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 const struct pnfs_commit_ops ff_layout_commit_ops;
36 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
37 struct nfs_pgio_header *hdr);
38 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
39 struct nfs42_layoutstat_devinfo *devinfo,
40 int dev_limit);
41 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
42 const struct nfs42_layoutstat_devinfo *devinfo,
43 struct nfs4_ff_layout_mirror *mirror);
44
45 static struct pnfs_layout_hdr *
ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)46 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
47 {
48 struct nfs4_flexfile_layout *ffl;
49
50 ffl = kzalloc(sizeof(*ffl), gfp_flags);
51 if (ffl) {
52 pnfs_init_ds_commit_info(&ffl->commit_info);
53 INIT_LIST_HEAD(&ffl->error_list);
54 INIT_LIST_HEAD(&ffl->mirrors);
55 ffl->last_report_time = ktime_get();
56 ffl->commit_info.ops = &ff_layout_commit_ops;
57 return &ffl->generic_hdr;
58 } else
59 return NULL;
60 }
61
62 static void
ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)63 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
64 {
65 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
66 struct nfs4_ff_layout_ds_err *err, *n;
67
68 list_for_each_entry_safe(err, n, &ffl->error_list, list) {
69 list_del(&err->list);
70 kfree(err);
71 }
72 kfree_rcu(ffl, generic_hdr.plh_rcu);
73 }
74
decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)75 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
76 {
77 __be32 *p;
78
79 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
80 if (unlikely(p == NULL))
81 return -ENOBUFS;
82 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
83 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
84 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
85 p[0], p[1], p[2], p[3]);
86 return 0;
87 }
88
decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)89 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
90 {
91 __be32 *p;
92
93 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
94 if (unlikely(!p))
95 return -ENOBUFS;
96 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
97 nfs4_print_deviceid(devid);
98 return 0;
99 }
100
decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)101 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
102 {
103 __be32 *p;
104
105 p = xdr_inline_decode(xdr, 4);
106 if (unlikely(!p))
107 return -ENOBUFS;
108 fh->size = be32_to_cpup(p++);
109 if (fh->size > NFS_MAXFHSIZE) {
110 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
111 fh->size);
112 return -EOVERFLOW;
113 }
114 /* fh.data */
115 p = xdr_inline_decode(xdr, fh->size);
116 if (unlikely(!p))
117 return -ENOBUFS;
118 memcpy(&fh->data, p, fh->size);
119 dprintk("%s: fh len %d\n", __func__, fh->size);
120
121 return 0;
122 }
123
124 /*
125 * Currently only stringified uids and gids are accepted.
126 * I.e., kerberos is not supported to the DSes, so no pricipals.
127 *
128 * That means that one common function will suffice, but when
129 * principals are added, this should be split to accomodate
130 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
131 */
132 static int
decode_name(struct xdr_stream *xdr, u32 *id)133 decode_name(struct xdr_stream *xdr, u32 *id)
134 {
135 __be32 *p;
136 int len;
137
138 /* opaque_length(4)*/
139 p = xdr_inline_decode(xdr, 4);
140 if (unlikely(!p))
141 return -ENOBUFS;
142 len = be32_to_cpup(p++);
143 if (len < 0)
144 return -EINVAL;
145
146 dprintk("%s: len %u\n", __func__, len);
147
148 /* opaque body */
149 p = xdr_inline_decode(xdr, len);
150 if (unlikely(!p))
151 return -ENOBUFS;
152
153 if (!nfs_map_string_to_numeric((char *)p, len, id))
154 return -EINVAL;
155
156 return 0;
157 }
158
ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1, const struct nfs4_ff_layout_mirror *m2)159 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
160 const struct nfs4_ff_layout_mirror *m2)
161 {
162 int i, j;
163
164 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
165 return false;
166 for (i = 0; i < m1->fh_versions_cnt; i++) {
167 bool found_fh = false;
168 for (j = 0; j < m2->fh_versions_cnt; j++) {
169 if (nfs_compare_fh(&m1->fh_versions[i],
170 &m2->fh_versions[j]) == 0) {
171 found_fh = true;
172 break;
173 }
174 }
175 if (!found_fh)
176 return false;
177 }
178 return true;
179 }
180
181 static struct nfs4_ff_layout_mirror *
ff_layout_add_mirror(struct pnfs_layout_hdr *lo, struct nfs4_ff_layout_mirror *mirror)182 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
183 struct nfs4_ff_layout_mirror *mirror)
184 {
185 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
186 struct nfs4_ff_layout_mirror *pos;
187 struct inode *inode = lo->plh_inode;
188
189 spin_lock(&inode->i_lock);
190 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
191 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
192 continue;
193 if (!ff_mirror_match_fh(mirror, pos))
194 continue;
195 if (refcount_inc_not_zero(&pos->ref)) {
196 spin_unlock(&inode->i_lock);
197 return pos;
198 }
199 }
200 list_add(&mirror->mirrors, &ff_layout->mirrors);
201 mirror->layout = lo;
202 spin_unlock(&inode->i_lock);
203 return mirror;
204 }
205
206 static void
ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)207 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
208 {
209 struct inode *inode;
210 if (mirror->layout == NULL)
211 return;
212 inode = mirror->layout->plh_inode;
213 spin_lock(&inode->i_lock);
214 list_del(&mirror->mirrors);
215 spin_unlock(&inode->i_lock);
216 mirror->layout = NULL;
217 }
218
ff_layout_alloc_mirror(gfp_t gfp_flags)219 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
220 {
221 struct nfs4_ff_layout_mirror *mirror;
222
223 mirror = kzalloc(sizeof(*mirror), gfp_flags);
224 if (mirror != NULL) {
225 spin_lock_init(&mirror->lock);
226 refcount_set(&mirror->ref, 1);
227 INIT_LIST_HEAD(&mirror->mirrors);
228 }
229 return mirror;
230 }
231
ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)232 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
233 {
234 const struct cred *cred;
235
236 ff_layout_remove_mirror(mirror);
237 kfree(mirror->fh_versions);
238 cred = rcu_access_pointer(mirror->ro_cred);
239 put_cred(cred);
240 cred = rcu_access_pointer(mirror->rw_cred);
241 put_cred(cred);
242 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
243 kfree(mirror);
244 }
245
ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)246 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
247 {
248 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
249 ff_layout_free_mirror(mirror);
250 }
251
ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)252 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
253 {
254 u32 i;
255
256 for (i = 0; i < fls->mirror_array_cnt; i++)
257 ff_layout_put_mirror(fls->mirror_array[i]);
258 }
259
_ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)260 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
261 {
262 if (fls) {
263 ff_layout_free_mirror_array(fls);
264 kfree(fls);
265 }
266 }
267
268 static bool
ff_lseg_match_mirrors(struct pnfs_layout_segment *l1, struct pnfs_layout_segment *l2)269 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
270 struct pnfs_layout_segment *l2)
271 {
272 const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
273 const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
274 u32 i;
275
276 if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
277 return false;
278 for (i = 0; i < fl1->mirror_array_cnt; i++) {
279 if (fl1->mirror_array[i] != fl2->mirror_array[i])
280 return false;
281 }
282 return true;
283 }
284
285 static bool
ff_lseg_range_is_after(const struct pnfs_layout_range *l1, const struct pnfs_layout_range *l2)286 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
287 const struct pnfs_layout_range *l2)
288 {
289 u64 end1, end2;
290
291 if (l1->iomode != l2->iomode)
292 return l1->iomode != IOMODE_READ;
293 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
294 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
295 if (end1 < l2->offset)
296 return false;
297 if (end2 < l1->offset)
298 return true;
299 return l2->offset <= l1->offset;
300 }
301
302 static bool
ff_lseg_merge(struct pnfs_layout_segment *new, struct pnfs_layout_segment *old)303 ff_lseg_merge(struct pnfs_layout_segment *new,
304 struct pnfs_layout_segment *old)
305 {
306 u64 new_end, old_end;
307
308 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
309 return false;
310 if (new->pls_range.iomode != old->pls_range.iomode)
311 return false;
312 old_end = pnfs_calc_offset_end(old->pls_range.offset,
313 old->pls_range.length);
314 if (old_end < new->pls_range.offset)
315 return false;
316 new_end = pnfs_calc_offset_end(new->pls_range.offset,
317 new->pls_range.length);
318 if (new_end < old->pls_range.offset)
319 return false;
320 if (!ff_lseg_match_mirrors(new, old))
321 return false;
322
323 /* Mergeable: copy info from 'old' to 'new' */
324 if (new_end < old_end)
325 new_end = old_end;
326 if (new->pls_range.offset < old->pls_range.offset)
327 new->pls_range.offset = old->pls_range.offset;
328 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
329 new_end);
330 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
331 set_bit(NFS_LSEG_ROC, &new->pls_flags);
332 return true;
333 }
334
335 static void
ff_layout_add_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg, struct list_head *free_me)336 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
337 struct pnfs_layout_segment *lseg,
338 struct list_head *free_me)
339 {
340 pnfs_generic_layout_insert_lseg(lo, lseg,
341 ff_lseg_range_is_after,
342 ff_lseg_merge,
343 free_me);
344 }
345
ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)346 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
347 {
348 int i, j;
349
350 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
351 for (j = i + 1; j < fls->mirror_array_cnt; j++)
352 if (fls->mirror_array[i]->efficiency <
353 fls->mirror_array[j]->efficiency)
354 swap(fls->mirror_array[i],
