_layouterror(hdr->lseg);
1359 hdr->pgio_mirror_idx = new_idx;
1360 out_eagain:
1361 rpc_restart_call_prepare(task);
1362 return -EAGAIN;
1363 }
1364
1365 static bool
ff_layout_need_layoutcommit(struct pnfs_layout_segment * lseg)1366 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1367 {
1368 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1369 }
1370
1371 /*
1372 * We reference the rpc_cred of the first WRITE that triggers the need for
1373 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1374 * rfc5661 is not clear about which credential should be used.
1375 *
1376 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1377 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1378 * we always send layoutcommit after DS writes.
1379 */
1380 static void
ff_layout_set_layoutcommit(struct inode * inode,struct pnfs_layout_segment * lseg,loff_t end_offset)1381 ff_layout_set_layoutcommit(struct inode *inode,
1382 struct pnfs_layout_segment *lseg,
1383 loff_t end_offset)
1384 {
1385 if (!ff_layout_need_layoutcommit(lseg))
1386 return;
1387
1388 pnfs_set_layoutcommit(inode, lseg, end_offset);
1389 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1390 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1391 }
1392
ff_layout_read_record_layoutstats_start(struct rpc_task * task,struct nfs_pgio_header * hdr)1393 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1394 struct nfs_pgio_header *hdr)
1395 {
1396 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1397 return;
1398 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1399 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1400 hdr->args.count,
1401 task->tk_start);
1402 }
1403
ff_layout_read_record_layoutstats_done(struct rpc_task * task,struct nfs_pgio_header * hdr)1404 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1405 struct nfs_pgio_header *hdr)
1406 {
1407 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1408 return;
1409 nfs4_ff_layout_stat_io_end_read(task,
1410 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1411 hdr->args.count,
1412 hdr->res.count);
1413 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1414 }
1415
ff_layout_read_prepare_common(struct rpc_task * task,struct nfs_pgio_header * hdr)1416 static int ff_layout_read_prepare_common(struct rpc_task *task,
1417 struct nfs_pgio_header *hdr)
1418 {
1419 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1420 rpc_exit(task, -EIO);
1421 return -EIO;
1422 }
1423
1424 ff_layout_read_record_layoutstats_start(task, hdr);
1425 return 0;
1426 }
1427
1428 /*
1429 * Call ops for the async read/write cases
1430 * In the case of dense layouts, the offset needs to be reset to its
1431 * original value.
1432 */
ff_layout_read_prepare_v3(struct rpc_task * task,void * data)1433 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1434 {
1435 struct nfs_pgio_header *hdr = data;
1436
1437 if (ff_layout_read_prepare_common(task, hdr))
1438 return;
1439
1440 rpc_call_start(task);
1441 }
1442
ff_layout_read_prepare_v4(struct rpc_task * task,void * data)1443 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1444 {
1445 struct nfs_pgio_header *hdr = data;
1446
1447 if (nfs4_setup_sequence(hdr->ds_clp,
1448 &hdr->args.seq_args,
1449 &hdr->res.seq_res,
1450 task))
1451 return;
1452
1453 ff_layout_read_prepare_common(task, hdr);
1454 }
1455
ff_layout_read_call_done(struct rpc_task * task,void * data)1456 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1457 {
1458 struct nfs_pgio_header *hdr = data;
1459
1460 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1461
1462 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1463 task->tk_status == 0) {
1464 nfs4_sequence_done(task, &hdr->res.seq_res);
1465 return;
1466 }
1467
1468 /* Note this may cause RPC to be resent */
1469 hdr->mds_ops->rpc_call_done(task, hdr);
1470 }
1471
ff_layout_read_count_stats(struct rpc_task * task,void * data)1472 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1473 {
1474 struct nfs_pgio_header *hdr = data;
1475
1476 ff_layout_read_record_layoutstats_done(task, hdr);
1477 rpc_count_iostats_metrics(task,
1478 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1479 }
1480
ff_layout_read_release(void * data)1481 static void ff_layout_read_release(void *data)
1482 {
1483 struct nfs_pgio_header *hdr = data;
1484
1485 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1486 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1487 ff_layout_send_layouterror(hdr->lseg);
1488 pnfs_read_resend_pnfs(hdr);
1489 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1490 ff_layout_reset_read(hdr);
1491 pnfs_generic_rw_release(data);
1492 }
1493
1494
ff_layout_write_done_cb(struct rpc_task * task,struct nfs_pgio_header * hdr)1495 static int ff_layout_write_done_cb(struct rpc_task *task,
1496 struct nfs_pgio_header *hdr)
1497 {
1498 loff_t end_offs = 0;
1499 int err;
1500
1501 trace_nfs4_pnfs_write(hdr, task->tk_status);
1502 if (task->tk_status < 0)
1503 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1504 hdr->args.offset, hdr->args.count,
1505 hdr->res.op_status, OP_WRITE,
1506 task->tk_status);
1507 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1508 hdr->ds_clp, hdr->lseg,
1509 hdr->pgio_mirror_idx);
1510
1511 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1512 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1513 switch (err) {
1514 case -NFS4ERR_RESET_TO_PNFS:
