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/file.h>
15 #include <linux/sched/mm.h>
16
17 #include <linux/sunrpc/metrics.h>
18
19 #include "flexfilelayout.h"
20 #include "../nfs4session.h"
21 #include "../nfs4idmap.h"
22 #include "../internal.h"
23 #include "../delegation.h"
24 #include "../nfs4trace.h"
25 #include "../iostat.h"
26 #include "../nfs.h"
27 #include "../nfs42.h"
28
29 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
30
31 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
32 #define FF_LAYOUTRETURN_MAXERR 20
33
34 enum nfs4_ff_op_type {
35 NFS4_FF_OP_LAYOUTSTATS,
36 NFS4_FF_OP_LAYOUTRETURN,
37 };
38
39 static unsigned short io_maxretrans;
40
41 static const struct pnfs_commit_ops ff_layout_commit_ops;
42 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
43 struct nfs_pgio_header *hdr);
44 static int
45 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
46 struct nfs42_layoutstat_devinfo *devinfo,
47 int dev_limit, enum nfs4_ff_op_type type);
48 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
49 const struct nfs42_layoutstat_devinfo *devinfo,
50 struct nfs4_ff_layout_mirror *mirror);
51
52 static struct pnfs_layout_hdr *
ff_layout_alloc_layout_hdr(struct inode * inode,gfp_t gfp_flags)53 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
54 {
55 struct nfs4_flexfile_layout *ffl;
56
57 ffl = kzalloc(sizeof(*ffl), gfp_flags);
58 if (ffl) {
59 pnfs_init_ds_commit_info(&ffl->commit_info);
60 INIT_LIST_HEAD(&ffl->error_list);
61 INIT_LIST_HEAD(&ffl->mirrors);
62 ffl->last_report_time = ktime_get();
63 ffl->commit_info.ops = &ff_layout_commit_ops;
64 return &ffl->generic_hdr;
65 } else
66 return NULL;
67 }
68
69 static void
ff_layout_free_layout_hdr(struct pnfs_layout_hdr * lo)70 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
71 {
72 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
73 struct nfs4_ff_layout_ds_err *err, *n;
74
75 list_for_each_entry_safe(err, n, &ffl->error_list, list) {
76 list_del(&err->list);
77 kfree(err);
78 }
79 kfree_rcu(ffl, generic_hdr.plh_rcu);
80 }
81
decode_pnfs_stateid(struct xdr_stream * xdr,nfs4_stateid * stateid)82 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
83 {
84 __be32 *p;
85
86 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
87 if (unlikely(p == NULL))
88 return -ENOBUFS;
89 stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
90 memcpy(stateid->data, p, NFS4_STATEID_SIZE);
91 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
92 p[0], p[1], p[2], p[3]);
93 return 0;
94 }
95
decode_deviceid(struct xdr_stream * xdr,struct nfs4_deviceid * devid)96 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
97 {
98 __be32 *p;
99
100 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
101 if (unlikely(!p))
102 return -ENOBUFS;
103 memcpy(devid, p, NFS4_DEVICEID4_SIZE);
104 nfs4_print_deviceid(devid);
105 return 0;
106 }
107
decode_nfs_fh(struct xdr_stream * xdr,struct nfs_fh * fh)108 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
109 {
110 __be32 *p;
111
112 p = xdr_inline_decode(xdr, 4);
113 if (unlikely(!p))
114 return -ENOBUFS;
115 fh->size = be32_to_cpup(p++);
116 if (fh->size > NFS_MAXFHSIZE) {
117 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
118 fh->size);
119 return -EOVERFLOW;
120 }
121 /* fh.data */
122 p = xdr_inline_decode(xdr, fh->size);
123 if (unlikely(!p))
124 return -ENOBUFS;
125 memcpy(&fh->data, p, fh->size);
126 dprintk("%s: fh len %d\n", __func__, fh->size);
127
128 return 0;
129 }
130
131 /*
132 * Currently only stringified uids and gids are accepted.
133 * I.e., kerberos is not supported to the DSes, so no pricipals.
134 *
135 * That means that one common function will suffice, but when
136 * principals are added, this should be split to accomodate
137 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
138 */
139 static int
decode_name(struct xdr_stream * xdr,u32 * id)140 decode_name(struct xdr_stream *xdr, u32 *id)
141 {
142 __be32 *p;
143 int len;
144
145 /* opaque_length(4)*/
146 p = xdr_inline_decode(xdr, 4);
147 if (unlikely(!p))
148 return -ENOBUFS;
149 len = be32_to_cpup(p++);
150 if (len < 0)
151 return -EINVAL;
152
153 dprintk("%s: len %u\n", __func__, len);
154
155 /* opaque body */
156 p = xdr_inline_decode(xdr, len);
157 if (unlikely(!p))
158 return -ENOBUFS;
159
160 if (!nfs_map_string_to_numeric((char *)p, len, id))
161 return -EINVAL;
162
163 return 0;
164 }
165
166 static struct nfsd_file *
ff_local_open_fh(struct nfs_client * clp,const struct cred * cred,struct nfs_fh * fh,fmode_t mode)167 ff_local_open_fh(struct nfs_client *clp, const struct cred *cred,
168 struct nfs_fh *fh, fmode_t mode)
169 {
170 if (mode & FMODE_WRITE) {
171 /*
172 * Always request read and write access since this corresponds
173 * to a rw layout.
174 */
175 mode |= FMODE_READ;
176 }
177
178 return nfs_local_open_fh(clp, cred, fh, mode);
179 }
180
ff_mirror_match_fh(const struct nfs4_ff_layout_mirror * m1,const struct nfs4_ff_layout_mirror * m2)181 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
182 const struct nfs4_ff_layout_mirror *m2)
183 {
184 int i, j;
185
186 if (m1->fh_versions_cnt != m2->fh_versions_cnt)
187 return false;
188 for (i = 0; i < m1->fh_versions_cnt; i++) {
189 bool found_fh = false;
190 for (j = 0; j < m2->fh_versions_cnt; j++) {
191 if (nfs_compare_fh(&m1->fh_versions[i],
192 &m2->fh_versions[j]) == 0) {
193 found_fh = true;
194 break;
195 }
196 }
197 if (!found_fh)
198 return false;
199 }
200 return true;
201 }
202
203 static struct nfs4_ff_layout_mirror *
ff_layout_add_mirror(struct pnfs_layout_hdr * lo,struct nfs4_ff_layout_mirror * mirror)204 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
205 struct nfs4_ff_layout_mirror *mirror)
206 {
207 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
208 struct nfs4_ff_layout_mirror *pos;
209 struct inode *inode = lo->plh_inode;
210
211 spin_lock(&inode->i_lock);
212 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
213 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
214 continue;
215 if (!ff_mirror_match_fh(mirror, pos))
216 continue;
217 if (refcount_inc_not_zero(&pos->ref)) {
218 spin_unlock(&inode->i_lock);
219 return pos;
220 }
221 }
222 list_add(&mirror->mirrors, &ff_layout->mirrors);
223 mirror->layout = lo;
224 spin_unlock(&inode->i_lock);
225 return mirror;
226 }
227
228 static void
ff_layout_remove_mirror(struct nfs4_ff_layout_mirror * mirror)229 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
230 {
231 struct inode *inode;
232 if (mirror->layout == NULL)
233 return;
234 inode = mirror->layout->plh_inode;
235 spin_lock(&inode->i_lock);
236 list_del(&mirror->mirrors);
237 spin_unlock(&inode->i_lock);
238 mirror->layout = NULL;
239 }
240
ff_layout_alloc_mirror(gfp_t gfp_flags)241 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
242 {
243 struct nfs4_ff_layout_mirror *mirror;
244
245 mirror = kzalloc(sizeof(*mirror), gfp_flags);
246 if (mirror != NULL) {
247 spin_lock_init(&mirror->lock);
248 refcount_set(&mirror->ref, 1);
249 INIT_LIST_HEAD(&mirror->mirrors);
250 }
251 return mirror;
252 }
253
ff_layout_free_mirror(struct nfs4_ff_layout_mirror * mirror)254 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
255 {
256 const struct cred *cred;
257
258 ff_layout_remove_mirror(mirror);
259 kfree(mirror->fh_versions);
260 cred = rcu_access_pointer(mirror->ro_cred);
261 put_cred(cred);
262 cred = rcu_access_pointer(mirror->rw_cred);
263 put_cred(cred);
264 nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
265 kfree(mirror);
266 }
267
ff_layout_put_mirror(struct nfs4_ff_layout_mirror * mirror)268 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
269 {
270 if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
271 ff_layout_free_mirror(mirror);
272 }
273
ff_layout_free_mirror_array(struct nfs4_ff_layout_segment * fls)274 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
275 {
276 u32 i;
277
278 for (i = 0; i < fls->mirror_array_cnt; i++)
279 ff_layout_put_mirror(fls->mirror_array[i]);
280 }
281
_ff_layout_free_lseg(struct nfs4_ff_layout_segment * fls)282 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
283 {
284 if (fls) {
285 ff_layout_free_mirror_array(fls);
286 kfree(fls);
287 }
288 }
289
290 static bool
ff_lseg_match_mirrors(struct pnfs_layout_segment * l1,struct pnfs_layout_segment * l2)291 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
292 struct pnfs_layout_segment *l2)
293 {
294 const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
295 const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l2);
296 u32 i;
297
298 if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
299 return false;
300 for (i = 0; i < fl1->mirror_array_cnt; i++) {
301 if (fl1->mirror_array[i] != fl2->mirror_array[i])
302 return false;
303 }
304 return true;
305 }
306
307 static bool
ff_lseg_range_is_after(const struct pnfs_layout_range * l1,const struct pnfs_layout_range * l2)308 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
309 const struct pnfs_layout_range *l2)
310 {
311 u64 end1, end2;
312
313 if (l1->iomode != l2->iomode)