355 fls->mirror_array[j]);
356 }
357 }
358
359 static struct pnfs_layout_segment *
ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh, struct nfs4_layoutget_res *lgr, gfp_t gfp_flags)360 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
361 struct nfs4_layoutget_res *lgr,
362 gfp_t gfp_flags)
363 {
364 struct pnfs_layout_segment *ret;
365 struct nfs4_ff_layout_segment *fls = NULL;
366 struct xdr_stream stream;
367 struct xdr_buf buf;
368 struct page *scratch;
369 u64 stripe_unit;
370 u32 mirror_array_cnt;
371 __be32 *p;
372 int i, rc;
373
374 dprintk("--> %s\n", __func__);
375 scratch = alloc_page(gfp_flags);
376 if (!scratch)
377 return ERR_PTR(-ENOMEM);
378
379 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
380 lgr->layoutp->len);
381 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
382
383 /* stripe unit and mirror_array_cnt */
384 rc = -EIO;
385 p = xdr_inline_decode(&stream, 8 + 4);
386 if (!p)
387 goto out_err_free;
388
389 p = xdr_decode_hyper(p, &stripe_unit);
390 mirror_array_cnt = be32_to_cpup(p++);
391 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
392 stripe_unit, mirror_array_cnt);
393
394 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
395 mirror_array_cnt == 0)
396 goto out_err_free;
397
398 rc = -ENOMEM;
399 fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
400 gfp_flags);
401 if (!fls)
402 goto out_err_free;
403
404 fls->mirror_array_cnt = mirror_array_cnt;
405 fls->stripe_unit = stripe_unit;
406
407 for (i = 0; i < fls->mirror_array_cnt; i++) {
408 struct nfs4_ff_layout_mirror *mirror;
409 struct cred *kcred;
410 const struct cred __rcu *cred;
411 kuid_t uid;
412 kgid_t gid;
413 u32 ds_count, fh_count, id;
414 int j;
415
416 rc = -EIO;
417 p = xdr_inline_decode(&stream, 4);
418 if (!p)
419 goto out_err_free;
420 ds_count = be32_to_cpup(p);
421
422 /* FIXME: allow for striping? */
423 if (ds_count != 1)
424 goto out_err_free;
425
426 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
427 if (fls->mirror_array[i] == NULL) {
428 rc = -ENOMEM;
429 goto out_err_free;
430 }
431
432 fls->mirror_array[i]->ds_count = ds_count;
433
434 /* deviceid */
435 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
436 if (rc)
437 goto out_err_free;
438
439 /* efficiency */
440 rc = -EIO;
441 p = xdr_inline_decode(&stream, 4);
442 if (!p)
443 goto out_err_free;
444 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
445
446 /* stateid */
447 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
448 if (rc)
449 goto out_err_free;
450
451 /* fh */
452 rc = -EIO;
453 p = xdr_inline_decode(&stream, 4);
454 if (!p)
455 goto out_err_free;
456 fh_count = be32_to_cpup(p);
457
458 fls->mirror_array[i]->fh_versions =
459 kcalloc(fh_count, sizeof(struct nfs_fh),
460 gfp_flags);
461 if (fls->mirror_array[i]->fh_versions == NULL) {
462 rc = -ENOMEM;
463 goto out_err_free;
464 }
465
466 for (j = 0; j < fh_count; j++) {
467 rc = decode_nfs_fh(&stream,
468 &fls->mirror_array[i]->fh_versions[j]);
469 if (rc)
470 goto out_err_free;
471 }
472
473 fls->mirror_array[i]->fh_versions_cnt = fh_count;
474
475 /* user */
476 rc = decode_name(&stream, &id);
477 if (rc)
478 goto out_err_free;
479
480 uid = make_kuid(&init_user_ns, id);
481
482 /* group */
483 rc = decode_name(&stream, &id);
484 if (rc)
485 goto out_err_free;
486
487 gid = make_kgid(&init_user_ns, id);
488
489 if (gfp_flags & __GFP_FS)
490 kcred = prepare_kernel_cred(NULL);
491 else {
492 unsigned int nofs_flags = memalloc_nofs_save();
493 kcred = prepare_kernel_cred(NULL);
494 memalloc_nofs_restore(nofs_flags);
495 }
496 rc = -ENOMEM;
497 if (!kcred)
498 goto out_err_free;
499 kcred->fsuid = uid;
500 kcred->fsgid = gid;
501 cred = RCU_INITIALIZER(kcred);
502
503 if (lgr->range.iomode == IOMODE_READ)
504 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
505 else
506 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
507
508 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
509 if (mirror != fls->mirror_array[i]) {
510 /* swap cred ptrs so free_mirror will clean up old */
511 if (lgr->range.iomode == IOMODE_READ) {
512 cred = xchg(&mirror->ro_cred, cred);
513 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
514 } else {
515 cred = xchg(&mirror->rw_cred, cred);
516 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
517 }
518 ff_layout_free_mirror(fls->mirror_array[i]);
519 fls->mirror_array[i] = mirror;
520 }
521
522 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
523 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
524 from_kuid(&init_user_ns, uid),
525 from_kgid(&init_user_ns, gid));
526 }
527
528 p = xdr_inline_decode(&stream, 4);
529 if (!p)
530 goto out_sort_mirrors;
531 fls->flags = be32_to_cpup(p);
532
533 p = xdr_inline_decode(&stream, 4);
534 if (!p)
535 goto out_sort_mirrors;
536 for (i=0; i < fls->mirror_array_cnt; i++)
537 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
538
539 out_sort_mirrors:
540 ff_layout_sort_mirrors(fls);
541 ret = &fls->generic_hdr;
542 dprintk("<-- %s (success)\n", __func__);
543 out_free_page:
544 __free_page(scratch);
545 return ret;
546 out_err_free:
547 _ff_layout_free_lseg(fls);
548 ret = ERR_PTR(rc);
549 dprintk("<-- %s (%d)\n", __func__, rc);
550 goto out_free_page;
551 }
552
553 static void
ff_layout_free_lseg(struct pnfs_layout_segment *lseg)554 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
555 {
556 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
557
558 dprintk("--> %s\n", __func__);
559
560 if (lseg->pls_range.iomode == IOMODE_RW) {
561 struct nfs4_flexfile_layout *ffl;
562 struct inode *inode;
563
564 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
565 inode = ffl->generic_hdr.plh_inode;
566 spin_lock(&inode->i_lock);
567 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
568 spin_unlock(&inode->i_lock);
569 }
570 _ff_layout_free_lseg(fls);
571 }
572
573 static void
nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)574 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
575 {
576 /* first IO request? */
577 if (atomic_inc_return(&timer->n_ops) == 1) {
578 timer->start_time = now;
579 }
580 }
581
582 static ktime_t
nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)583 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
584 {
585 ktime_t start;
586
587 if (atomic_dec_return(&timer->n_ops) < 0)
588 WARN_ON_ONCE(1);
589
590 start = timer->start_time;
591 timer->start_time = now;
592 return ktime_sub(now, start);
593 }
594
595 static bool
nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror, struct nfs4_ff_layoutstat *layoutstat, ktime_t now)596 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
597 struct nfs4_ff_layoutstat *layoutstat,
598 ktime_t now)
599 {
600 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
601 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
602
603 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
604 if (!mirror->start_time)
605 mirror->start_time = now;
606 if (mirror->report_interval != 0)
607 report_interval = (s64)mirror->report_interval * 1000LL;
608 else if (layoutstats_timer != 0)
609 report_interval = (s64)layoutstats_timer * 1000LL;
610 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
611 report_interval) {
612 ffl->last_report_time = now;
613 return true;
614 }
615
616 return false;
617 }
618
619 static void
nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat, __u64 requested)620 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
621 __u64 requested)
622 {
623 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
624
625 iostat->ops_requested++;
626 iostat->bytes_requested += requested;
627 }
628
629 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)630 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
631 __u64 requested,
632 __u64 completed,
633 ktime_t time_completed,
634 ktime_t time_started)
635 {
636 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
637 ktime_t completion_time = ktime_sub(time_completed, time_started);
638 ktime_t timer;
639
640 iostat->ops_completed++;
641 iostat->bytes_completed += completed;
642 iostat->bytes_not_delivered += requested - completed;
643
644 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
645 iostat->total_busy_time =
646 ktime_add(iostat->total_busy_time, timer);
647 iostat->aggregate_completion_time =
648 ktime_add(iostat->aggregate_completion_time,
649 completion_time);
650 }
651
652 static void
nfs4_ff_layout_stat_io_start_read(struct inode *inode, struct nfs4_ff_layout_mirror *mirror, __u64 requested, ktime_t now)653 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
654 struct nfs4_ff_layout_mirror *mirror,
655 __u64 requested, ktime_t now)
656 {
657 bool report;
658
659 spin_lock(&mirror->lock);
660 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
661 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
662 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
663 spin_unlock(&mirror->lock);
664
665 if (report)
666 pnfs_report_layoutstat(inode, GFP_KERNEL);
667 }
668
669 static void
nfs4_ff_layout_stat_io_end_read(struct rpc_task *task, struct nfs4_ff_layout_mirror *mirror, __u64 requested, __u64 completed)670 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
671 struct nfs4_ff_layout_mirror *mirror,
672 __u64 requested,
673 __u64 completed)
674 {
675 spin_lock(&mirror->lock);
676 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
677 requested, completed,
678 ktime_get(), task->tk_start);
679 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
680 spin_unlock(&mirror->lock);
681 }
682
683 static void
nfs4_ff_layout_stat_io_start_write(struct inode *inode, struct nfs4_ff_layout_mirror *mirror, __u64 requested, ktime_t now)684 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
685 struct nfs4_ff_layout_mirror *mirror,
686 __u64 requested, ktime_t now)
687 {
688 bool report;
689
690 spin_lock(&mirror->lock);
691 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
692 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
693 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