1515 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1516 return task->tk_status;
1517 case -NFS4ERR_RESET_TO_MDS:
1518 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1519 return task->tk_status;
1520 case -EAGAIN:
1521 return -EAGAIN;
1522 }
1523
1524 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1525 hdr->res.verf->committed == NFS_DATA_SYNC)
1526 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1527
1528 /* Note: if the write is unstable, don't set end_offs until commit */
1529 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1530
1531 /* zero out fattr since we don't care DS attr at all */
1532 hdr->fattr.valid = 0;
1533 if (task->tk_status >= 0)
1534 nfs_writeback_update_inode(hdr);
1535
1536 return 0;
1537 }
1538
ff_layout_commit_done_cb(struct rpc_task * task,struct nfs_commit_data * data)1539 static int ff_layout_commit_done_cb(struct rpc_task *task,
1540 struct nfs_commit_data *data)
1541 {
1542 int err;
1543
1544 trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1545 if (task->tk_status < 0)
1546 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1547 data->args.offset, data->args.count,
1548 data->res.op_status, OP_COMMIT,
1549 task->tk_status);
1550 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1551 data->lseg, data->ds_commit_index);
1552
1553 switch (err) {
1554 case -NFS4ERR_RESET_TO_PNFS:
1555 pnfs_generic_prepare_to_resend_writes(data);
1556 return -EAGAIN;
1557 case -NFS4ERR_RESET_TO_MDS:
1558 pnfs_generic_prepare_to_resend_writes(data);
1559 return -EAGAIN;
1560 case -EAGAIN:
1561 rpc_restart_call_prepare(task);
1562 return -EAGAIN;
1563 }
1564
1565 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1566
1567 return 0;
1568 }
1569
ff_layout_write_record_layoutstats_start(struct rpc_task * task,struct nfs_pgio_header * hdr)1570 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1571 struct nfs_pgio_header *hdr)
1572 {
1573 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1574 return;
1575 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1576 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1577 hdr->args.count,
1578 task->tk_start);
1579 }
1580
ff_layout_write_record_layoutstats_done(struct rpc_task * task,struct nfs_pgio_header * hdr)1581 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1582 struct nfs_pgio_header *hdr)
1583 {
1584 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1585 return;
1586 nfs4_ff_layout_stat_io_end_write(task,
1587 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1588 hdr->args.count, hdr->res.count,
1589 hdr->res.verf->committed);
1590 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1591 }
1592
ff_layout_write_prepare_common(struct rpc_task * task,struct nfs_pgio_header * hdr)1593 static int ff_layout_write_prepare_common(struct rpc_task *task,
1594 struct nfs_pgio_header *hdr)
1595 {
1596 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1597 rpc_exit(task, -EIO);
1598 return -EIO;
1599 }
1600
1601 ff_layout_write_record_layoutstats_start(task, hdr);
1602 return 0;
1603 }
1604
ff_layout_write_prepare_v3(struct rpc_task * task,void * data)1605 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1606 {
1607 struct nfs_pgio_header *hdr = data;
1608
1609 if (ff_layout_write_prepare_common(task, hdr))
1610 return;
1611
1612 rpc_call_start(task);
1613 }
1614
ff_layout_write_prepare_v4(struct rpc_task * task,void * data)1615 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1616 {
1617 struct nfs_pgio_header *hdr = data;
1618
1619 if (nfs4_setup_sequence(hdr->ds_clp,
1620 &hdr->args.seq_args,
1621 &hdr->res.seq_res,
1622 task))
1623 return;
1624
1625 ff_layout_write_prepare_common(task, hdr);
1626 }
1627
ff_layout_write_call_done(struct rpc_task * task,void * data)1628 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1629 {
1630 struct nfs_pgio_header *hdr = data;
1631
1632 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1633 task->tk_status == 0) {
1634 nfs4_sequence_done(task, &hdr->res.seq_res);
1635 return;
1636 }
1637
1638 /* Note this may cause RPC to be resent */
1639 hdr->mds_ops->rpc_call_done(task, hdr);
1640 }
1641
ff_layout_write_count_stats(struct rpc_task * task,void * data)1642 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1643 {
1644 struct nfs_pgio_header *hdr = data;
1645
1646 ff_layout_write_record_layoutstats_done(task, hdr);
1647 rpc_count_iostats_metrics(task,
1648 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1649 }
1650
ff_layout_write_release(void * data)1651 static void ff_layout_write_release(void *data)
1652 {
1653 struct nfs_pgio_header *hdr = data;
1654
1655 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1656 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1657 ff_layout_send_layouterror(hdr->lseg);
1658 ff_layout_reset_write(hdr, true);
1659 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1660 ff_layout_reset_write(hdr, false);
1661 pnfs_generic_rw_release(data);
1662 }
1663
ff_layout_commit_record_layoutstats_start(struct rpc_task * task,struct nfs_commit_data * cdata)1664 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1665 struct nfs_commit_data *cdata)
1666 {
1667 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1668 return;
1669 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1670 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1671 0, task->tk_start);
1672 }
1673