314 return l1->iomode != IOMODE_READ;
315 end1 = pnfs_calc_offset_end(l1->offset, l1->length);
316 end2 = pnfs_calc_offset_end(l2->offset, l2->length);
317 if (end1 < l2->offset)
318 return false;
319 if (end2 < l1->offset)
320 return true;
321 return l2->offset <= l1->offset;
322 }
323
324 static bool
ff_lseg_merge(struct pnfs_layout_segment * new,struct pnfs_layout_segment * old)325 ff_lseg_merge(struct pnfs_layout_segment *new,
326 struct pnfs_layout_segment *old)
327 {
328 u64 new_end, old_end;
329
330 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
331 return false;
332 if (new->pls_range.iomode != old->pls_range.iomode)
333 return false;
334 old_end = pnfs_calc_offset_end(old->pls_range.offset,
335 old->pls_range.length);
336 if (old_end < new->pls_range.offset)
337 return false;
338 new_end = pnfs_calc_offset_end(new->pls_range.offset,
339 new->pls_range.length);
340 if (new_end < old->pls_range.offset)
341 return false;
342 if (!ff_lseg_match_mirrors(new, old))
343 return false;
344
345 /* Mergeable: copy info from 'old' to 'new' */
346 if (new_end < old_end)
347 new_end = old_end;
348 if (new->pls_range.offset < old->pls_range.offset)
349 new->pls_range.offset = old->pls_range.offset;
350 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
351 new_end);
352 if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
353 set_bit(NFS_LSEG_ROC, &new->pls_flags);
354 return true;
355 }
356
357 static void
ff_layout_add_lseg(struct pnfs_layout_hdr * lo,struct pnfs_layout_segment * lseg,struct list_head * free_me)358 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
359 struct pnfs_layout_segment *lseg,
360 struct list_head *free_me)
361 {
362 pnfs_generic_layout_insert_lseg(lo, lseg,
363 ff_lseg_range_is_after,
364 ff_lseg_merge,
365 free_me);
366 }
367
ff_layout_sort_mirrors(struct nfs4_ff_layout_segment * fls)368 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
369 {
370 int i, j;
371
372 for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
373 for (j = i + 1; j < fls->mirror_array_cnt; j++)
374 if (fls->mirror_array[i]->efficiency <
375 fls->mirror_array[j]->efficiency)
376 swap(fls->mirror_array[i],
377 fls->mirror_array[j]);
378 }
379 }
380
381 static struct pnfs_layout_segment *
ff_layout_alloc_lseg(struct pnfs_layout_hdr * lh,struct nfs4_layoutget_res * lgr,gfp_t gfp_flags)382 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
383 struct nfs4_layoutget_res *lgr,
384 gfp_t gfp_flags)
385 {
386 struct pnfs_layout_segment *ret;
387 struct nfs4_ff_layout_segment *fls = NULL;
388 struct xdr_stream stream;
389 struct xdr_buf buf;
390 struct page *scratch;
391 u64 stripe_unit;
392 u32 mirror_array_cnt;
393 __be32 *p;
394 int i, rc;
395
396 dprintk("--> %s\n", __func__);
397 scratch = alloc_page(gfp_flags);
398 if (!scratch)
399 return ERR_PTR(-ENOMEM);
400
401 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
402 lgr->layoutp->len);
403 xdr_set_scratch_page(&stream, scratch);
404
405 /* stripe unit and mirror_array_cnt */
406 rc = -EIO;
407 p = xdr_inline_decode(&stream, 8 + 4);
408 if (!p)
409 goto out_err_free;
410
411 p = xdr_decode_hyper(p, &stripe_unit);
412 mirror_array_cnt = be32_to_cpup(p++);
413 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
414 stripe_unit, mirror_array_cnt);
415
416 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
417 mirror_array_cnt == 0)
418 goto out_err_free;
419
420 rc = -ENOMEM;
421 fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
422 gfp_flags);
423 if (!fls)
424 goto out_err_free;
425
426 fls->mirror_array_cnt = mirror_array_cnt;
427 fls->stripe_unit = stripe_unit;
428
429 for (i = 0; i < fls->mirror_array_cnt; i++) {
430 struct nfs4_ff_layout_mirror *mirror;
431 struct cred *kcred;
432 const struct cred __rcu *cred;
433 kuid_t uid;
434 kgid_t gid;
435 u32 ds_count, fh_count, id;
436 int j;
437
438 rc = -EIO;
439 p = xdr_inline_decode(&stream, 4);
440 if (!p)
441 goto out_err_free;
442 ds_count = be32_to_cpup(p);
443
444 /* FIXME: allow for striping? */
445 if (ds_count != 1)
446 goto out_err_free;
447
448 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
449 if (fls->mirror_array[i] == NULL) {
450 rc = -ENOMEM;
451 goto out_err_free;
452 }
453
454 fls->mirror_array[i]->ds_count = ds_count;
455
456 /* deviceid */
457 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
458 if (rc)
459 goto out_err_free;
460
461 /* efficiency */
462 rc = -EIO;
463 p = xdr_inline_decode(&stream, 4);
464 if (!p)
465 goto out_err_free;
466 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
467
468 /* stateid */
469 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
470 if (rc)
471 goto out_err_free;
472
473 /* fh */
474 rc = -EIO;
475 p = xdr_inline_decode(&stream, 4);
476 if (!p)
477 goto out_err_free;
478 fh_count = be32_to_cpup(p);
479
480 fls->mirror_array[i]->fh_versions =
481 kcalloc(fh_count, sizeof(struct nfs_fh),
482 gfp_flags);
483 if (fls->mirror_array[i]->fh_versions == NULL) {
484 rc = -ENOMEM;
485 goto out_err_free;
486 }
487
488 for (j = 0; j < fh_count; j++) {
489 rc = decode_nfs_fh(&stream,
490 &fls->mirror_array[i]->fh_versions[j]);
491 if (rc)
492 goto out_err_free;
493 }
494
495 fls->mirror_array[i]->fh_versions_cnt = fh_count;
496
497 /* user */
498 rc = decode_name(&stream, &id);
499 if (rc)
500 goto out_err_free;
501
502 uid = make_kuid(&init_user_ns, id);
503
504 /* group */
505 rc = decode_name(&stream, &id);
506 if (rc)
507 goto out_err_free;
508
509 gid = make_kgid(&init_user_ns, id);
510
511 if (gfp_flags & __GFP_FS)
512 kcred = prepare_kernel_cred(&init_task);
513 else {
514 unsigned int nofs_flags = memalloc_nofs_save();
515 kcred = prepare_kernel_cred(&init_task);
516 memalloc_nofs_restore(nofs_flags);
517 }
518 rc = -ENOMEM;
519 if (!kcred)
520 goto out_err_free;
521 kcred->fsuid = uid;
522 kcred->fsgid = gid;
523 cred = RCU_INITIALIZER(kcred);
524
525 if (lgr->range.iomode == IOMODE_READ)
526 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
527 else
528 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
529
530 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
531 if (mirror != fls->mirror_array[i]) {
532 /* swap cred ptrs so free_mirror will clean up old */
533 if (lgr->range.iomode == IOMODE_READ) {
534 cred = xchg(&mirror->ro_cred, cred);
535 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
536 } else {
537 cred = xchg(&mirror->rw_cred, cred);
538 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
539 }
540 ff_layout_free_mirror(fls->mirror_array[i]);
541 fls->mirror_array[i] = mirror;
542 }
543
544 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
545 lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
546 from_kuid(&init_user_ns, uid),
547 from_kgid(&init_user_ns, gid));
548 }
549
550 p = xdr_inline_decode(&stream, 4);
551 if (!p)
552 goto out_sort_mirrors;
553 fls->flags = be32_to_cpup(p);
554
555 p = xdr_inline_decode(&stream, 4);
556 if (!p)
557 goto out_sort_mirrors;
558 for (i=0; i < fls->mirror_array_cnt; i++)
559 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
560
561 out_sort_mirrors:
562 ff_layout_sort_mirrors(fls);
563 ret = &fls->generic_hdr;
564 dprintk("<-- %s (success)\n", __func__);
565 out_free_page:
566 __free_page(scratch);
567 return ret;
568 out_err_free:
569 _ff_layout_free_lseg(fls);
570 ret = ERR_PTR(rc);
571 dprintk("<-- %s (%d)\n", __func__, rc);
572 goto out_free_page;
573 }
574
575 static void
ff_layout_free_lseg(struct pnfs_layout_segment * lseg)576 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
577 {
578 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
579
580 dprintk("--> %s\n", __func__);
581
582 if (lseg->pls_range.iomode == IOMODE_RW) {
583 struct nfs4_flexfile_layout *ffl;
584 struct inode *inode;
585
586 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
587 inode = ffl->generic_hdr.plh_inode;
588 spin_lock(&inode->i_lock);
589 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
590 spin_unlock(&inode->i_lock);
591 }
592 _ff_layout_free_lseg(fls);
593 }
594
595 static void
nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer * timer,ktime_t now)596 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
597 {
598 /* first IO request? */
599 if (atomic_inc_return(&timer->n_ops) == 1) {
600 timer->start_time = now;
601 }
602 }
603
604 static ktime_t
nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer * timer,ktime_t now)605 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
606 {
607 ktime_t start;
608
609 if (atomic_dec_return(&timer->n_ops) < 0)
610 WARN_ON_ONCE(1);
611
612 start = timer->start_time;
613 timer->start_time = now;
614 return ktime_sub(now, start);
615 }
616
617 static bool
nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror * mirror,struct nfs4_ff_layoutstat * layoutstat,ktime_t now)618 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
619 struct nfs4_ff_layoutstat *layoutstat,
620 ktime_t now)
621 {
622 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
623 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
624