694 spin_unlock(&mirror->lock);
695
696 if (report)
697 pnfs_report_layoutstat(inode, GFP_NOIO);
698 }
699
700 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)701 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
702 struct nfs4_ff_layout_mirror *mirror,
703 __u64 requested,
704 __u64 completed,
705 enum nfs3_stable_how committed)
706 {
707 if (committed == NFS_UNSTABLE)
708 requested = completed = 0;
709
710 spin_lock(&mirror->lock);
711 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
712 requested, completed, ktime_get(), task->tk_start);
713 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
714 spin_unlock(&mirror->lock);
715 }
716
717 static void
ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)718 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
719 {
720 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
721
722 if (devid)
723 nfs4_mark_deviceid_unavailable(devid);
724 }
725
726 static void
ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)727 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
728 {
729 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
730
731 if (devid)
732 nfs4_mark_deviceid_available(devid);
733 }
734
735 static struct nfs4_pnfs_ds *
ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg, u32 start_idx, u32 *best_idx, bool check_device)736 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
737 u32 start_idx, u32 *best_idx,
738 bool check_device)
739 {
740 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
741 struct nfs4_ff_layout_mirror *mirror;
742 struct nfs4_pnfs_ds *ds;
743 bool fail_return = false;
744 u32 idx;
745
746 /* mirrors are initially sorted by efficiency */
747 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
748 if (idx+1 == fls->mirror_array_cnt)
749 fail_return = !check_device;
750
751 mirror = FF_LAYOUT_COMP(lseg, idx);
752 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
753 if (!ds)
754 continue;
755
756 if (check_device &&
757 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
758 continue;
759
760 *best_idx = idx;
761 return ds;
762 }
763
764 return NULL;
765 }
766
767 static struct nfs4_pnfs_ds *
ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg, u32 start_idx, u32 *best_idx)768 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
769 u32 start_idx, u32 *best_idx)
770 {
771 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
772 }
773
774 static struct nfs4_pnfs_ds *
ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg, u32 start_idx, u32 *best_idx)775 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
776 u32 start_idx, u32 *best_idx)
777 {
778 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
779 }
780
781 static struct nfs4_pnfs_ds *
ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, u32 start_idx, u32 *best_idx)782 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
783 u32 start_idx, u32 *best_idx)
784 {
785 struct nfs4_pnfs_ds *ds;
786
787 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
788 if (ds)
789 return ds;
790 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
791 }
792
793 static struct nfs4_pnfs_ds *
ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio, u32 *best_idx)794 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
795 u32 *best_idx)
796 {
797 struct pnfs_layout_segment *lseg = pgio->pg_lseg;
798 struct nfs4_pnfs_ds *ds;
799
800 ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
801 best_idx);
802 if (ds || !pgio->pg_mirror_idx)
803 return ds;
804 return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
805 }
806
807 static void
ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req, bool strict_iomode)808 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
809 struct nfs_page *req,
810 bool strict_iomode)
811 {
812 pnfs_put_lseg(pgio->pg_lseg);
813 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
814 nfs_req_openctx(req),
815 req_offset(req),
816 req->wb_bytes,
817 IOMODE_READ,
818 strict_iomode,
819 GFP_KERNEL);
820 if (IS_ERR(pgio->pg_lseg)) {
821 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
822 pgio->pg_lseg = NULL;
823 }
824 }
825
826 static void
ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)827 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
828 struct nfs_page *req)
829 {
830 pnfs_generic_pg_check_layout(pgio);
831 pnfs_generic_pg_check_range(pgio, req);
832 }
833
834 static void
ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)835 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
836 struct nfs_page *req)
837 {
838 struct nfs_pgio_mirror *pgm;
839 struct nfs4_ff_layout_mirror *mirror;
840 struct nfs4_pnfs_ds *ds;
841 u32 ds_idx;
842
843 retry:
844 ff_layout_pg_check_layout(pgio, req);
845 /* Use full layout for now */
846 if (!pgio->pg_lseg) {
847 ff_layout_pg_get_read(pgio, req, false);
848 if (!pgio->pg_lseg)
849 goto out_nolseg;
850 }
851 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
852 ff_layout_pg_get_read(pgio, req, true);
853 if (!pgio->pg_lseg)
854 goto out_nolseg;
855 }
856
857 ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
858 if (!ds) {
859 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
860 goto out_mds;
861 pnfs_generic_pg_cleanup(pgio);
862 /* Sleep for 1 second before retrying */
863 ssleep(1);
864 goto retry;
865 }
866
867 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
868 pgm = &pgio->pg_mirrors[0];
869 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
870
871 pgio->pg_mirror_idx = ds_idx;
872
873 if (NFS_SERVER(pgio->pg_inode)->flags &
874 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
875 pgio->pg_maxretrans = io_maxretrans;
876 return;
877 out_nolseg:
878 if (pgio->pg_error < 0)
879 return;
880 out_mds:
881 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
882 0, NFS4_MAX_UINT64, IOMODE_READ,
883 NFS_I(pgio->pg_inode)->layout,
884 pgio->pg_lseg);
885 pgio->pg_maxretrans = 0;
886 nfs_pageio_reset_read_mds(pgio);
887 }
888
889 static void
ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)890 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
891 struct nfs_page *req)
892 {
893 struct nfs4_ff_layout_mirror *mirror;
894 struct nfs_pgio_mirror *pgm;
895 struct nfs4_pnfs_ds *ds;
896 u32 i;
897
898 retry:
899 ff_layout_pg_check_layout(pgio, req);
900 if (!pgio->pg_lseg) {
901 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
902 nfs_req_openctx(req),
903 req_offset(req),
904 req->wb_bytes,
905 IOMODE_RW,
906 false,
907 GFP_NOFS);
908 if (IS_ERR(pgio->pg_lseg)) {
909 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
910 pgio->pg_lseg = NULL;
911 return;
912 }
913 }
914 /* If no lseg, fall back to write through mds */
915 if (pgio->pg_lseg == NULL)
916 goto out_mds;
917
918 /* Use a direct mapping of ds_idx to pgio mirror_idx */
919 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
920 goto out_eagain;
921
922 for (i = 0; i < pgio->pg_mirror_count; i++) {
923 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
924 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
925 if (!ds) {
926 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
927 goto out_mds;
928 pnfs_generic_pg_cleanup(pgio);
929 /* Sleep for 1 second before retrying */
930 ssleep(1);
931 goto retry;
932 }
933 pgm = &pgio->pg_mirrors[i];
934 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
935 }
936
937 if (NFS_SERVER(pgio->pg_inode)->flags &
938 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
939 pgio->pg_maxretrans = io_maxretrans;
940 return;
941 out_eagain:
942 pnfs_generic_pg_cleanup(pgio);
943 pgio->pg_error = -EAGAIN;
944 return;
945 out_mds:
946 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
947 0, NFS4_MAX_UINT64, IOMODE_RW,
948 NFS_I(pgio->pg_inode)->layout,
949 pgio->pg_lseg);
950 pgio->pg_maxretrans = 0;
951 nfs_pageio_reset_write_mds(pgio);
952 pgio->pg_error = -EAGAIN;
953 }
954
955 static unsigned int
ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)956 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
957 struct nfs_page *req)
958 {
959 if (!pgio->pg_lseg) {
960 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
961 nfs_req_openctx(req),
962 req_offset(req),
963 req->wb_bytes,
964 IOMODE_RW,
965 false,
966 GFP_NOFS);
967 if (IS_ERR(pgio->pg_lseg)) {
968 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
969 pgio->pg_lseg = NULL;
970 goto out;
971 }
972 }
973 if (pgio->pg_lseg)
974 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
975
976 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
977 0, NFS4_MAX_UINT64, IOMODE_RW,
978 NFS_I(pgio->pg_inode)->layout,
979 pgio->pg_lseg);
980 /* no lseg means that pnfs is not in use, so no mirroring here */
981 nfs_pageio_reset_write_mds(pgio);
982 out:
983 return 1;
984 }
985
986 static u32
ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)987 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
988 {
989 u32 old = desc->pg_mirror_idx;
990
991 desc->pg_mirror_idx = idx;
992 return old;
993 }
994
995 static struct nfs_pgio_mirror *
ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)996 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
997 {
998 return &desc->pg_mirrors[idx];
999 }
1000
1001 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1002 .pg_init = ff_layout_pg_init_read,
1003 .pg_test = pnfs_generic_pg_test,
1004 .pg_doio = pnfs_generic_pg_readpages,
1005 .pg_cleanup = pnfs_generic_pg_cleanup,
1006 };
1007
1008 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1009 .pg_init = ff_layout_pg_init_write,
1010 .pg_test = pnfs_generic_pg_test,
1011 .pg_doio = pnfs_generic_pg_writepages,
1012 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1013 .pg_cleanup = pnfs_generic_pg_cleanup,
1014 .pg_get_mirror = ff_layout_pg_get_mirror_write,
1015 .pg_set_mirror = ff_layout_pg_set_mirror_write,
1016 };
1017
ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)1018 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1019 {
1020 struct rpc_task *task = &hdr->task;
1021
1022 pnfs_layoutcommit_inode(hdr->inode, false);