ff_layout_commit_record_layoutstats_done(struct rpc_task * task,struct nfs_commit_data * cdata)1674 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1675 struct nfs_commit_data *cdata)
1676 {
1677 struct nfs_page *req;
1678 __u64 count = 0;
1679
1680 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1681 return;
1682
1683 if (task->tk_status == 0) {
1684 list_for_each_entry(req, &cdata->pages, wb_list)
1685 count += req->wb_bytes;
1686 }
1687 nfs4_ff_layout_stat_io_end_write(task,
1688 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1689 count, count, NFS_FILE_SYNC);
1690 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1691 }
1692
ff_layout_commit_prepare_common(struct rpc_task * task,struct nfs_commit_data * cdata)1693 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1694 struct nfs_commit_data *cdata)
1695 {
1696 ff_layout_commit_record_layoutstats_start(task, cdata);
1697 }
1698
ff_layout_commit_prepare_v3(struct rpc_task * task,void * data)1699 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1700 {
1701 ff_layout_commit_prepare_common(task, data);
1702 rpc_call_start(task);
1703 }
1704
ff_layout_commit_prepare_v4(struct rpc_task * task,void * data)1705 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1706 {
1707 struct nfs_commit_data *wdata = data;
1708
1709 if (nfs4_setup_sequence(wdata->ds_clp,
1710 &wdata->args.seq_args,
1711 &wdata->res.seq_res,
1712 task))
1713 return;
1714 ff_layout_commit_prepare_common(task, data);
1715 }
1716
ff_layout_commit_done(struct rpc_task * task,void * data)1717 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1718 {
1719 pnfs_generic_write_commit_done(task, data);
1720 }
1721
ff_layout_commit_count_stats(struct rpc_task * task,void * data)1722 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1723 {
1724 struct nfs_commit_data *cdata = data;
1725
1726 ff_layout_commit_record_layoutstats_done(task, cdata);
1727 rpc_count_iostats_metrics(task,
1728 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1729 }
1730
ff_layout_commit_release(void * data)1731 static void ff_layout_commit_release(void *data)
1732 {
1733 struct nfs_commit_data *cdata = data;
1734
1735 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1736 pnfs_generic_commit_release(data);
1737 }
1738
1739 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1740 .rpc_call_prepare = ff_layout_read_prepare_v3,
1741 .rpc_call_done = ff_layout_read_call_done,
1742 .rpc_count_stats = ff_layout_read_count_stats,
1743 .rpc_release = ff_layout_read_release,
1744 };
1745
1746 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1747 .rpc_call_prepare = ff_layout_read_prepare_v4,
1748 .rpc_call_done = ff_layout_read_call_done,
1749 .rpc_count_stats = ff_layout_read_count_stats,
1750 .rpc_release = ff_layout_read_release,
1751 };
1752
1753 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1754 .rpc_call_prepare = ff_layout_write_prepare_v3,
1755 .rpc_call_done = ff_layout_write_call_done,
1756 .rpc_count_stats = ff_layout_write_count_stats,
1757 .rpc_release = ff_layout_write_release,
1758 };
1759
1760 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1761 .rpc_call_prepare = ff_layout_write_prepare_v4,
1762 .rpc_call_done = ff_layout_write_call_done,
1763 .rpc_count_stats = ff_layout_write_count_stats,
1764 .rpc_release = ff_layout_write_release,
1765 };
1766
1767 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1768 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1769 .rpc_call_done = ff_layout_commit_done,
1770 .rpc_count_stats = ff_layout_commit_count_stats,
1771 .rpc_release = ff_layout_commit_release,
1772 };
1773
1774 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1775 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1776 .rpc_call_done = ff_layout_commit_done,
1777 .rpc_count_stats = ff_layout_commit_count_stats,
1778 .rpc_release = ff_layout_commit_release,
1779 };
1780
1781 static enum pnfs_try_status
ff_layout_read_pagelist(struct nfs_pgio_header * hdr)1782 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1783 {
1784 struct pnfs_layout_segment *lseg = hdr->lseg;
1785 struct nfs4_pnfs_ds *ds;
1786 struct rpc_clnt *ds_clnt;
1787 struct nfs4_ff_layout_mirror *mirror;
1788 const struct cred *ds_cred;
1789 loff_t offset = hdr->args.offset;
1790 u32 idx = hdr->pgio_mirror_idx;
1791 int vers;
1792 struct nfs_fh *fh;
1793
1794 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1795 __func__, hdr->inode->i_ino,
1796 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1797
1798 mirror = FF_LAYOUT_COMP(lseg, idx);
1799 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1800 if (!ds)
1801 goto out_failed;
1802
1803 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1804 hdr->inode);
1805 if (IS_ERR(ds_clnt))
1806 goto out_failed;
1807
1808 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1809 if (!ds_cred)
1810 goto out_failed;
1811
1812 vers = nfs4_ff_layout_ds_version(mirror);
1813
1814 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1815 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1816
1817 hdr->pgio_done_cb = ff_layout_read_done_cb;
1818 refcount_inc(&ds->ds_clp->cl_count);
1819 hdr->ds_clp = ds->ds_clp;
1820 fh = nfs4_ff_layout_select_ds_fh(mirror);
1821 if (fh)
1822 hdr->args.fh = fh;
1823
1824 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1825
1826 /*
1827 * Note that if we ever decide to split across DSes,
1828 * then we may need to handle dense-like offsets.