625 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
626 if (!mirror->start_time)
627 mirror->start_time = now;
628 if (mirror->report_interval != 0)
629 report_interval = (s64)mirror->report_interval * 1000LL;
630 else if (layoutstats_timer != 0)
631 report_interval = (s64)layoutstats_timer * 1000LL;
632 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
633 report_interval) {
634 ffl->last_report_time = now;
635 return true;
636 }
637
638 return false;
639 }
640
641 static void
nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat * layoutstat,__u64 requested)642 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
643 __u64 requested)
644 {
645 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
646
647 iostat->ops_requested++;
648 iostat->bytes_requested += requested;
649 }
650
651 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)652 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
653 __u64 requested,
654 __u64 completed,
655 ktime_t time_completed,
656 ktime_t time_started)
657 {
658 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
659 ktime_t completion_time = ktime_sub(time_completed, time_started);
660 ktime_t timer;
661
662 iostat->ops_completed++;
663 iostat->bytes_completed += completed;
664 iostat->bytes_not_delivered += requested - completed;
665
666 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
667 iostat->total_busy_time =
668 ktime_add(iostat->total_busy_time, timer);
669 iostat->aggregate_completion_time =
670 ktime_add(iostat->aggregate_completion_time,
671 completion_time);
672 }
673
674 static void
nfs4_ff_layout_stat_io_start_read(struct inode * inode,struct nfs4_ff_layout_mirror * mirror,__u64 requested,ktime_t now)675 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
676 struct nfs4_ff_layout_mirror *mirror,
677 __u64 requested, ktime_t now)
678 {
679 bool report;
680
681 spin_lock(&mirror->lock);
682 report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
683 nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
684 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
685 spin_unlock(&mirror->lock);
686
687 if (report)
688 pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
689 }
690
691 static void
nfs4_ff_layout_stat_io_end_read(struct rpc_task * task,struct nfs4_ff_layout_mirror * mirror,__u64 requested,__u64 completed)692 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
693 struct nfs4_ff_layout_mirror *mirror,
694 __u64 requested,
695 __u64 completed)
696 {
697 spin_lock(&mirror->lock);
698 nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
699 requested, completed,
700 ktime_get(), task->tk_start);
701 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
702 spin_unlock(&mirror->lock);
703 }
704
705 static void
nfs4_ff_layout_stat_io_start_write(struct inode * inode,struct nfs4_ff_layout_mirror * mirror,__u64 requested,ktime_t now)706 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
707 struct nfs4_ff_layout_mirror *mirror,
708 __u64 requested, ktime_t now)
709 {
710 bool report;
711
712 spin_lock(&mirror->lock);
713 report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
714 nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
715 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
716 spin_unlock(&mirror->lock);
717
718 if (report)
719 pnfs_report_layoutstat(inode, nfs_io_gfp_mask());
720 }
721
722 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)723 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
724 struct nfs4_ff_layout_mirror *mirror,
725 __u64 requested,
726 __u64 completed,
727 enum nfs3_stable_how committed)
728 {
729 if (committed == NFS_UNSTABLE)
730 requested = completed = 0;
731
732 spin_lock(&mirror->lock);
733 nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
734 requested, completed, ktime_get(), task->tk_start);
735 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
736 spin_unlock(&mirror->lock);
737 }
738
739 static void
ff_layout_mark_ds_unreachable(struct pnfs_layout_segment * lseg,u32 idx)740 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
741 {
742 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
743
744 if (devid)
745 nfs4_mark_deviceid_unavailable(devid);
746 }
747
748 static void
ff_layout_mark_ds_reachable(struct pnfs_layout_segment * lseg,u32 idx)749 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
750 {
751 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
752
753 if (devid)
754 nfs4_mark_deviceid_available(devid);
755 }
756
757 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)758 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
759 u32 start_idx, u32 *best_idx,
760 bool check_device)
761 {
762 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
763 struct nfs4_ff_layout_mirror *mirror;
764 struct nfs4_pnfs_ds *ds = ERR_PTR(-EAGAIN);
765 u32 idx;
766
767 /* mirrors are initially sorted by efficiency */
768 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
769 mirror = FF_LAYOUT_COMP(lseg, idx);
770 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
771 if (IS_ERR(ds))
772 continue;
773
774 if (check_device &&
775 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node)) {
776 // reinitialize the error state in case if this is the last iteration
777 ds = ERR_PTR(-EINVAL);
778 continue;
779 }
780
781 *best_idx = idx;
782 break;
783 }
784
785 return ds;
786 }
787
788 static struct nfs4_pnfs_ds *
ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment * lseg,u32 start_idx,u32 * best_idx)789 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
790 u32 start_idx, u32 *best_idx)
791 {
792 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
793 }
794
795 static struct nfs4_pnfs_ds *
ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment * lseg,u32 start_idx,u32 * best_idx)796 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
797 u32 start_idx, u32 *best_idx)
798 {
799 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
800 }
801
802 static struct nfs4_pnfs_ds *
ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment * lseg,u32 start_idx,u32 * best_idx)803 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
804 u32 start_idx, u32 *best_idx)
805 {
806 struct nfs4_pnfs_ds *ds;
807
808 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
809 if (!IS_ERR(ds))
810 return ds;
811 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
812 }
813
814 static struct nfs4_pnfs_ds *
ff_layout_get_ds_for_read(struct nfs_pageio_descriptor * pgio,u32 * best_idx)815 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
816 u32 *best_idx)
817 {
818 struct pnfs_layout_segment *lseg = pgio->pg_lseg;
819 struct nfs4_pnfs_ds *ds;
820
821 ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
822 best_idx);
823 if (!IS_ERR(ds) || !pgio->pg_mirror_idx)
824 return ds;
825 return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
826 }
827
828 static void
ff_layout_pg_get_read(struct nfs_pageio_descriptor * pgio,struct nfs_page * req,bool strict_iomode)829 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
830 struct nfs_page *req,
831 bool strict_iomode)
832 {
833 pnfs_put_lseg(pgio->pg_lseg);
834 pgio->pg_lseg =
835 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
836 req_offset(req), req->wb_bytes, IOMODE_READ,
837 strict_iomode, nfs_io_gfp_mask());
838 if (IS_ERR(pgio->pg_lseg)) {
839 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
840 pgio->pg_lseg = NULL;
841 }
842 }
843
844 static void
ff_layout_pg_init_read(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)845 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
846 struct nfs_page *req)
847 {
848 struct nfs_pgio_mirror *pgm;
849 struct nfs4_ff_layout_mirror *mirror;
850 struct nfs4_pnfs_ds *ds;
851 u32 ds_idx;
852
853 if (NFS_SERVER(pgio->pg_inode)->flags &
854 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
855 pgio->pg_maxretrans = io_maxretrans;
856 retry:
857 pnfs_generic_pg_check_layout(pgio, req);
858 /* Use full layout for now */
859 if (!pgio->pg_lseg) {
860 ff_layout_pg_get_read(pgio, req, false);
861 if (!pgio->pg_lseg)
862 goto out_nolseg;
863 }
864 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
865 ff_layout_pg_get_read(pgio, req, true);
866 if (!pgio->pg_lseg)
867 goto out_nolseg;
868 }
869 /* Reset wb_nio, since getting layout segment was successful */
870 req->wb_nio = 0;
871
872 ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
873 if (IS_ERR(ds)) {
874 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
875 goto out_mds;
876 pnfs_generic_pg_cleanup(pgio);
877 /* Sleep for 1 second before retrying */
878 ssleep(1);
879 goto retry;
880 }
881
882 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
883 pgm = &pgio->pg_mirrors[0];
884 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
885
886 pgio->pg_mirror_idx = ds_idx;
887 return;
888 out_nolseg:
889 if (pgio->pg_error < 0) {
890 if (pgio->pg_error != -EAGAIN)
891 return;
892 /* Retry getting layout segment if lower layer returned -EAGAIN */
893 if (pgio->pg_maxretrans && req->wb_nio++ > pgio->pg_maxretrans) {