1023
1024 if (retry_pnfs) {
1025 dprintk("%s Reset task %5u for i/o through pNFS "
1026 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1027 hdr->task.tk_pid,
1028 hdr->inode->i_sb->s_id,
1029 (unsigned long long)NFS_FILEID(hdr->inode),
1030 hdr->args.count,
1031 (unsigned long long)hdr->args.offset);
1032
1033 hdr->completion_ops->reschedule_io(hdr);
1034 return;
1035 }
1036
1037 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1038 dprintk("%s Reset task %5u for i/o through MDS "
1039 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1040 hdr->task.tk_pid,
1041 hdr->inode->i_sb->s_id,
1042 (unsigned long long)NFS_FILEID(hdr->inode),
1043 hdr->args.count,
1044 (unsigned long long)hdr->args.offset);
1045
1046 trace_pnfs_mds_fallback_write_done(hdr->inode,
1047 hdr->args.offset, hdr->args.count,
1048 IOMODE_RW, NFS_I(hdr->inode)->layout,
1049 hdr->lseg);
1050 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1051 }
1052 }
1053
ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)1054 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1055 {
1056 u32 idx = hdr->pgio_mirror_idx + 1;
1057 u32 new_idx = 0;
1058
1059 if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx))
1060 ff_layout_send_layouterror(hdr->lseg);
1061 else
1062 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1063 pnfs_read_resend_pnfs(hdr, new_idx);
1064 }
1065
ff_layout_reset_read(struct nfs_pgio_header *hdr)1066 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1067 {
1068 struct rpc_task *task = &hdr->task;
1069
1070 pnfs_layoutcommit_inode(hdr->inode, false);
1071 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1072
1073 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1074 dprintk("%s Reset task %5u for i/o through MDS "
1075 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1076 hdr->task.tk_pid,
1077 hdr->inode->i_sb->s_id,
1078 (unsigned long long)NFS_FILEID(hdr->inode),
1079 hdr->args.count,
1080 (unsigned long long)hdr->args.offset);
1081
1082 trace_pnfs_mds_fallback_read_done(hdr->inode,
1083 hdr->args.offset, hdr->args.count,
1084 IOMODE_READ, NFS_I(hdr->inode)->layout,
1085 hdr->lseg);
1086 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1087 }
1088 }
1089
ff_layout_async_handle_error_v4(struct rpc_task *task, struct nfs4_state *state, struct nfs_client *clp, struct pnfs_layout_segment *lseg, u32 idx)1090 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1091 struct nfs4_state *state,
1092 struct nfs_client *clp,
1093 struct pnfs_layout_segment *lseg,
1094 u32 idx)
1095 {
1096 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1097 struct inode *inode = lo->plh_inode;
1098 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1099 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1100
1101 switch (task->tk_status) {
1102 case -NFS4ERR_BADSESSION:
1103 case -NFS4ERR_BADSLOT:
1104 case -NFS4ERR_BAD_HIGH_SLOT:
1105 case -NFS4ERR_DEADSESSION:
1106 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1107 case -NFS4ERR_SEQ_FALSE_RETRY:
1108 case -NFS4ERR_SEQ_MISORDERED:
1109 dprintk("%s ERROR %d, Reset session. Exchangeid "
1110 "flags 0x%x\n", __func__, task->tk_status,
1111 clp->cl_exchange_flags);
1112 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1113 break;
1114 case -NFS4ERR_DELAY:
1115 case -NFS4ERR_GRACE:
1116 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1117 break;
1118 case -NFS4ERR_RETRY_UNCACHED_REP:
1119 break;
1120 /* Invalidate Layout errors */
1121 case -NFS4ERR_PNFS_NO_LAYOUT:
1122 case -ESTALE: /* mapped NFS4ERR_STALE */
1123 case -EBADHANDLE: /* mapped NFS4ERR_BADHANDLE */
1124 case -EISDIR: /* mapped NFS4ERR_ISDIR */
1125 case -NFS4ERR_FHEXPIRED:
1126 case -NFS4ERR_WRONG_TYPE:
1127 dprintk("%s Invalid layout error %d\n", __func__,
1128 task->tk_status);
1129 /*
1130 * Destroy layout so new i/o will get a new layout.
1131 * Layout will not be destroyed until all current lseg
1132 * references are put. Mark layout as invalid to resend failed
1133 * i/o and all i/o waiting on the slot table to the MDS until
1134 * layout is destroyed and a new valid layout is obtained.
1135 */
1136 pnfs_destroy_layout(NFS_I(inode));
1137 rpc_wake_up(&tbl->slot_tbl_waitq);
1138 goto reset;
1139 /* RPC connection errors */
1140 case -ECONNREFUSED:
1141 case -EHOSTDOWN:
1142 case -EHOSTUNREACH:
1143 case -ENETUNREACH:
1144 case -EIO:
1145 case -ETIMEDOUT:
1146 case -EPIPE:
1147 dprintk("%s DS connection error %d\n", __func__,
1148 task->tk_status);
1149 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1150 &devid->deviceid);
1151 rpc_wake_up(&tbl->slot_tbl_waitq);
1152 fallthrough;
1153 default:
1154 if (ff_layout_avoid_mds_available_ds(lseg))
1155 return -NFS4ERR_RESET_TO_PNFS;
1156 reset:
1157 dprintk("%s Retry through MDS. Error %d\n", __func__,
1158 task->tk_status);
1159 return -NFS4ERR_RESET_TO_MDS;
1160 }
1161 task->tk_status = 0;
1162 return -EAGAIN;
1163 }
1164
1165 /* 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, u32 idx)1166 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1167 struct pnfs_layout_segment *lseg,
1168 u32 idx)
1169 {
1170 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1171
1172 switch (task->tk_status) {
1173 /* File access problems. Don't mark the device as unavailable */
1174 case -EACCES:
1175 case -ESTALE:
1176 case -EISDIR:
1177 case -EBADHANDLE:
1178 case -ELOOP:
1179 case -ENOSPC:
1180 break;
1181 case -EJUKEBOX:
1182 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1183 goto out_retry;
1184 default:
1185 dprintk("%s DS connection error %d\n", __func__,
1186 task->tk_status);
1187 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1188 &devid->deviceid);
1189 }
1190 /* FIXME: Need to prevent infinite looping here. */
1191 return -NFS4ERR_RESET_TO_PNFS;
1192 out_retry:
1193 task->tk_status = 0;
1194 rpc_restart_call_prepare(task);
1195 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1196 return -EAGAIN;
1197 }
1198
ff_layout_async_handle_error(struct rpc_task *task, struct nfs4_state *state, struct nfs_client *clp, struct pnfs_layout_segment *lseg, u32 idx)1199 static int ff_layout_async_handle_error(struct rpc_task *task,
1200 struct nfs4_state *state,
1201 struct nfs_client *clp,
1202 struct pnfs_layout_segment *lseg,
1203 u32 idx)
1204 {
1205 int vers = clp->cl_nfs_mod->rpc_vers->number;
1206
1207 if (task->tk_status >= 0) {
1208 ff_layout_mark_ds_reachable(lseg, idx);
1209 return 0;
1210 }
1211
1212 /* Handle the case of an invalid layout segment */
1213 if (!pnfs_is_valid_lseg(lseg))
1214 return -NFS4ERR_RESET_TO_PNFS;
1215
1216 switch (vers) {
1217 case 3:
1218 return ff_layout_async_handle_error_v3(task, lseg, idx);
1219 case 4:
1220 return ff_layout_async_handle_error_v4(task, state, clp,
1221 lseg, idx);
1222 default:
1223 /* should never happen */
1224 WARN_ON_ONCE(1);
1225 return 0;
1226 }
1227 }
1228
ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, u32 idx, u64 offset, u64 length, u32 *op_status, int opnum, int error)1229 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1230 u32 idx, u64 offset, u64 length,
1231 u32 *op_status, int opnum, int error)
1232 {
1233 struct nfs4_ff_layout_mirror *mirror;
1234 u32 status = *op_status;
1235 int err;
1236
1237 if (status == 0) {
1238 switch (error) {
1239 case -ETIMEDOUT:
1240 case -EPFNOSUPPORT:
1241 case -EPROTONOSUPPORT:
1242 case -EOPNOTSUPP:
1243 case -EINVAL:
1244 case -ECONNREFUSED:
1245 case -ECONNRESET:
1246 case -EHOSTDOWN:
1247 case -EHOSTUNREACH:
1248 case -ENETUNREACH:
1249 case -EADDRINUSE:
1250 case -ENOBUFS:
1251 case -EPIPE:
1252 case -EPERM:
1253 *op_status = status = NFS4ERR_NXIO;
1254 break;
1255 case -EACCES:
1256 *op_status = status = NFS4ERR_ACCESS;
1257 break;
1258 default:
1259 return;
1260 }
1261 }
1262
1263 mirror = FF_LAYOUT_COMP(lseg, idx);
1264 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1265 mirror, offset, length, status, opnum,
1266 GFP_NOIO);
1267
1268 switch (status) {
1269 case NFS4ERR_DELAY:
1270 case NFS4ERR_GRACE:
1271 break;
1272 case NFS4ERR_NXIO:
1273 ff_layout_mark_ds_unreachable(lseg, idx);
1274 /*
1275 * Don't return the layout if this is a read and we still
1276 * have layouts to try
1277 */
1278 if (opnum == OP_READ)
1279 break;
1280 fallthrough;
1281 default:
1282 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1283 lseg);
1284 }
1285
1286 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1287 }
1288
1289 /* NFS_PROTO call done callback routines */
ff_layout_read_done_cb(struct rpc_task *task, struct nfs_pgio_header *hdr)1290 static int ff_layout_read_done_cb(struct rpc_task *task,
1291 struct nfs_pgio_header *hdr)
1292 {
1293 int err;
1294
1295 if (task->tk_status < 0) {
1296 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1297 hdr->args.offset, hdr->args.count,
1298 &hdr->res.op_status, OP_READ,
1299 task->tk_status);
1300 trace_ff_layout_read_error(hdr);
1301 }
1302
1303 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1304 hdr->ds_clp, hdr->lseg,
1305 hdr->pgio_mirror_idx);
1306
1307 trace_nfs4_pnfs_read(hdr, err);
1308 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1309 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1310 switch (err) {
1311 case -NFS4ERR_RESET_TO_PNFS:
1312 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1313 return task->tk_status;
1314 case -NFS4ERR_RESET_TO_MDS:
1315 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1316 return task->tk_status;
1317 case -EAGAIN:
1318 goto out_eagain;
1319 }
1320
1321 return 0;
1322 out_eagain:
1323 rpc_restart_call_prepare(task);
1324 return -EAGAIN;
1325 }
1326
1327 static bool
ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)1328 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1329 {
1330 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1331 }
1332
1333 /*
1334 * We reference the rpc_cred of the first WRITE that triggers the need for
1335 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1336 * rfc5661 is not clear about which credential should be used.