1829 */
1830 hdr->args.offset = offset;
1831 hdr->mds_offset = offset;
1832
1833 /* Perform an asynchronous read to ds */
1834 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1835 vers == 3 ? &ff_layout_read_call_ops_v3 :
1836 &ff_layout_read_call_ops_v4,
1837 0, RPC_TASK_SOFTCONN);
1838 put_cred(ds_cred);
1839 return PNFS_ATTEMPTED;
1840
1841 out_failed:
1842 if (ff_layout_avoid_mds_available_ds(lseg))
1843 return PNFS_TRY_AGAIN;
1844 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1845 hdr->args.offset, hdr->args.count,
1846 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1847 return PNFS_NOT_ATTEMPTED;
1848 }
1849
1850 /* Perform async writes. */
1851 static enum pnfs_try_status
ff_layout_write_pagelist(struct nfs_pgio_header * hdr,int sync)1852 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1853 {
1854 struct pnfs_layout_segment *lseg = hdr->lseg;
1855 struct nfs4_pnfs_ds *ds;
1856 struct rpc_clnt *ds_clnt;
1857 struct nfs4_ff_layout_mirror *mirror;
1858 const struct cred *ds_cred;
1859 loff_t offset = hdr->args.offset;
1860 int vers;
1861 struct nfs_fh *fh;
1862 int idx = hdr->pgio_mirror_idx;
1863
1864 mirror = FF_LAYOUT_COMP(lseg, idx);
1865 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1866 if (!ds)
1867 goto out_failed;
1868
1869 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1870 hdr->inode);
1871 if (IS_ERR(ds_clnt))
1872 goto out_failed;
1873
1874 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1875 if (!ds_cred)
1876 goto out_failed;
1877
1878 vers = nfs4_ff_layout_ds_version(mirror);
1879
1880 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1881 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1882 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1883 vers);
1884
1885 hdr->pgio_done_cb = ff_layout_write_done_cb;
1886 refcount_inc(&ds->ds_clp->cl_count);
1887 hdr->ds_clp = ds->ds_clp;
1888 hdr->ds_commit_idx = idx;
1889 fh = nfs4_ff_layout_select_ds_fh(mirror);
1890 if (fh)
1891 hdr->args.fh = fh;
1892
1893 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1894
1895 /*
1896 * Note that if we ever decide to split across DSes,
1897 * then we may need to handle dense-like offsets.
1898 */
1899 hdr->args.offset = offset;
1900
1901 /* Perform an asynchronous write */
1902 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1903 vers == 3 ? &ff_layout_write_call_ops_v3 :
1904 &ff_layout_write_call_ops_v4,
1905 sync, RPC_TASK_SOFTCONN);
1906 put_cred(ds_cred);
1907 return PNFS_ATTEMPTED;
1908
1909 out_failed:
1910 if (ff_layout_avoid_mds_available_ds(lseg))
1911 return PNFS_TRY_AGAIN;
1912 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1913 hdr->args.offset, hdr->args.count,
1914 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1915 return PNFS_NOT_ATTEMPTED;
1916 }
1917
calc_ds_index_from_commit(struct pnfs_layout_segment * lseg,u32 i)1918 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1919 {
1920 return i;
1921 }
1922
1923 static struct nfs_fh *
select_ds_fh_from_commit(struct pnfs_layout_segment * lseg,u32 i)1924 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1925 {
1926 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1927
1928 /* FIXME: Assume that there is only one NFS version available
1929 * for the DS.