894 if (NFS_SERVER(pgio->pg_inode)->flags & NFS_MOUNT_SOFTERR)
895 pgio->pg_error = -ETIMEDOUT;
896 else
897 pgio->pg_error = -EIO;
898 return;
899 }
900 pgio->pg_error = 0;
901 /* Sleep for 1 second before retrying */
902 ssleep(1);
903 goto retry;
904 }
905 out_mds:
906 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
907 0, NFS4_MAX_UINT64, IOMODE_READ,
908 NFS_I(pgio->pg_inode)->layout,
909 pgio->pg_lseg);
910 pgio->pg_maxretrans = 0;
911 nfs_pageio_reset_read_mds(pgio);
912 }
913
914 static void
ff_layout_pg_init_write(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)915 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
916 struct nfs_page *req)
917 {
918 struct nfs4_ff_layout_mirror *mirror;
919 struct nfs_pgio_mirror *pgm;
920 struct nfs4_pnfs_ds *ds;
921 u32 i;
922
923 retry:
924 pnfs_generic_pg_check_layout(pgio, req);
925 if (!pgio->pg_lseg) {
926 pgio->pg_lseg =
927 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
928 req_offset(req), req->wb_bytes,
929 IOMODE_RW, false, nfs_io_gfp_mask());
930 if (IS_ERR(pgio->pg_lseg)) {
931 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
932 pgio->pg_lseg = NULL;
933 return;
934 }
935 }
936 /* If no lseg, fall back to write through mds */
937 if (pgio->pg_lseg == NULL)
938 goto out_mds;
939
940 /* Use a direct mapping of ds_idx to pgio mirror_idx */
941 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
942 goto out_eagain;
943
944 for (i = 0; i < pgio->pg_mirror_count; i++) {
945 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
946 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
947 if (IS_ERR(ds)) {
948 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
949 goto out_mds;
950 pnfs_generic_pg_cleanup(pgio);
951 /* Sleep for 1 second before retrying */
952 ssleep(1);
953 goto retry;
954 }
955 pgm = &pgio->pg_mirrors[i];
956 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
957 }
958
959 if (NFS_SERVER(pgio->pg_inode)->flags &
960 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
961 pgio->pg_maxretrans = io_maxretrans;
962 return;
963 out_eagain:
964 pnfs_generic_pg_cleanup(pgio);
965 pgio->pg_error = -EAGAIN;
966 return;
967 out_mds:
968 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
969 0, NFS4_MAX_UINT64, IOMODE_RW,
970 NFS_I(pgio->pg_inode)->layout,
971 pgio->pg_lseg);
972 pgio->pg_maxretrans = 0;
973 nfs_pageio_reset_write_mds(pgio);
974 pgio->pg_error = -EAGAIN;
975 }
976
977 static unsigned int
ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor * pgio,struct nfs_page * req)978 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
979 struct nfs_page *req)
980 {
981 if (!pgio->pg_lseg) {
982 pgio->pg_lseg =
983 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req),
984 req_offset(req), req->wb_bytes,
985 IOMODE_RW, false, nfs_io_gfp_mask());
986 if (IS_ERR(pgio->pg_lseg)) {
987 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
988 pgio->pg_lseg = NULL;
989 goto out;
990 }
991 }
992 if (pgio->pg_lseg)
993 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
994
995 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
996 0, NFS4_MAX_UINT64, IOMODE_RW,
997 NFS_I(pgio->pg_inode)->layout,
998 pgio->pg_lseg);
999 /* no lseg means that pnfs is not in use, so no mirroring here */
1000 nfs_pageio_reset_write_mds(pgio);
1001 out:
1002 return 1;
1003 }
1004
1005 static u32
ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor * desc,u32 idx)1006 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
1007 {
1008 u32 old = desc->pg_mirror_idx;
1009
1010 desc->pg_mirror_idx = idx;
1011 return old;
1012 }
1013
1014 static struct nfs_pgio_mirror *
ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor * desc,u32 idx)1015 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx)
1016 {
1017 return &desc->pg_mirrors[idx];
1018 }
1019
1020 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1021 .pg_init = ff_layout_pg_init_read,
1022 .pg_test = pnfs_generic_pg_test,
1023 .pg_doio = pnfs_generic_pg_readpages,
1024 .pg_cleanup = pnfs_generic_pg_cleanup,
1025 };
1026
1027 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1028 .pg_init = ff_layout_pg_init_write,
1029 .pg_test = pnfs_generic_pg_test,
1030 .pg_doio = pnfs_generic_pg_writepages,
1031 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1032 .pg_cleanup = pnfs_generic_pg_cleanup,
1033 .pg_get_mirror = ff_layout_pg_get_mirror_write,
1034 .pg_set_mirror = ff_layout_pg_set_mirror_write,
1035 };
1036
ff_layout_reset_write(struct nfs_pgio_header * hdr,bool retry_pnfs)1037 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1038 {
1039 struct rpc_task *task = &hdr->task;
1040
1041 pnfs_layoutcommit_inode(hdr->inode, false);
1042
1043 if (retry_pnfs) {
1044 dprintk("%s Reset task %5u for i/o through pNFS "
1045 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1046 hdr->task.tk_pid,
1047 hdr->inode->i_sb->s_id,
1048 (unsigned long long)NFS_FILEID(hdr->inode),
1049 hdr->args.count,
1050 (unsigned long long)hdr->args.offset);
1051
1052 hdr->completion_ops->reschedule_io(hdr);
1053 return;
1054 }
1055
1056 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1057 dprintk("%s Reset task %5u for i/o through MDS "
1058 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1059 hdr->task.tk_pid,
1060 hdr->inode->i_sb->s_id,
1061 (unsigned long long)NFS_FILEID(hdr->inode),
1062 hdr->args.count,
1063 (unsigned long long)hdr->args.offset);
1064
1065 trace_pnfs_mds_fallback_write_done(hdr->inode,
1066 hdr->args.offset, hdr->args.count,
1067 IOMODE_RW, NFS_I(hdr->inode)->layout,
1068 hdr->lseg);
1069 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1070 }
1071 }
1072
ff_layout_resend_pnfs_read(struct nfs_pgio_header * hdr)1073 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1074 {
1075 u32 idx = hdr->pgio_mirror_idx + 1;
1076 u32 new_idx = 0;
1077 struct nfs4_pnfs_ds *ds;
1078
1079 ds = ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx);
1080 if (IS_ERR(ds))
1081 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1082 else
1083 ff_layout_send_layouterror(hdr->lseg);
1084 pnfs_read_resend_pnfs(hdr, new_idx);
1085 }
1086
ff_layout_reset_read(struct nfs_pgio_header * hdr)1087 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1088 {
1089 struct rpc_task *task = &hdr->task;
1090
1091 pnfs_layoutcommit_inode(hdr->inode, false);
1092 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1093
1094 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1095 dprintk("%s Reset task %5u for i/o through MDS "
1096 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1097 hdr->task.tk_pid,
1098 hdr->inode->i_sb->s_id,
1099 (unsigned long long)NFS_FILEID(hdr->inode),
1100 hdr->args.count,
1101 (unsigned long long)hdr->args.offset);
1102
1103 trace_pnfs_mds_fallback_read_done(hdr->inode,
1104 hdr->args.offset, hdr->args.count,
1105 IOMODE_READ, NFS_I(hdr->inode)->layout,
1106 hdr->lseg);
1107 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1108 }
1109 }
1110
ff_layout_async_handle_error_v4(struct rpc_task * task,u32 op_status,struct nfs4_state * state,struct nfs_client * clp,struct pnfs_layout_segment * lseg,u32 idx)1111 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1112 u32 op_status,
1113 struct nfs4_state *state,
1114 struct nfs_client *clp,
1115 struct pnfs_layout_segment *lseg,
1116 u32 idx)
1117 {
1118 struct pnfs_layout_hdr *lo = lseg->pls_layout;
1119 struct inode *inode = lo->plh_inode;
1120 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1121 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1122
1123 switch (op_status) {
1124 case NFS4_OK:
1125 case NFS4ERR_NXIO:
1126 break;
1127 case NFSERR_PERM:
1128 if (!task->tk_xprt)
1129 break;
1130 xprt_force_disconnect(task->tk_xprt);
1131 goto out_retry;
1132 case NFS4ERR_BADSESSION:
1133 case NFS4ERR_BADSLOT:
1134 case NFS4ERR_BAD_HIGH_SLOT:
1135 case NFS4ERR_DEADSESSION:
1136 case NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1137 case NFS4ERR_SEQ_FALSE_RETRY:
1138 case NFS4ERR_SEQ_MISORDERED:
1139 dprintk("%s ERROR %d, Reset session. Exchangeid "
1140 "flags 0x%x\n", __func__, task->tk_status,
1141 clp->cl_exchange_flags);
1142 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1143 goto out_retry;
1144 case NFS4ERR_DELAY:
1145 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1146 fallthrough;
1147 case NFS4ERR_GRACE:
1148 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1149 goto out_retry;
1150 case NFS4ERR_RETRY_UNCACHED_REP:
1151 goto out_retry;
1152 /* Invalidate Layout errors */
1153 case NFS4ERR_PNFS_NO_LAYOUT:
1154 case NFS4ERR_STALE:
1155 case NFS4ERR_BADHANDLE:
1156 case NFS4ERR_ISDIR:
1157 case NFS4ERR_FHEXPIRED:
1158 case NFS4ERR_WRONG_TYPE:
1159 dprintk("%s Invalid layout error %d\n", __func__,
1160 task->tk_status);
1161 /*
1162 * Destroy layout so new i/o will get a new layout.
1163 * Layout will not be destroyed until all current lseg
1164 * references are put. Mark layout as invalid to resend failed
1165 * i/o and all i/o waiting on the slot table to the MDS until
1166 * layout is destroyed and a new valid layout is obtained.