1337 *
1338 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1339 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1340 * we always send layoutcommit after DS writes.
1341 */
1342 static void
ff_layout_set_layoutcommit(struct inode *inode, struct pnfs_layout_segment *lseg, loff_t end_offset)1343 ff_layout_set_layoutcommit(struct inode *inode,
1344 struct pnfs_layout_segment *lseg,
1345 loff_t end_offset)
1346 {
1347 if (!ff_layout_need_layoutcommit(lseg))
1348 return;
1349
1350 pnfs_set_layoutcommit(inode, lseg, end_offset);
1351 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1352 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1353 }
1354
ff_layout_read_record_layoutstats_start(struct rpc_task *task, struct nfs_pgio_header *hdr)1355 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1356 struct nfs_pgio_header *hdr)
1357 {
1358 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1359 return;
1360 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1361 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1362 hdr->args.count,
1363 task->tk_start);
1364 }
1365
ff_layout_read_record_layoutstats_done(struct rpc_task *task, struct nfs_pgio_header *hdr)1366 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1367 struct nfs_pgio_header *hdr)
1368 {
1369 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1370 return;
1371 nfs4_ff_layout_stat_io_end_read(task,
1372 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1373 hdr->args.count,
1374 hdr->res.count);
1375 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1376 }
1377
ff_layout_read_prepare_common(struct rpc_task *task, struct nfs_pgio_header *hdr)1378 static int ff_layout_read_prepare_common(struct rpc_task *task,
1379 struct nfs_pgio_header *hdr)
1380 {
1381 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1382 rpc_exit(task, -EIO);
1383 return -EIO;
1384 }
1385
1386 ff_layout_read_record_layoutstats_start(task, hdr);
1387 return 0;
1388 }
1389
1390 /*
1391 * Call ops for the async read/write cases
1392 * In the case of dense layouts, the offset needs to be reset to its
1393 * original value.
1394 */
ff_layout_read_prepare_v3(struct rpc_task *task, void *data)1395 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1396 {
1397 struct nfs_pgio_header *hdr = data;
1398
1399 if (ff_layout_read_prepare_common(task, hdr))
1400 return;
1401
1402 rpc_call_start(task);
1403 }
1404
ff_layout_read_prepare_v4(struct rpc_task *task, void *data)1405 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1406 {
1407 struct nfs_pgio_header *hdr = data;
1408
1409 if (nfs4_setup_sequence(hdr->ds_clp,
1410 &hdr->args.seq_args,
1411 &hdr->res.seq_res,
1412 task))
1413 return;
1414
1415 ff_layout_read_prepare_common(task, hdr);
1416 }
1417
ff_layout_read_call_done(struct rpc_task *task, void *data)1418 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1419 {
1420 struct nfs_pgio_header *hdr = data;
1421
1422 dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1423
1424 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1425 task->tk_status == 0) {
1426 nfs4_sequence_done(task, &hdr->res.seq_res);
1427 return;
1428 }
1429
1430 /* Note this may cause RPC to be resent */
1431 hdr->mds_ops->rpc_call_done(task, hdr);
1432 }
1433
ff_layout_read_count_stats(struct rpc_task *task, void *data)1434 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1435 {
1436 struct nfs_pgio_header *hdr = data;
1437
1438 ff_layout_read_record_layoutstats_done(task, hdr);
1439 rpc_count_iostats_metrics(task,
1440 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1441 }
1442
ff_layout_read_release(void *data)1443 static void ff_layout_read_release(void *data)
1444 {
1445 struct nfs_pgio_header *hdr = data;
1446
1447 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1448 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1449 ff_layout_resend_pnfs_read(hdr);
1450 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1451 ff_layout_reset_read(hdr);
1452 pnfs_generic_rw_release(data);
1453 }
1454
1455
ff_layout_write_done_cb(struct rpc_task *task, struct nfs_pgio_header *hdr)1456 static int ff_layout_write_done_cb(struct rpc_task *task,
1457 struct nfs_pgio_header *hdr)
1458 {
1459 loff_t end_offs = 0;
1460 int err;
1461
1462 if (task->tk_status < 0) {
1463 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1464 hdr->args.offset, hdr->args.count,
1465 &hdr->res.op_status, OP_WRITE,
1466 task->tk_status);
1467 trace_ff_layout_write_error(hdr);
1468 }
1469
1470 err = ff_layout_async_handle_error(task, hdr->args.context->state,
1471 hdr->ds_clp, hdr->lseg,
1472 hdr->pgio_mirror_idx);
1473
1474 trace_nfs4_pnfs_write(hdr, err);
1475 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1476 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1477 switch (err) {
1478 case -NFS4ERR_RESET_TO_PNFS:
1479 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1480 return task->tk_status;
1481 case -NFS4ERR_RESET_TO_MDS:
1482 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1483 return task->tk_status;
1484 case -EAGAIN:
1485 return -EAGAIN;
1486 }
1487
1488 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1489 hdr->res.verf->committed == NFS_DATA_SYNC)
1490 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1491
1492 /* Note: if the write is unstable, don't set end_offs until commit */
1493 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1494
1495 /* zero out fattr since we don't care DS attr at all */
1496 hdr->fattr.valid = 0;
1497 if (task->tk_status >= 0)
1498 nfs_writeback_update_inode(hdr);
1499
1500 return 0;
1501 }
1502
ff_layout_commit_done_cb(struct rpc_task *task, struct nfs_commit_data *data)1503 static int ff_layout_commit_done_cb(struct rpc_task *task,
1504 struct nfs_commit_data *data)
1505 {
1506 int err;
1507
1508 if (task->tk_status < 0) {
1509 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1510 data->args.offset, data->args.count,
1511 &data->res.op_status, OP_COMMIT,
1512 task->tk_status);
1513 trace_ff_layout_commit_error(data);
1514 }
1515
1516 err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1517 data->lseg, data->ds_commit_index);
1518
1519 trace_nfs4_pnfs_commit_ds(data, err);
1520 switch (err) {
1521 case -NFS4ERR_RESET_TO_PNFS:
1522 pnfs_generic_prepare_to_resend_writes(data);
1523 return -EAGAIN;
1524 case -NFS4ERR_RESET_TO_MDS:
1525 pnfs_generic_prepare_to_resend_writes(data);
1526 return -EAGAIN;
1527 case -EAGAIN:
1528 rpc_restart_call_prepare(task);
1529 return -EAGAIN;
1530 }
1531
1532 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1533
1534 return 0;
1535 }
1536
ff_layout_write_record_layoutstats_start(struct rpc_task *task, struct nfs_pgio_header *hdr)1537 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1538 struct nfs_pgio_header *hdr)
1539 {
1540 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1541 return;
1542 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1543 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1544 hdr->args.count,
1545 task->tk_start);
1546 }
1547
ff_layout_write_record_layoutstats_done(struct rpc_task *task, struct nfs_pgio_header *hdr)1548 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1549 struct nfs_pgio_header *hdr)
1550 {
1551 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1552 return;
1553 nfs4_ff_layout_stat_io_end_write(task,
1554 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1555 hdr->args.count, hdr->res.count,
1556 hdr->res.verf->committed);
1557 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1558 }
1559
ff_layout_write_prepare_common(struct rpc_task *task, struct nfs_pgio_header *hdr)1560 static int ff_layout_write_prepare_common(struct rpc_task *task,
1561 struct nfs_pgio_header *hdr)
1562 {
1563 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1564 rpc_exit(task, -EIO);
1565 return -EIO;
1566 }
1567
1568 ff_layout_write_record_layoutstats_start(task, hdr);
1569 return 0;
1570 }
1571
ff_layout_write_prepare_v3(struct rpc_task *task, void *data)1572 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1573 {
1574 struct nfs_pgio_header *hdr = data;
1575
1576 if (ff_layout_write_prepare_common(task, hdr))
1577 return;
1578
1579 rpc_call_start(task);
1580 }
1581
ff_layout_write_prepare_v4(struct rpc_task *task, void *data)1582 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1583 {
1584 struct nfs_pgio_header *hdr = data;
1585
1586 if (nfs4_setup_sequence(hdr->ds_clp,
1587 &hdr->args.seq_args,
1588 &hdr->res.seq_res,
1589 task))
1590 return;
1591
1592 ff_layout_write_prepare_common(task, hdr);
1593 }
1594
ff_layout_write_call_done(struct rpc_task *task, void *data)1595 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1596 {
1597 struct nfs_pgio_header *hdr = data;
1598
1599 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1600 task->tk_status == 0) {
1601 nfs4_sequence_done(task, &hdr->res.seq_res);
1602 return;
1603 }
1604
1605 /* Note this may cause RPC to be resent */
1606 hdr->mds_ops->rpc_call_done(task, hdr);
1607 }
1608
ff_layout_write_count_stats(struct rpc_task *task, void *data)1609 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1610 {
1611 struct nfs_pgio_header *hdr = data;
1612
1613 ff_layout_write_record_layoutstats_done(task, hdr);
1614 rpc_count_iostats_metrics(task,
1615 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1616 }
1617
ff_layout_write_release(void *data)1618 static void ff_layout_write_release(void *data)
1619 {
1620 struct nfs_pgio_header *hdr = data;