1930 */
1931 return &flseg->mirror_array[i]->fh_versions[0];
1932 }
1933
ff_layout_initiate_commit(struct nfs_commit_data * data,int how)1934 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1935 {
1936 struct pnfs_layout_segment *lseg = data->lseg;
1937 struct nfs4_pnfs_ds *ds;
1938 struct rpc_clnt *ds_clnt;
1939 struct nfs4_ff_layout_mirror *mirror;
1940 const struct cred *ds_cred;
1941 u32 idx;
1942 int vers, ret;
1943 struct nfs_fh *fh;
1944
1945 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1946 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1947 goto out_err;
1948
1949 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1950 mirror = FF_LAYOUT_COMP(lseg, idx);
1951 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1952 if (!ds)
1953 goto out_err;
1954
1955 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1956 data->inode);
1957 if (IS_ERR(ds_clnt))
1958 goto out_err;
1959
1960 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1961 if (!ds_cred)
1962 goto out_err;
1963
1964 vers = nfs4_ff_layout_ds_version(mirror);
1965
1966 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1967 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1968 vers);
1969 data->commit_done_cb = ff_layout_commit_done_cb;
1970 data->cred = ds_cred;
1971 refcount_inc(&ds->ds_clp->cl_count);
1972 data->ds_clp = ds->ds_clp;
1973 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1974 if (fh)
1975 data->args.fh = fh;
1976
1977 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1978 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1979 &ff_layout_commit_call_ops_v4,
1980 how, RPC_TASK_SOFTCONN);
1981 put_cred(ds_cred);
1982 return ret;
1983 out_err:
1984 pnfs_generic_prepare_to_resend_writes(data);
1985 pnfs_generic_commit_release(data);
1986 return -EAGAIN;
1987 }
1988
1989 static int
ff_layout_commit_pagelist(struct inode * inode,struct list_head * mds_pages,int how,struct nfs_commit_info * cinfo)1990 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1991 int how, struct nfs_commit_info *cinfo)
1992 {
1993 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1994 ff_layout_initiate_commit);
1995 }
1996
1997 static struct pnfs_ds_commit_info *
ff_layout_get_ds_info(struct inode * inode)1998 ff_layout_get_ds_info(struct inode *inode)
1999 {
2000 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2001
2002 if (layout == NULL)
2003 return NULL;
2004
2005 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2006 }
2007
2008 static void
ff_layout_free_deviceid_node(struct nfs4_deviceid_node * d)2009 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2010 {
2011 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2012 id_node));
2013 }
2014
ff_layout_encode_ioerr(struct xdr_stream * xdr,const struct nfs4_layoutreturn_args * args,const struct nfs4_flexfile_layoutreturn_args * ff_args)2015 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2016 const struct nfs4_layoutreturn_args *args,
2017 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2018 {
2019 __be32 *start;
2020
2021 start = xdr_reserve_space(xdr, 4);
2022 if (unlikely(!start))
2023 return -E2BIG;
2024
2025 *start = cpu_to_be32(ff_args->num_errors);
2026 /* This assume we always return _ALL_ layouts */
2027 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2028 }
2029
2030 static void
encode_opaque_fixed(struct xdr_stream * xdr,const void * buf,size_t len)2031 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2032 {
2033 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2034 }
2035
2036 static void
ff_layout_encode_ff_iostat_head(struct xdr_stream * xdr,const nfs4_stateid * stateid,const struct nfs42_layoutstat_devinfo * devinfo)2037 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2038 const nfs4_stateid *stateid,
2039 const struct nfs42_layoutstat_devinfo *devinfo)
2040 {
2041 __be32 *p;
2042
2043 p = xdr_reserve_space(xdr, 8 + 8);
2044 p = xdr_encode_hyper(p, devinfo->offset);
2045 p = xdr_encode_hyper(p, devinfo->length);
2046 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2047 p = xdr_reserve_space(xdr, 4*8);
2048 p = xdr_encode_hyper(p, devinfo->read_count);
2049 p = xdr_encode_hyper(p, devinfo->read_bytes);
2050 p = xdr_encode_hyper(p, devinfo->write_count);
2051 p = xdr_encode_hyper(p, devinfo->write_bytes);
2052 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2053 }
2054
2055 static void
ff_layout_encode_ff_iostat(struct xdr_stream * xdr,const nfs4_stateid * stateid,const struct nfs42_layoutstat_devinfo * devinfo)2056 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2057 const nfs4_stateid *stateid,
2058 const struct nfs42_layoutstat_devinfo *devinfo)
2059 {
2060 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2061 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2062 devinfo->ld_private.data);
2063 }
2064
2065 /* report nothing for now */
ff_layout_encode_iostats_array(struct xdr_stream * xdr,const struct nfs4_layoutreturn_args * args,struct nfs4_flexfile_layoutreturn_args * ff_args)2066 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2067 const struct nfs4_layoutreturn_args *args,
2068 struct nfs4_flexfile_layoutreturn_args *ff_args)
2069 {
2070 __be32 *p;
2071 int i;
2072
2073 p = xdr_reserve_space(xdr, 4);
2074 *p = cpu_to_be32(ff_args->num_dev);
2075 for (i = 0; i < ff_args->num_dev; i++)
2076 ff_layout_encode_ff_iostat(xdr,
2077 &args->layout->plh_stateid,
2078 &ff_args->devinfo[i]);
2079 }
2080
2081 static void
ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo * devinfo,unsigned int num_entries)2082 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2083 unsigned int num_entries)
2084 {
2085 unsigned int i;
2086
2087 for (i = 0; i < num_entries; i++) {
2088 if (!devinfo[i].ld_private.ops)