1167 */
1168 pnfs_destroy_layout(NFS_I(inode));
1169 rpc_wake_up(&tbl->slot_tbl_waitq);
1170 goto reset;
1171 default:
1172 break;
1173 }
1174
1175 switch (task->tk_status) {
1176 /* RPC connection errors */
1177 case -ECONNREFUSED:
1178 case -EHOSTDOWN:
1179 case -EHOSTUNREACH:
1180 case -ENETUNREACH:
1181 case -EIO:
1182 case -ETIMEDOUT:
1183 case -EPIPE:
1184 case -EPROTO:
1185 case -ENODEV:
1186 dprintk("%s DS connection error %d\n", __func__,
1187 task->tk_status);
1188 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1189 &devid->deviceid);
1190 rpc_wake_up(&tbl->slot_tbl_waitq);
1191 break;
1192 default:
1193 break;
1194 }
1195
1196 if (ff_layout_avoid_mds_available_ds(lseg))
1197 return -NFS4ERR_RESET_TO_PNFS;
1198 reset:
1199 dprintk("%s Retry through MDS. Error %d\n", __func__,
1200 task->tk_status);
1201 return -NFS4ERR_RESET_TO_MDS;
1202
1203 out_retry:
1204 task->tk_status = 0;
1205 return -EAGAIN;
1206 }
1207
1208 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
ff_layout_async_handle_error_v3(struct rpc_task * task,u32 op_status,struct nfs_client * clp,struct pnfs_layout_segment * lseg,u32 idx)1209 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1210 u32 op_status,
1211 struct nfs_client *clp,
1212 struct pnfs_layout_segment *lseg,
1213 u32 idx)
1214 {
1215 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1216
1217 switch (op_status) {
1218 case NFS_OK:
1219 case NFSERR_NXIO:
1220 break;
1221 case NFSERR_PERM:
1222 if (!task->tk_xprt)
1223 break;
1224 xprt_force_disconnect(task->tk_xprt);
1225 goto out_retry;
1226 case NFSERR_ACCES:
1227 case NFSERR_BADHANDLE:
1228 case NFSERR_FBIG:
1229 case NFSERR_IO:
1230 case NFSERR_NOSPC:
1231 case NFSERR_ROFS:
1232 case NFSERR_STALE:
1233 goto out_reset_to_pnfs;
1234 case NFSERR_JUKEBOX:
1235 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1236 goto out_retry;
1237 default:
1238 break;
1239 }
1240
1241 switch (task->tk_status) {
1242 /* File access problems. Don't mark the device as unavailable */
1243 case -EACCES:
1244 case -ESTALE:
1245 case -EISDIR:
1246 case -EBADHANDLE:
1247 case -ELOOP:
1248 case -ENOSPC:
1249 break;
1250 case -EJUKEBOX:
1251 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1252 goto out_retry;
1253 default:
1254 dprintk("%s DS connection error %d\n", __func__,
1255 task->tk_status);
1256 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1257 &devid->deviceid);
1258 }
1259 out_reset_to_pnfs:
1260 /* FIXME: Need to prevent infinite looping here. */
1261 return -NFS4ERR_RESET_TO_PNFS;
1262 out_retry:
1263 task->tk_status = 0;
1264 rpc_restart_call_prepare(task);
1265 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1266 return -EAGAIN;
1267 }
1268
ff_layout_async_handle_error(struct rpc_task * task,u32 op_status,struct nfs4_state * state,struct nfs_client * clp,struct pnfs_layout_segment * lseg,u32 idx)1269 static int ff_layout_async_handle_error(struct rpc_task *task,
1270 u32 op_status,
1271 struct nfs4_state *state,
1272 struct nfs_client *clp,
1273 struct pnfs_layout_segment *lseg,
1274 u32 idx)
1275 {
1276 int vers = clp->cl_nfs_mod->rpc_vers->number;
1277
1278 if (task->tk_status >= 0) {
1279 ff_layout_mark_ds_reachable(lseg, idx);
1280 return 0;
1281 }
1282
1283 /* Handle the case of an invalid layout segment */
1284 if (!pnfs_is_valid_lseg(lseg))
1285 return -NFS4ERR_RESET_TO_PNFS;
1286
1287 switch (vers) {
1288 case 3:
1289 return ff_layout_async_handle_error_v3(task, op_status, clp,
1290 lseg, idx);
1291 case 4:
1292 return ff_layout_async_handle_error_v4(task, op_status, state,
1293 clp, lseg, idx);
1294 default:
1295 /* should never happen */
1296 WARN_ON_ONCE(1);
1297 return 0;
1298 }
1299 }
1300
ff_layout_io_track_ds_error(struct pnfs_layout_segment * lseg,u32 idx,u64 offset,u64 length,u32 * op_status,int opnum,int error)1301 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1302 u32 idx, u64 offset, u64 length,
1303 u32 *op_status, int opnum, int error)
1304 {
1305 struct nfs4_ff_layout_mirror *mirror;
1306 u32 status = *op_status;
1307 int err;
1308
1309 if (status == 0) {
1310 switch (error) {
1311 case -ETIMEDOUT:
1312 case -EPFNOSUPPORT:
1313 case -EPROTONOSUPPORT:
1314 case -EOPNOTSUPP:
1315 case -EINVAL:
1316 case -ECONNREFUSED:
1317 case -ECONNRESET:
1318 case -EHOSTDOWN:
1319 case -EHOSTUNREACH:
1320 case -ENETDOWN:
1321 case -ENETUNREACH:
1322 case -EADDRINUSE:
1323 case -ENOBUFS:
1324 case -EPIPE:
1325 case -EPERM:
1326 case -EPROTO:
1327 case -ENODEV:
1328 *op_status = status = NFS4ERR_NXIO;
1329 break;
1330 case -EACCES:
1331 *op_status = status = NFS4ERR_ACCESS;
1332 break;
1333 default:
1334 return;
1335 }
1336 }
1337
1338 mirror = FF_LAYOUT_COMP(lseg, idx);
1339 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1340 mirror, offset, length, status, opnum,
1341 nfs_io_gfp_mask());
1342
1343 switch (status) {
1344 case NFS4ERR_DELAY:
1345 case NFS4ERR_GRACE:
1346 case NFS4ERR_PERM:
1347 break;
1348 case NFS4ERR_NXIO:
1349 ff_layout_mark_ds_unreachable(lseg, idx);
1350 /*
1351 * Don't return the layout if this is a read and we still
1352 * have layouts to try
1353 */
1354 if (opnum == OP_READ)
1355 break;
1356 fallthrough;
1357 default:
1358 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1359 lseg);
1360 }
1361
1362 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1363 }
1364
1365 /* NFS_PROTO call done callback routines */
ff_layout_read_done_cb(struct rpc_task * task,struct nfs_pgio_header * hdr)1366 static int ff_layout_read_done_cb(struct rpc_task *task,
1367 struct nfs_pgio_header *hdr)
1368 {
1369 int err;
1370
1371 if (task->tk_status < 0) {
1372 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1373 hdr->args.offset, hdr->args.count,
1374 &hdr->res.op_status, OP_READ,
1375 task->tk_status);
1376 trace_ff_layout_read_error(hdr);
1377 }
1378
1379 err = ff_layout_async_handle_error(task, hdr->res.op_status,
1380 hdr->args.context->state,
1381 hdr->ds_clp, hdr->lseg,
1382 hdr->pgio_mirror_idx);
1383
1384 trace_nfs4_pnfs_read(hdr, err);
1385 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1386 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1387 switch (err) {
1388 case -NFS4ERR_RESET_TO_PNFS:
1389 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1390 return task->tk_status;
1391 case -NFS4ERR_RESET_TO_MDS:
1392 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1393 return task->tk_status;
1394 case -EAGAIN:
1395 goto out_eagain;
1396 }
1397
1398 return 0;
1399 out_eagain:
1400 rpc_restart_call_prepare(task);
1401 return -EAGAIN;
1402 }
1403
1404 static bool
ff_layout_need_layoutcommit(struct pnfs_layout_segment * lseg)1405 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1406 {
1407 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1408 }
1409
1410 /*
1411 * We reference the rpc_cred of the first WRITE that triggers the need for
1412 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1413 * rfc5661 is not clear about which credential should be used.
1414 *
1415 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1416 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1417 * we always send layoutcommit after DS writes.
1418 */
1419 static void
ff_layout_set_layoutcommit(struct inode * inode,struct pnfs_layout_segment * lseg,loff_t end_offset)1420 ff_layout_set_layoutcommit(struct inode *inode,
1421 struct pnfs_layout_segment *lseg,
1422 loff_t end_offset)
1423 {
1424 if (!ff_layout_need_layoutcommit(lseg))
1425 return;
1426
1427 pnfs_set_layoutcommit(inode, lseg, end_offset);
1428 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1429 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1430 }
1431
ff_layout_read_record_layoutstats_start(struct rpc_task * task,struct nfs_pgio_header * hdr)1432 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1433 struct nfs_pgio_header *hdr)
1434 {
1435 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1436 return;
1437 nfs4_ff_layout_stat_io_start_read(hdr->inode,
1438 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1439 hdr->args.count,
1440 task->tk_start);
1441 }
1442
ff_layout_read_record_layoutstats_done(struct rpc_task * task,struct nfs_pgio_header * hdr)1443 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1444 struct nfs_pgio_header *hdr)
1445 {
1446 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1447 return;
1448 nfs4_ff_layout_stat_io_end_read(task,
1449 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1450 hdr->args.count,
1451 hdr->res.count);
1452 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1453 }
1454
ff_layout_read_prepare_common(struct rpc_task * task,struct nfs_pgio_header * hdr)1455 static int ff_layout_read_prepare_common(struct rpc_task *task,
1456 struct nfs_pgio_header *hdr)
1457 {
1458 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1459 rpc_exit(task, -EIO);
1460 return -EIO;
1461 }
1462
1463 if (!pnfs_is_valid_lseg(hdr->lseg)) {
1464 rpc_exit(task, -EAGAIN);
1465 return -EAGAIN;
1466 }
1467
1468 ff_layout_read_record_layoutstats_start(task, hdr);
1469 return 0;
1470 }
1471
1472 /*
1473 * Call ops for the async read/write cases
1474 * In the case of dense layouts, the offset needs to be reset to its
1475 * original value.