1621
1622 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1623 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1624 ff_layout_send_layouterror(hdr->lseg);
1625 ff_layout_reset_write(hdr, true);
1626 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1627 ff_layout_reset_write(hdr, false);
1628 pnfs_generic_rw_release(data);
1629 }
1630
ff_layout_commit_record_layoutstats_start(struct rpc_task *task, struct nfs_commit_data *cdata)1631 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1632 struct nfs_commit_data *cdata)
1633 {
1634 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1635 return;
1636 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1637 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1638 0, task->tk_start);
1639 }
1640
ff_layout_commit_record_layoutstats_done(struct rpc_task *task, struct nfs_commit_data *cdata)1641 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1642 struct nfs_commit_data *cdata)
1643 {
1644 struct nfs_page *req;
1645 __u64 count = 0;
1646
1647 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1648 return;
1649
1650 if (task->tk_status == 0) {
1651 list_for_each_entry(req, &cdata->pages, wb_list)
1652 count += req->wb_bytes;
1653 }
1654 nfs4_ff_layout_stat_io_end_write(task,
1655 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1656 count, count, NFS_FILE_SYNC);
1657 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1658 }
1659
ff_layout_commit_prepare_common(struct rpc_task *task, struct nfs_commit_data *cdata)1660 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1661 struct nfs_commit_data *cdata)
1662 {
1663 ff_layout_commit_record_layoutstats_start(task, cdata);
1664 }
1665
ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)1666 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1667 {
1668 ff_layout_commit_prepare_common(task, data);
1669 rpc_call_start(task);
1670 }
1671
ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)1672 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1673 {
1674 struct nfs_commit_data *wdata = data;
1675
1676 if (nfs4_setup_sequence(wdata->ds_clp,
1677 &wdata->args.seq_args,
1678 &wdata->res.seq_res,
1679 task))
1680 return;
1681 ff_layout_commit_prepare_common(task, data);
1682 }
1683
ff_layout_commit_done(struct rpc_task *task, void *data)1684 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1685 {
1686 pnfs_generic_write_commit_done(task, data);
1687 }
1688
ff_layout_commit_count_stats(struct rpc_task *task, void *data)1689 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1690 {
1691 struct nfs_commit_data *cdata = data;
1692
1693 ff_layout_commit_record_layoutstats_done(task, cdata);
1694 rpc_count_iostats_metrics(task,
1695 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1696 }
1697
ff_layout_commit_release(void *data)1698 static void ff_layout_commit_release(void *data)
1699 {
1700 struct nfs_commit_data *cdata = data;
1701
1702 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1703 pnfs_generic_commit_release(data);
1704 }
1705
1706 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1707 .rpc_call_prepare = ff_layout_read_prepare_v3,
1708 .rpc_call_done = ff_layout_read_call_done,
1709 .rpc_count_stats = ff_layout_read_count_stats,
1710 .rpc_release = ff_layout_read_release,
1711 };
1712
1713 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1714 .rpc_call_prepare = ff_layout_read_prepare_v4,
1715 .rpc_call_done = ff_layout_read_call_done,
1716 .rpc_count_stats = ff_layout_read_count_stats,
1717 .rpc_release = ff_layout_read_release,
1718 };
1719
1720 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1721 .rpc_call_prepare = ff_layout_write_prepare_v3,
1722 .rpc_call_done = ff_layout_write_call_done,
1723 .rpc_count_stats = ff_layout_write_count_stats,
1724 .rpc_release = ff_layout_write_release,
1725 };
1726
1727 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1728 .rpc_call_prepare = ff_layout_write_prepare_v4,
1729 .rpc_call_done = ff_layout_write_call_done,
1730 .rpc_count_stats = ff_layout_write_count_stats,
1731 .rpc_release = ff_layout_write_release,
1732 };
1733
1734 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1735 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1736 .rpc_call_done = ff_layout_commit_done,
1737 .rpc_count_stats = ff_layout_commit_count_stats,
1738 .rpc_release = ff_layout_commit_release,
1739 };
1740
1741 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1742 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1743 .rpc_call_done = ff_layout_commit_done,
1744 .rpc_count_stats = ff_layout_commit_count_stats,
1745 .rpc_release = ff_layout_commit_release,
1746 };
1747
1748 static enum pnfs_try_status
ff_layout_read_pagelist(struct nfs_pgio_header *hdr)1749 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1750 {
1751 struct pnfs_layout_segment *lseg = hdr->lseg;
1752 struct nfs4_pnfs_ds *ds;
1753 struct rpc_clnt *ds_clnt;
1754 struct nfs4_ff_layout_mirror *mirror;
1755 const struct cred *ds_cred;
1756 loff_t offset = hdr->args.offset;
1757 u32 idx = hdr->pgio_mirror_idx;
1758 int vers;
1759 struct nfs_fh *fh;
1760
1761 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1762 __func__, hdr->inode->i_ino,
1763 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1764
1765 mirror = FF_LAYOUT_COMP(lseg, idx);
1766 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1767 if (!ds)
1768 goto out_failed;
1769
1770 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1771 hdr->inode);
1772 if (IS_ERR(ds_clnt))
1773 goto out_failed;
1774
1775 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1776 if (!ds_cred)
1777 goto out_failed;
1778
1779 vers = nfs4_ff_layout_ds_version(mirror);
1780
1781 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1782 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1783
1784 hdr->pgio_done_cb = ff_layout_read_done_cb;
1785 refcount_inc(&ds->ds_clp->cl_count);
1786 hdr->ds_clp = ds->ds_clp;
1787 fh = nfs4_ff_layout_select_ds_fh(mirror);
1788 if (fh)
1789 hdr->args.fh = fh;
1790
1791 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1792
1793 /*
1794 * Note that if we ever decide to split across DSes,
1795 * then we may need to handle dense-like offsets.
1796 */
1797 hdr->args.offset = offset;
1798 hdr->mds_offset = offset;
1799
1800 /* Perform an asynchronous read to ds */
1801 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1802 vers == 3 ? &ff_layout_read_call_ops_v3 :
1803 &ff_layout_read_call_ops_v4,
1804 0, RPC_TASK_SOFTCONN);
1805 put_cred(ds_cred);
1806 return PNFS_ATTEMPTED;
1807
1808 out_failed:
1809 if (ff_layout_avoid_mds_available_ds(lseg))
1810 return PNFS_TRY_AGAIN;
1811 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1812 hdr->args.offset, hdr->args.count,
1813 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1814 return PNFS_NOT_ATTEMPTED;
1815 }
1816
1817 /* Perform async writes. */
1818 static enum pnfs_try_status
ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)1819 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1820 {
1821 struct pnfs_layout_segment *lseg = hdr->lseg;
1822 struct nfs4_pnfs_ds *ds;
1823 struct rpc_clnt *ds_clnt;
1824 struct nfs4_ff_layout_mirror *mirror;
1825 const struct cred *ds_cred;
1826 loff_t offset = hdr->args.offset;
1827 int vers;
1828 struct nfs_fh *fh;
1829 u32 idx = hdr->pgio_mirror_idx;
1830
1831 mirror = FF_LAYOUT_COMP(lseg, idx);
1832 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1833 if (!ds)
1834 goto out_failed;
1835
1836 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1837 hdr->inode);
1838 if (IS_ERR(ds_clnt))
1839 goto out_failed;
1840
1841 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1842 if (!ds_cred)
1843 goto out_failed;
1844
1845 vers = nfs4_ff_layout_ds_version(mirror);
1846
1847 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1848 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1849 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1850 vers);
1851
1852 hdr->pgio_done_cb = ff_layout_write_done_cb;
1853 refcount_inc(&ds->ds_clp->cl_count);
1854 hdr->ds_clp = ds->ds_clp;
1855 hdr->ds_commit_idx = idx;
1856 fh = nfs4_ff_layout_select_ds_fh(mirror);
1857 if (fh)
1858 hdr->args.fh = fh;
1859
1860 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1861
1862 /*
1863 * Note that if we ever decide to split across DSes,
1864 * then we may need to handle dense-like offsets.
1865 */
1866 hdr->args.offset = offset;
1867
1868 /* Perform an asynchronous write */
1869 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1870 vers == 3 ? &ff_layout_write_call_ops_v3 :
1871 &ff_layout_write_call_ops_v4,
1872 sync, RPC_TASK_SOFTCONN);
1873 put_cred(ds_cred);
1874 return PNFS_ATTEMPTED;
1875
1876 out_failed:
1877 if (ff_layout_avoid_mds_available_ds(lseg))
1878 return PNFS_TRY_AGAIN;
1879 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1880 hdr->args.offset, hdr->args.count,
1881 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1882 return PNFS_NOT_ATTEMPTED;
1883 }
1884
calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)1885 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1886 {
1887 return i;
1888 }
1889
1890 static struct nfs_fh *
select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)1891 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1892 {
1893 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1894
1895 /* FIXME: Assume that there is only one NFS version available
1896 * for the DS.