2089 continue;
2090 if (!devinfo[i].ld_private.ops->free)
2091 continue;
2092 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2093 }
2094 }
2095
2096 static struct nfs4_deviceid_node *
ff_layout_alloc_deviceid_node(struct nfs_server * server,struct pnfs_device * pdev,gfp_t gfp_flags)2097 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2098 struct pnfs_device *pdev, gfp_t gfp_flags)
2099 {
2100 struct nfs4_ff_layout_ds *dsaddr;
2101
2102 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2103 if (!dsaddr)
2104 return NULL;
2105 return &dsaddr->id_node;
2106 }
2107
2108 static void
ff_layout_encode_layoutreturn(struct xdr_stream * xdr,const void * voidargs,const struct nfs4_xdr_opaque_data * ff_opaque)2109 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2110 const void *voidargs,
2111 const struct nfs4_xdr_opaque_data *ff_opaque)
2112 {
2113 const struct nfs4_layoutreturn_args *args = voidargs;
2114 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2115 struct xdr_buf tmp_buf = {
2116 .head = {
2117 [0] = {
2118 .iov_base = page_address(ff_args->pages[0]),
2119 },
2120 },
2121 .buflen = PAGE_SIZE,
2122 };
2123 struct xdr_stream tmp_xdr;
2124 __be32 *start;
2125
2126 dprintk("%s: Begin\n", __func__);
2127
2128 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2129
2130 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2131 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2132
2133 start = xdr_reserve_space(xdr, 4);
2134 *start = cpu_to_be32(tmp_buf.len);
2135 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2136
2137 dprintk("%s: Return\n", __func__);
2138 }
2139
2140 static void
ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data * args)2141 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2142 {
2143 struct nfs4_flexfile_layoutreturn_args *ff_args;
2144
2145 if (!args->data)
2146 return;
2147 ff_args = args->data;
2148 args->data = NULL;
2149
2150 ff_layout_free_ds_ioerr(&ff_args->errors);
2151 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2152
2153 put_page(ff_args->pages[0]);
2154 kfree(ff_args);
2155 }
2156
2157 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2158 .encode = ff_layout_encode_layoutreturn,
2159 .free = ff_layout_free_layoutreturn,
2160 };
2161
2162 static int
ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args * args)2163 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2164 {
2165 struct nfs4_flexfile_layoutreturn_args *ff_args;
2166 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2167
2168 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2169 if (!ff_args)
2170 goto out_nomem;
2171 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2172 if (!ff_args->pages[0])
2173 goto out_nomem_free;
2174
2175 INIT_LIST_HEAD(&ff_args->errors);
2176 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2177 &args->range, &ff_args->errors,
2178 FF_LAYOUTRETURN_MAXERR);
2179
2180 spin_lock(&args->inode->i_lock);
2181 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2182 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2183 spin_unlock(&args->inode->i_lock);
2184
2185 args->ld_private->ops = &layoutreturn_ops;
2186 args->ld_private->data = ff_args;
2187 return 0;
2188 out_nomem_free:
2189 kfree(ff_args);
2190 out_nomem:
2191 return -ENOMEM;
2192 }
2193
2194 #ifdef CONFIG_NFS_V4_2
2195 void
ff_layout_send_layouterror(struct pnfs_layout_segment * lseg)2196 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2197 {
2198 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2199 struct nfs42_layout_error *errors;
2200 LIST_HEAD(head);
2201
2202 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2203 return;
2204 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2205 if (list_empty(&head))
2206 return;
2207
2208 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2209 sizeof(*errors), GFP_NOFS);
2210 if (errors != NULL) {
2211 const struct nfs4_ff_layout_ds_err *pos;
2212 size_t n = 0;
2213
2214 list_for_each_entry(pos, &head, list) {
2215 errors[n].offset = pos->offset;
2216 errors[n].length = pos->length;
2217 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2218 errors[n].errors[0].dev_id = pos->deviceid;
2219 errors[n].errors[0].status = pos->status;
2220 errors[n].errors[0].opnum = pos->opnum;
2221 n++;
2222 if (!list_is_last(&pos->list, &head) &&
2223 n < NFS42_LAYOUTERROR_MAX)
2224 continue;
2225 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2226 break;
2227 n = 0;
2228 }
2229 kfree(errors);
2230 }
2231 ff_layout_free_ds_ioerr(&head);
2232 }
2233 #else
2234 void
ff_layout_send_layouterror(struct pnfs_layout_segment * lseg)2235 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2236 {
2237 }
2238 #endif
2239
2240 static int
ff_layout_ntop4(const struct sockaddr * sap,char * buf,const size_t buflen)2241 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2242 {
2243 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2244
2245 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2246 }
2247
2248 static size_t
ff_layout_ntop6_noscopeid(const struct sockaddr * sap,char * buf,const int buflen)2249 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2250 const int buflen)
2251 {
2252 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2253 const struct in6_addr *addr = &sin6->sin6_addr;
2254
2255 /*
2256 * RFC 4291, Section 2.2.2
2257 *
2258 * Shorthanded ANY address
2259 */
2260 if (ipv6_addr_any(addr))
2261 return snprintf(buf, buflen, "::");
2262
2263 /*
2264 * RFC 4291, Section 2.2.2
2265 *
2266 * Shorthanded loopback address
2267 */
2268 if (ipv6_addr_loopback(addr))
2269 return snprintf(buf, buflen, "::1");
2270
2271 /*
2272 * RFC 4291, Section 2.2.3
2273 *
2274 * Special presentation address format for mapped v4
2275 * addresses.