1476 */
ff_layout_read_prepare_v3(struct rpc_task * task,void * data)1477 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1478 {
1479 struct nfs_pgio_header *hdr = data;
1480
1481 if (ff_layout_read_prepare_common(task, hdr))
1482 return;
1483
1484 rpc_call_start(task);
1485 }
1486
ff_layout_read_prepare_v4(struct rpc_task * task,void * data)1487 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1488 {
1489 struct nfs_pgio_header *hdr = data;
1490
1491 if (nfs4_setup_sequence(hdr->ds_clp,
1492 &hdr->args.seq_args,
1493 &hdr->res.seq_res,
1494 task))
1495 return;
1496
1497 ff_layout_read_prepare_common(task, hdr);
1498 }
1499
ff_layout_read_call_done(struct rpc_task * task,void * data)1500 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1501 {
1502 struct nfs_pgio_header *hdr = data;
1503
1504 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1505 task->tk_status == 0) {
1506 nfs4_sequence_done(task, &hdr->res.seq_res);
1507 return;
1508 }
1509
1510 /* Note this may cause RPC to be resent */
1511 hdr->mds_ops->rpc_call_done(task, hdr);
1512 }
1513
ff_layout_read_count_stats(struct rpc_task * task,void * data)1514 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1515 {
1516 struct nfs_pgio_header *hdr = data;
1517
1518 ff_layout_read_record_layoutstats_done(task, hdr);
1519 rpc_count_iostats_metrics(task,
1520 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1521 }
1522
ff_layout_read_release(void * data)1523 static void ff_layout_read_release(void *data)
1524 {
1525 struct nfs_pgio_header *hdr = data;
1526
1527 ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1528 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1529 ff_layout_resend_pnfs_read(hdr);
1530 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1531 ff_layout_reset_read(hdr);
1532 pnfs_generic_rw_release(data);
1533 }
1534
1535
ff_layout_write_done_cb(struct rpc_task * task,struct nfs_pgio_header * hdr)1536 static int ff_layout_write_done_cb(struct rpc_task *task,
1537 struct nfs_pgio_header *hdr)
1538 {
1539 loff_t end_offs = 0;
1540 int err;
1541
1542 if (task->tk_status < 0) {
1543 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1544 hdr->args.offset, hdr->args.count,
1545 &hdr->res.op_status, OP_WRITE,
1546 task->tk_status);
1547 trace_ff_layout_write_error(hdr);
1548 }
1549
1550 err = ff_layout_async_handle_error(task, hdr->res.op_status,
1551 hdr->args.context->state,
1552 hdr->ds_clp, hdr->lseg,
1553 hdr->pgio_mirror_idx);
1554
1555 trace_nfs4_pnfs_write(hdr, err);
1556 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1557 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1558 switch (err) {
1559 case -NFS4ERR_RESET_TO_PNFS:
1560 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1561 return task->tk_status;
1562 case -NFS4ERR_RESET_TO_MDS:
1563 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1564 return task->tk_status;
1565 case -EAGAIN:
1566 return -EAGAIN;
1567 }
1568
1569 if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1570 hdr->res.verf->committed == NFS_DATA_SYNC)
1571 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1572
1573 /* Note: if the write is unstable, don't set end_offs until commit */
1574 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1575
1576 /* zero out fattr since we don't care DS attr at all */
1577 hdr->fattr.valid = 0;
1578 if (task->tk_status >= 0)
1579 nfs_writeback_update_inode(hdr);
1580
1581 return 0;
1582 }
1583
ff_layout_commit_done_cb(struct rpc_task * task,struct nfs_commit_data * data)1584 static int ff_layout_commit_done_cb(struct rpc_task *task,
1585 struct nfs_commit_data *data)
1586 {
1587 int err;
1588
1589 if (task->tk_status < 0) {
1590 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1591 data->args.offset, data->args.count,
1592 &data->res.op_status, OP_COMMIT,
1593 task->tk_status);
1594 trace_ff_layout_commit_error(data);
1595 }
1596
1597 err = ff_layout_async_handle_error(task, data->res.op_status,
1598 NULL, data->ds_clp, data->lseg,
1599 data->ds_commit_index);
1600
1601 trace_nfs4_pnfs_commit_ds(data, err);
1602 switch (err) {
1603 case -NFS4ERR_RESET_TO_PNFS:
1604 pnfs_generic_prepare_to_resend_writes(data);
1605 return -EAGAIN;
1606 case -NFS4ERR_RESET_TO_MDS:
1607 pnfs_generic_prepare_to_resend_writes(data);
1608 return -EAGAIN;
1609 case -EAGAIN:
1610 rpc_restart_call_prepare(task);
1611 return -EAGAIN;
1612 }
1613
1614 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1615
1616 return 0;
1617 }
1618
ff_layout_write_record_layoutstats_start(struct rpc_task * task,struct nfs_pgio_header * hdr)1619 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1620 struct nfs_pgio_header *hdr)
1621 {
1622 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1623 return;
1624 nfs4_ff_layout_stat_io_start_write(hdr->inode,
1625 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1626 hdr->args.count,
1627 task->tk_start);
1628 }
1629
ff_layout_write_record_layoutstats_done(struct rpc_task * task,struct nfs_pgio_header * hdr)1630 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1631 struct nfs_pgio_header *hdr)
1632 {
1633 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1634 return;
1635 nfs4_ff_layout_stat_io_end_write(task,
1636 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1637 hdr->args.count, hdr->res.count,
1638 hdr->res.verf->committed);
1639 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1640 }
1641
ff_layout_write_prepare_common(struct rpc_task * task,struct nfs_pgio_header * hdr)1642 static int ff_layout_write_prepare_common(struct rpc_task *task,
1643 struct nfs_pgio_header *hdr)
1644 {
1645 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1646 rpc_exit(task, -EIO);
1647 return -EIO;
1648 }
1649
1650 if (!pnfs_is_valid_lseg(hdr->lseg)) {
1651 rpc_exit(task, -EAGAIN);
1652 return -EAGAIN;
1653 }
1654
1655 ff_layout_write_record_layoutstats_start(task, hdr);
1656 return 0;
1657 }
1658
ff_layout_write_prepare_v3(struct rpc_task * task,void * data)1659 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1660 {
1661 struct nfs_pgio_header *hdr = data;
1662
1663 if (ff_layout_write_prepare_common(task, hdr))
1664 return;
1665
1666 rpc_call_start(task);
1667 }
1668
ff_layout_write_prepare_v4(struct rpc_task * task,void * data)1669 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1670 {
1671 struct nfs_pgio_header *hdr = data;
1672
1673 if (nfs4_setup_sequence(hdr->ds_clp,
1674 &hdr->args.seq_args,
1675 &hdr->res.seq_res,
1676 task))
1677 return;
1678
1679 ff_layout_write_prepare_common(task, hdr);
1680 }
1681
ff_layout_write_call_done(struct rpc_task * task,void * data)1682 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1683 {
1684 struct nfs_pgio_header *hdr = data;
1685
1686 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1687 task->tk_status == 0) {
1688 nfs4_sequence_done(task, &hdr->res.seq_res);
1689 return;
1690 }
1691
1692 /* Note this may cause RPC to be resent */
1693 hdr->mds_ops->rpc_call_done(task, hdr);
1694 }
1695
ff_layout_write_count_stats(struct rpc_task * task,void * data)1696 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1697 {
1698 struct nfs_pgio_header *hdr = data;
1699
1700 ff_layout_write_record_layoutstats_done(task, hdr);
1701 rpc_count_iostats_metrics(task,
1702 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1703 }
1704
ff_layout_write_release(void * data)1705 static void ff_layout_write_release(void *data)
1706 {
1707 struct nfs_pgio_header *hdr = data;
1708
1709 ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1710 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1711 ff_layout_send_layouterror(hdr->lseg);
1712 ff_layout_reset_write(hdr, true);
1713 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1714 ff_layout_reset_write(hdr, false);
1715 pnfs_generic_rw_release(data);
1716 }
1717
ff_layout_commit_record_layoutstats_start(struct rpc_task * task,struct nfs_commit_data * cdata)1718 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1719 struct nfs_commit_data *cdata)
1720 {
1721 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1722 return;
1723 nfs4_ff_layout_stat_io_start_write(cdata->inode,
1724 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1725 0, task->tk_start);
1726 }
1727
ff_layout_commit_record_layoutstats_done(struct rpc_task * task,struct nfs_commit_data * cdata)1728 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1729 struct nfs_commit_data *cdata)
1730 {
1731 struct nfs_page *req;
1732 __u64 count = 0;
1733
1734 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1735 return;
1736
1737 if (task->tk_status == 0) {
1738 list_for_each_entry(req, &cdata->pages, wb_list)
1739 count += req->wb_bytes;
1740 }
1741 nfs4_ff_layout_stat_io_end_write(task,
1742 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1743 count, count, NFS_FILE_SYNC);
1744 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1745 }
1746
ff_layout_commit_prepare_common(struct rpc_task * task,struct nfs_commit_data * cdata)1747 static int ff_layout_commit_prepare_common(struct rpc_task *task,
1748 struct nfs_commit_data *cdata)
1749 {
1750 if (!pnfs_is_valid_lseg(cdata->lseg)) {
1751 rpc_exit(task, -EAGAIN);
1752 return -EAGAIN;
1753 }
1754
1755 ff_layout_commit_record_layoutstats_start(task, cdata);
1756 return 0;
1757 }
1758
ff_layout_commit_prepare_v3(struct rpc_task * task,void * data)1759 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1760 {
1761 if (ff_layout_commit_prepare_common(task, data))
1762 return;
1763
1764 rpc_call_start(task);
1765 }
1766
ff_layout_commit_prepare_v4(struct rpc_task * task,void * data)1767 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1768 {
1769 struct nfs_commit_data *wdata = data;
1770
1771 if (nfs4_setup_sequence(wdata->ds_clp,
1772 &wdata->args.seq_args,
1773 &wdata->res.seq_res,
1774 task))
1775 return;
1776 ff_layout_commit_prepare_common(task, data);
1777 }
1778
ff_layout_commit_done(struct rpc_task * task,void * data)1779 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1780 {
1781 pnfs_generic_write_commit_done(task, data);
1782 }
1783
ff_layout_commit_count_stats(struct rpc_task * task,void * data)1784 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1785 {
1786 struct nfs_commit_data *cdata = data;
1787
1788 ff_layout_commit_record_layoutstats_done(task, cdata);
1789 rpc_count_iostats_metrics(task,
1790 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1791 }
1792
ff_layout_commit_release(void * data)1793 static void ff_layout_commit_release(void *data)
1794 {
1795 struct nfs_commit_data *cdata = data;
1796
1797 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1798 pnfs_generic_commit_release(data);
1799 }
1800
1801 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1802 .rpc_call_prepare = ff_layout_read_prepare_v3,
1803 .rpc_call_done = ff_layout_read_call_done,
1804 .rpc_count_stats = ff_layout_read_count_stats,
1805 .rpc_release = ff_layout_read_release,
1806 };
1807
1808 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1809 .rpc_call_prepare = ff_layout_read_prepare_v4,
1810 .rpc_call_done = ff_layout_read_call_done,
1811 .rpc_count_stats = ff_layout_read_count_stats,
1812 .rpc_release = ff_layout_read_release,
1813 };
1814
1815 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1816 .rpc_call_prepare = ff_layout_write_prepare_v3,
1817 .rpc_call_done = ff_layout_write_call_done,
1818 .rpc_count_stats = ff_layout_write_count_stats,
1819 .rpc_release = ff_layout_write_release,
1820 };
1821
1822 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1823 .rpc_call_prepare = ff_layout_write_prepare_v4,
1824 .rpc_call_done = ff_layout_write_call_done,
1825 .rpc_count_stats = ff_layout_write_count_stats,
1826 .rpc_release = ff_layout_write_release,
1827 };
1828
1829 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1830 .rpc_call_prepare = ff_layout_commit_prepare_v3,
1831 .rpc_call_done = ff_layout_commit_done,
1832 .rpc_count_stats = ff_layout_commit_count_stats,
1833 .rpc_release = ff_layout_commit_release,
1834 };
1835
1836 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1837 .rpc_call_prepare = ff_layout_commit_prepare_v4,
1838 .rpc_call_done = ff_layout_commit_done,
1839 .rpc_count_stats = ff_layout_commit_count_stats,
1840 .rpc_release = ff_layout_commit_release,
1841 };
1842
1843 static enum pnfs_try_status
ff_layout_read_pagelist(struct nfs_pgio_header * hdr)1844 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1845 {
1846 struct pnfs_layout_segment *lseg = hdr->lseg;
1847 struct nfs4_pnfs_ds *ds;
1848 struct rpc_clnt *ds_clnt;
1849 struct nfsd_file *localio;
1850 struct nfs4_ff_layout_mirror *mirror;
1851 const struct cred *ds_cred;
1852 loff_t offset = hdr->args.offset;
1853 u32 idx = hdr->pgio_mirror_idx;
1854 int vers;
1855 struct nfs_fh *fh;
1856 bool ds_fatal_error = false;
1857
1858 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1859 __func__, hdr->inode->i_ino,
1860 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1861
1862 mirror = FF_LAYOUT_COMP(lseg, idx);
1863 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1864 if (IS_ERR(ds)) {
1865 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds));
1866 goto out_failed;
1867 }
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 USE DS: %s cl_count %d vers %d\n", __func__,
1881 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1882
1883 hdr->pgio_done_cb = ff_layout_read_done_cb;
1884 refcount_inc(&ds->ds_clp->cl_count);
1885 hdr->ds_clp = ds->ds_clp;
1886 fh = nfs4_ff_layout_select_ds_fh(mirror);
1887 if (fh)
1888 hdr->args.fh = fh;
1889
1890 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1891
1892 /*
1893 * Note that if we ever decide to split across DSes,
1894 * then we may need to handle dense-like offsets.