1897 */
1898 return &flseg->mirror_array[i]->fh_versions[0];
1899 }
1900
ff_layout_initiate_commit(struct nfs_commit_data *data, int how)1901 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1902 {
1903 struct pnfs_layout_segment *lseg = data->lseg;
1904 struct nfs4_pnfs_ds *ds;
1905 struct rpc_clnt *ds_clnt;
1906 struct nfs4_ff_layout_mirror *mirror;
1907 const struct cred *ds_cred;
1908 u32 idx;
1909 int vers, ret;
1910 struct nfs_fh *fh;
1911
1912 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1913 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1914 goto out_err;
1915
1916 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1917 mirror = FF_LAYOUT_COMP(lseg, idx);
1918 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1919 if (!ds)
1920 goto out_err;
1921
1922 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1923 data->inode);
1924 if (IS_ERR(ds_clnt))
1925 goto out_err;
1926
1927 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1928 if (!ds_cred)
1929 goto out_err;
1930
1931 vers = nfs4_ff_layout_ds_version(mirror);
1932
1933 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1934 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1935 vers);
1936 data->commit_done_cb = ff_layout_commit_done_cb;
1937 data->cred = ds_cred;
1938 refcount_inc(&ds->ds_clp->cl_count);
1939 data->ds_clp = ds->ds_clp;
1940 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1941 if (fh)
1942 data->args.fh = fh;
1943
1944 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1945 vers == 3 ? &ff_layout_commit_call_ops_v3 :
1946 &ff_layout_commit_call_ops_v4,
1947 how, RPC_TASK_SOFTCONN);
1948 put_cred(ds_cred);
1949 return ret;
1950 out_err:
1951 pnfs_generic_prepare_to_resend_writes(data);
1952 pnfs_generic_commit_release(data);
1953 return -EAGAIN;
1954 }
1955
1956 static int
ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, int how, struct nfs_commit_info *cinfo)1957 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1958 int how, struct nfs_commit_info *cinfo)
1959 {
1960 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1961 ff_layout_initiate_commit);
1962 }
1963
1964 static struct pnfs_ds_commit_info *
ff_layout_get_ds_info(struct inode *inode)1965 ff_layout_get_ds_info(struct inode *inode)
1966 {
1967 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1968
1969 if (layout == NULL)
1970 return NULL;
1971
1972 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1973 }
1974
1975 static void
ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo, struct pnfs_layout_segment *lseg)1976 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1977 struct pnfs_layout_segment *lseg)
1978 {
1979 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1980 struct inode *inode = lseg->pls_layout->plh_inode;
1981 struct pnfs_commit_array *array, *new;
1982
1983 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1984 if (new) {
1985 spin_lock(&inode->i_lock);
1986 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1987 spin_unlock(&inode->i_lock);
1988 if (array != new)
1989 pnfs_free_commit_array(new);
1990 }
1991 }
1992
1993 static void
ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo, struct inode *inode)1994 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1995 struct inode *inode)
1996 {
1997 spin_lock(&inode->i_lock);
1998 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1999 spin_unlock(&inode->i_lock);
2000 }
2001
2002 static void
ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)2003 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2004 {
2005 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2006 id_node));
2007 }
2008
ff_layout_encode_ioerr(struct xdr_stream *xdr, const struct nfs4_layoutreturn_args *args, const struct nfs4_flexfile_layoutreturn_args *ff_args)2009 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2010 const struct nfs4_layoutreturn_args *args,
2011 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2012 {
2013 __be32 *start;
2014
2015 start = xdr_reserve_space(xdr, 4);
2016 if (unlikely(!start))
2017 return -E2BIG;
2018
2019 *start = cpu_to_be32(ff_args->num_errors);
2020 /* This assume we always return _ALL_ layouts */
2021 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2022 }
2023
2024 static void
encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)2025 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2026 {
2027 WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2028 }
2029
2030 static void
ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, const nfs4_stateid *stateid, const struct nfs42_layoutstat_devinfo *devinfo)2031 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2032 const nfs4_stateid *stateid,
2033 const struct nfs42_layoutstat_devinfo *devinfo)
2034 {
2035 __be32 *p;
2036
2037 p = xdr_reserve_space(xdr, 8 + 8);
2038 p = xdr_encode_hyper(p, devinfo->offset);
2039 p = xdr_encode_hyper(p, devinfo->length);
2040 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2041 p = xdr_reserve_space(xdr, 4*8);
2042 p = xdr_encode_hyper(p, devinfo->read_count);
2043 p = xdr_encode_hyper(p, devinfo->read_bytes);
2044 p = xdr_encode_hyper(p, devinfo->write_count);
2045 p = xdr_encode_hyper(p, devinfo->write_bytes);
2046 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2047 }
2048
2049 static void
ff_layout_encode_ff_iostat(struct xdr_stream *xdr, const nfs4_stateid *stateid, const struct nfs42_layoutstat_devinfo *devinfo)2050 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2051 const nfs4_stateid *stateid,
2052 const struct nfs42_layoutstat_devinfo *devinfo)
2053 {
2054 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2055 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2056 devinfo->ld_private.data);
2057 }
2058
2059 /* 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)2060 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2061 const struct nfs4_layoutreturn_args *args,
2062 struct nfs4_flexfile_layoutreturn_args *ff_args)
2063 {
2064 __be32 *p;
2065 int i;
2066
2067 p = xdr_reserve_space(xdr, 4);
2068 *p = cpu_to_be32(ff_args->num_dev);
2069 for (i = 0; i < ff_args->num_dev; i++)
2070 ff_layout_encode_ff_iostat(xdr,
2071 &args->layout->plh_stateid,
2072 &ff_args->devinfo[i]);
2073 }
2074
2075 static void
ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, unsigned int num_entries)2076 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2077 unsigned int num_entries)
2078 {
2079 unsigned int i;
2080
2081 for (i = 0; i < num_entries; i++) {
2082 if (!devinfo[i].ld_private.ops)
2083 continue;
2084 if (!devinfo[i].ld_private.ops->free)
2085 continue;
2086 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2087 }
2088 }
2089
2090 static struct nfs4_deviceid_node *
ff_layout_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev, gfp_t gfp_flags)2091 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2092 struct pnfs_device *pdev, gfp_t gfp_flags)
2093 {
2094 struct nfs4_ff_layout_ds *dsaddr;
2095
2096 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2097 if (!dsaddr)
2098 return NULL;
2099 return &dsaddr->id_node;
2100 }
2101
2102 static void
ff_layout_encode_layoutreturn(struct xdr_stream *xdr, const void *voidargs, const struct nfs4_xdr_opaque_data *ff_opaque)2103 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2104 const void *voidargs,
2105 const struct nfs4_xdr_opaque_data *ff_opaque)
2106 {
2107 const struct nfs4_layoutreturn_args *args = voidargs;
2108 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2109 struct xdr_buf tmp_buf = {
2110 .head = {
2111 [0] = {
2112 .iov_base = page_address(ff_args->pages[0]),
2113 },
2114 },
2115 .buflen = PAGE_SIZE,
2116 };
2117 struct xdr_stream tmp_xdr;
2118 __be32 *start;
2119
2120 dprintk("%s: Begin\n", __func__);
2121
2122 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2123
2124 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2125 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2126
2127 start = xdr_reserve_space(xdr, 4);
2128 *start = cpu_to_be32(tmp_buf.len);
2129 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2130
2131 dprintk("%s: Return\n", __func__);
2132 }
2133
2134 static void
ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)2135 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2136 {
2137 struct nfs4_flexfile_layoutreturn_args *ff_args;
2138
2139 if (!args->data)
2140 return;
2141 ff_args = args->data;
2142 args->data = NULL;
2143
2144 ff_layout_free_ds_ioerr(&ff_args->errors);
2145 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2146
2147 put_page(ff_args->pages[0]);
2148 kfree(ff_args);
2149 }
2150
2151 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2152 .encode = ff_layout_encode_layoutreturn,
2153 .free = ff_layout_free_layoutreturn,
2154 };
2155
2156 static int
ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)2157 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2158 {
2159 struct nfs4_flexfile_layoutreturn_args *ff_args;
2160 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2161
2162 ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2163 if (!ff_args)
2164 goto out_nomem;
2165 ff_args->pages[0] = alloc_page(GFP_KERNEL);
2166 if (!ff_args->pages[0])
2167 goto out_nomem_free;
2168
2169 INIT_LIST_HEAD(&ff_args->errors);
2170 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2171 &args->range, &ff_args->errors,
2172 FF_LAYOUTRETURN_MAXERR);
2173
2174 spin_lock(&args->inode->i_lock);
2175 ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2176 &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2177 spin_unlock(&args->inode->i_lock);
2178
2179 args->ld_private->ops = &layoutreturn_ops;
2180 args->ld_private->data = ff_args;
2181 return 0;
2182 out_nomem_free:
2183 kfree(ff_args);
2184 out_nomem:
2185 return -ENOMEM;
2186 }
2187
2188 #ifdef CONFIG_NFS_V4_2
2189 void
ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)2190 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2191 {
2192 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2193 struct nfs42_layout_error *errors;
2194 LIST_HEAD(head);
2195
2196 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2197 return;
2198 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2199 if (list_empty(&head))
2200 return;
2201
2202 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2203 sizeof(*errors), GFP_NOFS);
2204 if (errors != NULL) {
2205 const struct nfs4_ff_layout_ds_err *pos;
2206 size_t n = 0;
2207
2208 list_for_each_entry(pos, &head, list) {
2209 errors[n].offset = pos->offset;
2210 errors[n].length = pos->length;
2211 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2212 errors[n].errors[0].dev_id = pos->deviceid;
2213 errors[n].errors[0].status = pos->status;
2214 errors[n].errors[0].opnum = pos->opnum;
2215 n++;
2216 if (!list_is_last(&pos->list, &head) &&
2217 n < NFS42_LAYOUTERROR_MAX)
2218 continue;
2219 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2220 break;
2221 n = 0;
2222 }
2223 kfree(errors);
2224 }
2225 ff_layout_free_ds_ioerr(&head);
2226 }
2227 #else
2228 void
ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)2229 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2230 {
2231 }
2232 #endif
2233
2234 static int
ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)2235 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2236 {
2237 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2238
2239 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2240 }
2241
2242 static size_t
ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, const int buflen)2243 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2244 const int buflen)
2245 {
2246 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2247 const struct in6_addr *addr = &sin6->sin6_addr;
2248
2249 /*
2250 * RFC 4291, Section 2.2.2
2251 *
2252 * Shorthanded ANY address
2253 */
2254 if (ipv6_addr_any(addr))
2255 return snprintf(buf, buflen, "::");
2256
2257 /*
2258 * RFC 4291, Section 2.2.2
2259 *
2260 * Shorthanded loopback address
2261 */
2262 if (ipv6_addr_loopback(addr))
2263 return snprintf(buf, buflen, "::1");
2264
2265 /*
2266 * RFC 4291, Section 2.2.3
2267 *
2268 * Special presentation address format for mapped v4
2269 * addresses.