2276 */
2277 if (ipv6_addr_v4mapped(addr))
2278 return snprintf(buf, buflen, "::ffff:%pI4",
2279 &addr->s6_addr32[3]);
2280
2281 /*
2282 * RFC 4291, Section 2.2.1
2283 */
2284 return snprintf(buf, buflen, "%pI6c", addr);
2285 }
2286
2287 /* Derived from rpc_sockaddr2uaddr */
2288 static void
ff_layout_encode_netaddr(struct xdr_stream * xdr,struct nfs4_pnfs_ds_addr * da)2289 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2290 {
2291 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2292 char portbuf[RPCBIND_MAXUADDRPLEN];
2293 char addrbuf[RPCBIND_MAXUADDRLEN];
2294 char *netid;
2295 unsigned short port;
2296 int len, netid_len;
2297 __be32 *p;
2298
2299 switch (sap->sa_family) {
2300 case AF_INET:
2301 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2302 return;
2303 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2304 netid = "tcp";
2305 netid_len = 3;
2306 break;
2307 case AF_INET6:
2308 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2309 return;
2310 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2311 netid = "tcp6";
2312 netid_len = 4;
2313 break;
2314 default:
2315 /* we only support tcp and tcp6 */
2316 WARN_ON_ONCE(1);
2317 return;
2318 }
2319
2320 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2321 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2322
2323 p = xdr_reserve_space(xdr, 4 + netid_len);
2324 xdr_encode_opaque(p, netid, netid_len);
2325
2326 p = xdr_reserve_space(xdr, 4 + len);
2327 xdr_encode_opaque(p, addrbuf, len);
2328 }
2329
2330 static void
ff_layout_encode_nfstime(struct xdr_stream * xdr,ktime_t t)2331 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2332 ktime_t t)
2333 {
2334 struct timespec64 ts;
2335 __be32 *p;
2336
2337 p = xdr_reserve_space(xdr, 12);
2338 ts = ktime_to_timespec64(t);
2339 p = xdr_encode_hyper(p, ts.tv_sec);
2340 *p++ = cpu_to_be32(ts.tv_nsec);
2341 }
2342
2343 static void
ff_layout_encode_io_latency(struct xdr_stream * xdr,struct nfs4_ff_io_stat * stat)2344 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2345 struct nfs4_ff_io_stat *stat)
2346 {
2347 __be32 *p;
2348
2349 p = xdr_reserve_space(xdr, 5 * 8);
2350 p = xdr_encode_hyper(p, stat->ops_requested);
2351 p = xdr_encode_hyper(p, stat->bytes_requested);
2352 p = xdr_encode_hyper(p, stat->ops_completed);
2353 p = xdr_encode_hyper(p, stat->bytes_completed);
2354 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2355 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2356 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2357 }
2358
2359 static void
ff_layout_encode_ff_layoutupdate(struct xdr_stream * xdr,const struct nfs42_layoutstat_devinfo * devinfo,struct nfs4_ff_layout_mirror * mirror)2360 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2361 const struct nfs42_layoutstat_devinfo *devinfo,
2362 struct nfs4_ff_layout_mirror *mirror)
2363 {
2364 struct nfs4_pnfs_ds_addr *da;
2365 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2366 struct nfs_fh *fh = &mirror->fh_versions[0];
2367 __be32 *p;
2368
2369 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2370 dprintk("%s: DS %s: encoding address %s\n",
2371 __func__, ds->ds_remotestr, da->da_remotestr);
2372 /* netaddr4 */
2373 ff_layout_encode_netaddr(xdr, da);
2374 /* nfs_fh4 */
2375 p = xdr_reserve_space(xdr, 4 + fh->size);
2376 xdr_encode_opaque(p, fh->data, fh->size);
2377 /* ff_io_latency4 read */
2378 spin_lock(&mirror->lock);
2379 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2380 /* ff_io_latency4 write */
2381 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2382 spin_unlock(&mirror->lock);
2383 /* nfstime4 */
2384 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2385 /* bool */
2386 p = xdr_reserve_space(xdr, 4);
2387 *p = cpu_to_be32(false);
2388 }
2389
2390 static void
ff_layout_encode_layoutstats(struct xdr_stream * xdr,const void * args,const struct nfs4_xdr_opaque_data * opaque)2391 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2392 const struct nfs4_xdr_opaque_data *opaque)
2393 {
2394 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2395 struct nfs42_layoutstat_devinfo, ld_private);
2396 __be32 *start;
2397
2398 /* layoutupdate length */
2399 start = xdr_reserve_space(xdr, 4);
2400 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2401
2402 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2403 }
2404
2405 static void
ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data * opaque)2406 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2407 {
2408 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2409
2410 ff_layout_put_mirror(mirror);
2411 }
2412
2413 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2414 .encode = ff_layout_encode_layoutstats,
2415 .free = ff_layout_free_layoutstats,