1895 */
1896 hdr->args.offset = offset;
1897 hdr->mds_offset = offset;
1898
1899 /* Start IO accounting for local read */
1900 localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh, FMODE_READ);
1901 if (localio) {
1902 hdr->task.tk_start = ktime_get();
1903 ff_layout_read_record_layoutstats_start(&hdr->task, hdr);
1904 }
1905
1906 /* Perform an asynchronous read to ds */
1907 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1908 vers == 3 ? &ff_layout_read_call_ops_v3 :
1909 &ff_layout_read_call_ops_v4,
1910 0, RPC_TASK_SOFTCONN, localio);
1911 put_cred(ds_cred);
1912 return PNFS_ATTEMPTED;
1913
1914 out_failed:
1915 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error)
1916 return PNFS_TRY_AGAIN;
1917 trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1918 hdr->args.offset, hdr->args.count,
1919 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1920 return PNFS_NOT_ATTEMPTED;
1921 }
1922
1923 /* Perform async writes. */
1924 static enum pnfs_try_status
ff_layout_write_pagelist(struct nfs_pgio_header * hdr,int sync)1925 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1926 {
1927 struct pnfs_layout_segment *lseg = hdr->lseg;
1928 struct nfs4_pnfs_ds *ds;
1929 struct rpc_clnt *ds_clnt;
1930 struct nfsd_file *localio;
1931 struct nfs4_ff_layout_mirror *mirror;
1932 const struct cred *ds_cred;
1933 loff_t offset = hdr->args.offset;
1934 int vers;
1935 struct nfs_fh *fh;
1936 u32 idx = hdr->pgio_mirror_idx;
1937 bool ds_fatal_error = false;
1938
1939 mirror = FF_LAYOUT_COMP(lseg, idx);
1940 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1941 if (IS_ERR(ds)) {
1942 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds));
1943 goto out_failed;
1944 }
1945
1946 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1947 hdr->inode);
1948 if (IS_ERR(ds_clnt))
1949 goto out_failed;
1950
1951 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1952 if (!ds_cred)
1953 goto out_failed;
1954
1955 vers = nfs4_ff_layout_ds_version(mirror);
1956
1957 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1958 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1959 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1960 vers);
1961
1962 hdr->pgio_done_cb = ff_layout_write_done_cb;
1963 refcount_inc(&ds->ds_clp->cl_count);
1964 hdr->ds_clp = ds->ds_clp;
1965 hdr->ds_commit_idx = idx;
1966 fh = nfs4_ff_layout_select_ds_fh(mirror);
1967 if (fh)
1968 hdr->args.fh = fh;
1969
1970 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1971
1972 /*
1973 * Note that if we ever decide to split across DSes,
1974 * then we may need to handle dense-like offsets.
1975 */
1976 hdr->args.offset = offset;
1977
1978 /* Start IO accounting for local write */
1979 localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh,
1980 FMODE_READ|FMODE_WRITE);
1981 if (localio) {
1982 hdr->task.tk_start = ktime_get();
1983 ff_layout_write_record_layoutstats_start(&hdr->task, hdr);
1984 }
1985
1986 /* Perform an asynchronous write */
1987 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1988 vers == 3 ? &ff_layout_write_call_ops_v3 :
1989 &ff_layout_write_call_ops_v4,
1990 sync, RPC_TASK_SOFTCONN, localio);
1991 put_cred(ds_cred);
1992 return PNFS_ATTEMPTED;
1993
1994 out_failed:
1995 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error)
1996 return PNFS_TRY_AGAIN;
1997 trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1998 hdr->args.offset, hdr->args.count,
1999 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
2000 return PNFS_NOT_ATTEMPTED;
2001 }
2002
calc_ds_index_from_commit(struct pnfs_layout_segment * lseg,u32 i)2003 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
2004 {
2005 return i;
2006 }
2007
2008 static struct nfs_fh *
select_ds_fh_from_commit(struct pnfs_layout_segment * lseg,u32 i)2009 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
2010 {
2011 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2012
2013 /* FIXME: Assume that there is only one NFS version available
2014 * for the DS.
2015 */
2016 return &flseg->mirror_array[i]->fh_versions[0];
2017 }
2018
ff_layout_initiate_commit(struct nfs_commit_data * data,int how)2019 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
2020 {
2021 struct pnfs_layout_segment *lseg = data->lseg;
2022 struct nfs4_pnfs_ds *ds;
2023 struct rpc_clnt *ds_clnt;
2024 struct nfsd_file *localio;
2025 struct nfs4_ff_layout_mirror *mirror;
2026 const struct cred *ds_cred;
2027 u32 idx;
2028 int vers, ret;
2029 struct nfs_fh *fh;
2030
2031 if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
2032 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
2033 goto out_err;
2034
2035 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
2036 mirror = FF_LAYOUT_COMP(lseg, idx);
2037 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
2038 if (IS_ERR(ds))
2039 goto out_err;
2040
2041 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
2042 data->inode);
2043 if (IS_ERR(ds_clnt))
2044 goto out_err;
2045
2046 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
2047 if (!ds_cred)
2048 goto out_err;
2049
2050 vers = nfs4_ff_layout_ds_version(mirror);
2051
2052 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
2053 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
2054 vers);
2055 data->commit_done_cb = ff_layout_commit_done_cb;
2056 data->cred = ds_cred;
2057 refcount_inc(&ds->ds_clp->cl_count);
2058 data->ds_clp = ds->ds_clp;
2059 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
2060 if (fh)
2061 data->args.fh = fh;
2062
2063 /* Start IO accounting for local commit */
2064 localio = ff_local_open_fh(ds->ds_clp, ds_cred, fh,
2065 FMODE_READ|FMODE_WRITE);
2066 if (localio) {
2067 data->task.tk_start = ktime_get();
2068 ff_layout_commit_record_layoutstats_start(&data->task, data);
2069 }
2070
2071 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
2072 vers == 3 ? &ff_layout_commit_call_ops_v3 :
2073 &ff_layout_commit_call_ops_v4,
2074 how, RPC_TASK_SOFTCONN, localio);
2075 put_cred(ds_cred);
2076 return ret;
2077 out_err:
2078 pnfs_generic_prepare_to_resend_writes(data);
2079 pnfs_generic_commit_release(data);
2080 return -EAGAIN;
2081 }
2082
2083 static int
ff_layout_commit_pagelist(struct inode * inode,struct list_head * mds_pages,int how,struct nfs_commit_info * cinfo)2084 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
2085 int how, struct nfs_commit_info *cinfo)
2086 {
2087 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
2088 ff_layout_initiate_commit);
2089 }
2090
ff_layout_match_rw(const struct rpc_task * task,const struct nfs_pgio_header * hdr,const struct pnfs_layout_segment * lseg)2091 static bool ff_layout_match_rw(const struct rpc_task *task,
2092 const struct nfs_pgio_header *hdr,
2093 const struct pnfs_layout_segment *lseg)
2094 {
2095 return hdr->lseg == lseg;
2096 }
2097
ff_layout_match_commit(const struct rpc_task * task,const struct nfs_commit_data * cdata,const struct pnfs_layout_segment * lseg)2098 static bool ff_layout_match_commit(const struct rpc_task *task,
2099 const struct nfs_commit_data *cdata,
2100 const struct pnfs_layout_segment *lseg)
2101 {
2102 return cdata->lseg == lseg;
2103 }
2104
ff_layout_match_io(const struct rpc_task * task,const void * data)2105 static bool ff_layout_match_io(const struct rpc_task *task, const void *data)
2106 {
2107 const struct rpc_call_ops *ops = task->tk_ops;
2108
2109 if (ops == &ff_layout_read_call_ops_v3 ||
2110 ops == &ff_layout_read_call_ops_v4 ||
2111 ops == &ff_layout_write_call_ops_v3 ||
2112 ops == &ff_layout_write_call_ops_v4)
2113 return ff_layout_match_rw(task, task->tk_calldata, data);
2114 if (ops == &ff_layout_commit_call_ops_v3 ||
2115 ops == &ff_layout_commit_call_ops_v4)
2116 return ff_layout_match_commit(task, task->tk_calldata, data);
2117 return false;
2118 }
2119
ff_layout_cancel_io(struct pnfs_layout_segment * lseg)2120 static void ff_layout_cancel_io(struct pnfs_layout_segment *lseg)
2121 {
2122 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2123 struct nfs4_ff_layout_mirror *mirror;
2124 struct nfs4_ff_layout_ds *mirror_ds;
2125 struct nfs4_pnfs_ds *ds;
2126 struct nfs_client *ds_clp;
2127 struct rpc_clnt *clnt;
2128 u32 idx;
2129
2130 for (idx = 0; idx < flseg->mirror_array_cnt; idx++) {
2131 mirror = flseg->mirror_array[idx];
2132 mirror_ds = mirror->mirror_ds;
2133 if (IS_ERR_OR_NULL(mirror_ds))
2134 continue;
2135 ds = mirror->mirror_ds->ds;
2136 if (!ds)
2137 continue;
2138 ds_clp = ds->ds_clp;
2139 if (!ds_clp)
2140 continue;
2141 clnt = ds_clp->cl_rpcclient;
2142 if (!clnt)
2143 continue;
2144 if (!rpc_cancel_tasks(clnt, -EAGAIN, ff_layout_match_io, lseg))
2145 continue;
2146 rpc_clnt_disconnect(clnt);
2147 }
2148 }
2149
2150 static struct pnfs_ds_commit_info *