2270 */
2271 if (ipv6_addr_v4mapped(addr))
2272 return snprintf(buf, buflen, "::ffff:%pI4",
2273 &addr->s6_addr32[3]);
2274
2275 /*
2276 * RFC 4291, Section 2.2.1
2277 */
2278 return snprintf(buf, buflen, "%pI6c", addr);
2279 }
2280
2281 /* Derived from rpc_sockaddr2uaddr */
2282 static void
ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)2283 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2284 {
2285 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2286 char portbuf[RPCBIND_MAXUADDRPLEN];
2287 char addrbuf[RPCBIND_MAXUADDRLEN];
2288 char *netid;
2289 unsigned short port;
2290 int len, netid_len;
2291 __be32 *p;
2292
2293 switch (sap->sa_family) {
2294 case AF_INET:
2295 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2296 return;
2297 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2298 netid = "tcp";
2299 netid_len = 3;
2300 break;
2301 case AF_INET6:
2302 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2303 return;
2304 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2305 netid = "tcp6";
2306 netid_len = 4;
2307 break;
2308 default:
2309 /* we only support tcp and tcp6 */
2310 WARN_ON_ONCE(1);
2311 return;
2312 }
2313
2314 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2315 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2316
2317 p = xdr_reserve_space(xdr, 4 + netid_len);
2318 xdr_encode_opaque(p, netid, netid_len);
2319
2320 p = xdr_reserve_space(xdr, 4 + len);
2321 xdr_encode_opaque(p, addrbuf, len);
2322 }
2323
2324 static void
ff_layout_encode_nfstime(struct xdr_stream *xdr, ktime_t t)2325 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2326 ktime_t t)
2327 {
2328 struct timespec64 ts;
2329 __be32 *p;
2330
2331 p = xdr_reserve_space(xdr, 12);
2332 ts = ktime_to_timespec64(t);
2333 p = xdr_encode_hyper(p, ts.tv_sec);
2334 *p++ = cpu_to_be32(ts.tv_nsec);
2335 }
2336
2337 static void
ff_layout_encode_io_latency(struct xdr_stream *xdr, struct nfs4_ff_io_stat *stat)2338 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2339 struct nfs4_ff_io_stat *stat)
2340 {
2341 __be32 *p;
2342
2343 p = xdr_reserve_space(xdr, 5 * 8);
2344 p = xdr_encode_hyper(p, stat->ops_requested);
2345 p = xdr_encode_hyper(p, stat->bytes_requested);
2346 p = xdr_encode_hyper(p, stat->ops_completed);
2347 p = xdr_encode_hyper(p, stat->bytes_completed);
2348 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2349 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2350 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2351 }
2352
2353 static void
ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, const struct nfs42_layoutstat_devinfo *devinfo, struct nfs4_ff_layout_mirror *mirror)2354 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2355 const struct nfs42_layoutstat_devinfo *devinfo,
2356 struct nfs4_ff_layout_mirror *mirror)
2357 {
2358 struct nfs4_pnfs_ds_addr *da;
2359 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2360 struct nfs_fh *fh = &mirror->fh_versions[0];
2361 __be32 *p;
2362
2363 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2364 dprintk("%s: DS %s: encoding address %s\n",
2365 __func__, ds->ds_remotestr, da->da_remotestr);
2366 /* netaddr4 */
2367 ff_layout_encode_netaddr(xdr, da);
2368 /* nfs_fh4 */
2369 p = xdr_reserve_space(xdr, 4 + fh->size);
2370 xdr_encode_opaque(p, fh->data, fh->size);
2371 /* ff_io_latency4 read */
2372 spin_lock(&mirror->lock);
2373 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2374 /* ff_io_latency4 write */
2375 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2376 spin_unlock(&mirror->lock);
2377 /* nfstime4 */
2378 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2379 /* bool */
2380 p = xdr_reserve_space(xdr, 4);
2381 *p = cpu_to_be32(false);
2382 }
2383
2384 static void
ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, const struct nfs4_xdr_opaque_data *opaque)2385 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2386 const struct nfs4_xdr_opaque_data *opaque)
2387 {
2388 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2389 struct nfs42_layoutstat_devinfo, ld_private);
2390 __be32 *start;
2391
2392 /* layoutupdate length */
2393 start = xdr_reserve_space(xdr, 4);
2394 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2395
2396 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2397 }
2398
2399 static void
ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)2400 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2401 {
2402 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2403
2404 ff_layout_put_mirror(mirror);
2405 }
2406
2407 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2408 .encode = ff_layout_encode_layoutstats,
2409 .free = ff_layout_free_layoutstats,
2410 };
2411
2412 static int
ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, struct nfs42_layoutstat_devinfo *devinfo, int dev_limit)2413 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2414 struct nfs42_layoutstat_devinfo *devinfo,
2415 int dev_limit)
2416 {
2417 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2418 struct nfs4_ff_layout_mirror *mirror;
2419 struct nfs4_deviceid_node *dev;
2420 int i = 0;
2421
2422 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2423 if (i >= dev_limit)
2424 break;
2425 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2426 continue;
2427 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2428 continue;
2429 /* mirror refcount put in cleanup_layoutstats */
2430 if (!refcount_inc_not_zero(&mirror->ref))
2431 continue;
2432 dev = &mirror->mirror_ds->id_node;
2433 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2434 devinfo->offset = 0;
2435 devinfo->length = NFS4_MAX_UINT64;
2436 spin_lock(&mirror->lock);
2437 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2438 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2439 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2440 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2441 spin_unlock(&mirror->lock);
2442 devinfo->layout_type = LAYOUT_FLEX_FILES;
2443 devinfo->ld_private.ops = &layoutstat_ops;
2444 devinfo->ld_private.data = mirror;
2445
2446 devinfo++;
2447 i++;
2448 }
2449 return i;
2450 }
2451
2452 static int
ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)2453 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2454 {
2455 struct nfs4_flexfile_layout *ff_layout;
2456 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2457
2458 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2459 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2460 if (!args->devinfo)
2461 return -ENOMEM;
2462
2463 spin_lock(&args->inode->i_lock);
2464 ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2465 args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2466 &args->devinfo[0], dev_count);
2467 spin_unlock(&args->inode->i_lock);
2468 if (!args->num_dev) {
2469 kfree(args->devinfo);
2470 args->devinfo = NULL;
2471 return -ENOENT;
2472 }
2473
2474 return 0;
2475 }
2476
2477 static int
ff_layout_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *dummy)2478 ff_layout_set_layoutdriver(struct nfs_server *server,
2479 const struct nfs_fh *dummy)
2480 {
2481 #if IS_ENABLED(CONFIG_NFS_V4_2)
2482 server->caps |= NFS_CAP_LAYOUTSTATS;
2483 #endif
2484 return 0;
2485 }
2486
2487 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2488 .setup_ds_info = ff_layout_setup_ds_info,
2489 .release_ds_info = ff_layout_release_ds_info,
2490 .mark_request_commit = pnfs_layout_mark_request_commit,
2491 .clear_request_commit = pnfs_generic_clear_request_commit,
2492 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2493 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2494 .commit_pagelist = ff_layout_commit_pagelist,
2495 };
2496
2497 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2498 .id = LAYOUT_FLEX_FILES,
2499 .name = "LAYOUT_FLEX_FILES",
2500 .owner = THIS_MODULE,
2501 .flags = PNFS_LAYOUTGET_ON_OPEN,
2502 .max_layoutget_response = 4096, /* 1 page or so... */
2503 .set_layoutdriver = ff_layout_set_layoutdriver,
2504 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2505 .free_layout_hdr = ff_layout_free_layout_hdr,
2506 .alloc_lseg = ff_layout_alloc_lseg,
2507 .free_lseg = ff_layout_free_lseg,
2508 .add_lseg = ff_layout_add_lseg,
2509 .pg_read_ops = &ff_layout_pg_read_ops,
2510 .pg_write_ops = &ff_layout_pg_write_ops,
2511 .get_ds_info = ff_layout_get_ds_info,
2512 .free_deviceid_node = ff_layout_free_deviceid_node,
2513 .read_pagelist = ff_layout_read_pagelist,
2514 .write_pagelist = ff_layout_write_pagelist,
2515 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2516 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2517 .sync = pnfs_nfs_generic_sync,
2518 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2519 };
2520
nfs4flexfilelayout_init(void)2521 static int __init nfs4flexfilelayout_init(void)
2522 {
2523 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2524 __func__);
2525 return pnfs_register_layoutdriver(&flexfilelayout_type);
2526 }
2527
nfs4flexfilelayout_exit(void)2528 static void __exit nfs4flexfilelayout_exit(void)
2529 {
2530 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2531 __func__);
2532 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2533 }
2534
2535 MODULE_ALIAS("nfs-layouttype4-4");
2536
2537 MODULE_LICENSE("GPL");
2538 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2539
2540 module_init(nfs4flexfilelayout_init);
2541 module_exit(nfs4flexfilelayout_exit);
2542
2543 module_param(io_maxretrans, ushort, 0644);
2544 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2545 "retries an I/O request before returning an error. ");
2546