2416 };
2417
2418 static int
ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr * lo,struct nfs42_layoutstat_devinfo * devinfo,int dev_limit)2419 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2420 struct nfs42_layoutstat_devinfo *devinfo,
2421 int dev_limit)
2422 {
2423 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2424 struct nfs4_ff_layout_mirror *mirror;
2425 struct nfs4_deviceid_node *dev;
2426 int i = 0;
2427
2428 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2429 if (i >= dev_limit)
2430 break;
2431 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2432 continue;
2433 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2434 continue;
2435 /* mirror refcount put in cleanup_layoutstats */
2436 if (!refcount_inc_not_zero(&mirror->ref))
2437 continue;
2438 dev = &mirror->mirror_ds->id_node;
2439 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2440 devinfo->offset = 0;
2441 devinfo->length = NFS4_MAX_UINT64;
2442 spin_lock(&mirror->lock);
2443 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2444 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2445 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2446 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2447 spin_unlock(&mirror->lock);
2448 devinfo->layout_type = LAYOUT_FLEX_FILES;
2449 devinfo->ld_private.ops = &layoutstat_ops;
2450 devinfo->ld_private.data = mirror;
2451
2452 devinfo++;
2453 i++;
2454 }
2455 return i;
2456 }
2457
2458 static int
ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args * args)2459 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2460 {
2461 struct nfs4_flexfile_layout *ff_layout;
2462 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2463
2464 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2465 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2466 if (!args->devinfo)
2467 return -ENOMEM;
2468
2469 spin_lock(&args->inode->i_lock);
2470 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2471 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2472 &args->devinfo[0], dev_count);
2473 spin_unlock(&args->inode->i_lock);
2474 if (!args->num_dev) {
2475 kfree(args->devinfo);
2476 args->devinfo = NULL;
2477 return -ENOENT;
2478 }
2479
2480 return 0;
2481 }
2482
2483 static int
ff_layout_set_layoutdriver(struct nfs_server * server,const struct nfs_fh * dummy)2484 ff_layout_set_layoutdriver(struct nfs_server *server,
2485 const struct nfs_fh *dummy)
2486 {
2487 #if IS_ENABLED(CONFIG_NFS_V4_2)
2488 server->caps |= NFS_CAP_LAYOUTSTATS;
2489 #endif
2490 return 0;
2491 }
2492
2493 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2494 .id = LAYOUT_FLEX_FILES,
2495 .name = "LAYOUT_FLEX_FILES",
2496 .owner = THIS_MODULE,
2497 .flags = PNFS_LAYOUTGET_ON_OPEN,
2498 .max_layoutget_response = 4096, /* 1 page or so... */
2499 .set_layoutdriver = ff_layout_set_layoutdriver,
2500 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2501 .free_layout_hdr = ff_layout_free_layout_hdr,
2502 .alloc_lseg = ff_layout_alloc_lseg,
2503 .free_lseg = ff_layout_free_lseg,
2504 .add_lseg = ff_layout_add_lseg,
2505 .pg_read_ops = &ff_layout_pg_read_ops,
2506 .pg_write_ops = &ff_layout_pg_write_ops,
2507 .get_ds_info = ff_layout_get_ds_info,
2508 .free_deviceid_node = ff_layout_free_deviceid_node,
2509 .mark_request_commit = pnfs_layout_mark_request_commit,
2510 .clear_request_commit = pnfs_generic_clear_request_commit,
2511 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2512 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2513 .commit_pagelist = ff_layout_commit_pagelist,
2514 .read_pagelist = ff_layout_read_pagelist,
2515 .write_pagelist = ff_layout_write_pagelist,
2516 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2517 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2518 .sync = pnfs_nfs_generic_sync,
2519 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2520 };
2521
nfs4flexfilelayout_init(void)2522 static int __init nfs4flexfilelayout_init(void)
2523 {
2524 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2525 __func__);
2526 return pnfs_register_layoutdriver(&flexfilelayout_type);
2527 }
2528
nfs4flexfilelayout_exit(void)2529 static void __exit nfs4flexfilelayout_exit(void)
2530 {
2531 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2532 __func__);
2533 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2534 }
2535
2536 MODULE_ALIAS("nfs-layouttype4-4");
2537
2538 MODULE_LICENSE("GPL");
2539 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2540
2541 module_init(nfs4flexfilelayout_init);
2542 module_exit(nfs4flexfilelayout_exit);
2543
2544 module_param(io_maxretrans, ushort, 0644);
2545 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2546 "retries an I/O request before returning an error. ");
2547