ff_layout_get_ds_info(struct inode * inode)2151 ff_layout_get_ds_info(struct inode *inode)
2152 {
2153 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
2154
2155 if (layout == NULL)
2156 return NULL;
2157
2158 return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
2159 }
2160
2161 static void
ff_layout_setup_ds_info(struct pnfs_ds_commit_info * fl_cinfo,struct pnfs_layout_segment * lseg)2162 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2163 struct pnfs_layout_segment *lseg)
2164 {
2165 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
2166 struct inode *inode = lseg->pls_layout->plh_inode;
2167 struct pnfs_commit_array *array, *new;
2168
2169 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt,
2170 nfs_io_gfp_mask());
2171 if (new) {
2172 spin_lock(&inode->i_lock);
2173 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
2174 spin_unlock(&inode->i_lock);
2175 if (array != new)
2176 pnfs_free_commit_array(new);
2177 }
2178 }
2179
2180 static void
ff_layout_release_ds_info(struct pnfs_ds_commit_info * fl_cinfo,struct inode * inode)2181 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
2182 struct inode *inode)
2183 {
2184 spin_lock(&inode->i_lock);
2185 pnfs_generic_ds_cinfo_destroy(fl_cinfo);
2186 spin_unlock(&inode->i_lock);
2187 }
2188
2189 static void
ff_layout_free_deviceid_node(struct nfs4_deviceid_node * d)2190 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2191 {
2192 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2193 id_node));
2194 }
2195
ff_layout_encode_ioerr(struct xdr_stream * xdr,const struct nfs4_layoutreturn_args * args,const struct nfs4_flexfile_layoutreturn_args * ff_args)2196 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2197 const struct nfs4_layoutreturn_args *args,
2198 const struct nfs4_flexfile_layoutreturn_args *ff_args)
2199 {
2200 __be32 *start;
2201
2202 start = xdr_reserve_space(xdr, 4);
2203 if (unlikely(!start))
2204 return -E2BIG;
2205
2206 *start = cpu_to_be32(ff_args->num_errors);
2207 /* This assume we always return _ALL_ layouts */
2208 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2209 }
2210
2211 static void
ff_layout_encode_ff_iostat_head(struct xdr_stream * xdr,const nfs4_stateid * stateid,const struct nfs42_layoutstat_devinfo * devinfo)2212 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2213 const nfs4_stateid *stateid,
2214 const struct nfs42_layoutstat_devinfo *devinfo)
2215 {
2216 __be32 *p;
2217
2218 p = xdr_reserve_space(xdr, 8 + 8);
2219 p = xdr_encode_hyper(p, devinfo->offset);
2220 p = xdr_encode_hyper(p, devinfo->length);
2221 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2222 p = xdr_reserve_space(xdr, 4*8);
2223 p = xdr_encode_hyper(p, devinfo->read_count);
2224 p = xdr_encode_hyper(p, devinfo->read_bytes);
2225 p = xdr_encode_hyper(p, devinfo->write_count);
2226 p = xdr_encode_hyper(p, devinfo->write_bytes);
2227 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2228 }
2229
2230 static void
ff_layout_encode_ff_iostat(struct xdr_stream * xdr,const nfs4_stateid * stateid,const struct nfs42_layoutstat_devinfo * devinfo)2231 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2232 const nfs4_stateid *stateid,
2233 const struct nfs42_layoutstat_devinfo *devinfo)
2234 {
2235 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2236 ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2237 devinfo->ld_private.data);
2238 }
2239
2240 /* 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)2241 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2242 const struct nfs4_layoutreturn_args *args,
2243 struct nfs4_flexfile_layoutreturn_args *ff_args)
2244 {
2245 __be32 *p;
2246 int i;
2247
2248 p = xdr_reserve_space(xdr, 4);
2249 *p = cpu_to_be32(ff_args->num_dev);
2250 for (i = 0; i < ff_args->num_dev; i++)
2251 ff_layout_encode_ff_iostat(xdr,
2252 &args->layout->plh_stateid,
2253 &ff_args->devinfo[i]);
2254 }
2255
2256 static void
ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo * devinfo,unsigned int num_entries)2257 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2258 unsigned int num_entries)
2259 {
2260 unsigned int i;
2261
2262 for (i = 0; i < num_entries; i++) {
2263 if (!devinfo[i].ld_private.ops)
2264 continue;
2265 if (!devinfo[i].ld_private.ops->free)
2266 continue;
2267 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2268 }
2269 }
2270
2271 static struct nfs4_deviceid_node *
ff_layout_alloc_deviceid_node(struct nfs_server * server,struct pnfs_device * pdev,gfp_t gfp_flags)2272 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2273 struct pnfs_device *pdev, gfp_t gfp_flags)
2274 {
2275 struct nfs4_ff_layout_ds *dsaddr;
2276
2277 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2278 if (!dsaddr)
2279 return NULL;
2280 return &dsaddr->id_node;
2281 }
2282
2283 static void
ff_layout_encode_layoutreturn(struct xdr_stream * xdr,const void * voidargs,const struct nfs4_xdr_opaque_data * ff_opaque)2284 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2285 const void *voidargs,
2286 const struct nfs4_xdr_opaque_data *ff_opaque)
2287 {
2288 const struct nfs4_layoutreturn_args *args = voidargs;
2289 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2290 struct xdr_buf tmp_buf = {
2291 .head = {
2292 [0] = {
2293 .iov_base = page_address(ff_args->pages[0]),
2294 },
2295 },
2296 .buflen = PAGE_SIZE,
2297 };
2298 struct xdr_stream tmp_xdr;
2299 __be32 *start;
2300
2301 dprintk("%s: Begin\n", __func__);
2302
2303 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2304
2305 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2306 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2307
2308 start = xdr_reserve_space(xdr, 4);
2309 *start = cpu_to_be32(tmp_buf.len);
2310 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2311
2312 dprintk("%s: Return\n", __func__);
2313 }
2314
2315 static void
ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data * args)2316 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2317 {
2318 struct nfs4_flexfile_layoutreturn_args *ff_args;
2319
2320 if (!args->data)
2321 return;
2322 ff_args = args->data;
2323 args->data = NULL;
2324
2325 ff_layout_free_ds_ioerr(&ff_args->errors);
2326 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2327
2328 put_page(ff_args->pages[0]);
2329 kfree(ff_args);
2330 }
2331
2332 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2333 .encode = ff_layout_encode_layoutreturn,
2334 .free = ff_layout_free_layoutreturn,
2335 };
2336
2337 static int
ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args * args)2338 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2339 {
2340 struct nfs4_flexfile_layoutreturn_args *ff_args;
2341 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2342
2343 ff_args = kmalloc(sizeof(*ff_args), nfs_io_gfp_mask());
2344 if (!ff_args)
2345 goto out_nomem;
2346 ff_args->pages[0] = alloc_page(nfs_io_gfp_mask());
2347 if (!ff_args->pages[0])
2348 goto out_nomem_free;
2349
2350 INIT_LIST_HEAD(&ff_args->errors);
2351 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2352 &args->range, &ff_args->errors,
2353 FF_LAYOUTRETURN_MAXERR);
2354
2355 spin_lock(&args->inode->i_lock);
2356 ff_args->num_dev = ff_layout_mirror_prepare_stats(
2357 &ff_layout->generic_hdr, &ff_args->devinfo[0],
2358 ARRAY_SIZE(ff_args->devinfo), NFS4_FF_OP_LAYOUTRETURN);
2359 spin_unlock(&args->inode->i_lock);
2360
2361 args->ld_private->ops = &layoutreturn_ops;
2362 args->ld_private->data = ff_args;
2363 return 0;
2364 out_nomem_free:
2365 kfree(ff_args);
2366 out_nomem:
2367 return -ENOMEM;
2368 }
2369
2370 #ifdef CONFIG_NFS_V4_2
2371 void
ff_layout_send_layouterror(struct pnfs_layout_segment * lseg)2372 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2373 {
2374 struct pnfs_layout_hdr *lo = lseg->pls_layout;
2375 struct nfs42_layout_error *errors;
2376 LIST_HEAD(head);
2377
2378 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2379 return;
2380 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2381 if (list_empty(&head))
2382 return;
2383
2384 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX, sizeof(*errors),
2385 nfs_io_gfp_mask());
2386 if (errors != NULL) {
2387 const struct nfs4_ff_layout_ds_err *pos;
2388 size_t n = 0;
2389
2390 list_for_each_entry(pos, &head, list) {
2391 errors[n].offset = pos->offset;
2392 errors[n].length = pos->length;
2393 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2394 errors[n].errors[0].dev_id = pos->deviceid;
2395 errors[n].errors[0].status = pos->status;
2396 errors[n].errors[0].opnum = pos->opnum;
2397 n++;
2398 if (!list_is_last(&pos->list, &head) &&
2399 n < NFS42_LAYOUTERROR_MAX)
2400 continue;
2401 if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2402 break;
2403 n = 0;
2404 }
2405 kfree(errors);
2406 }
2407 ff_layout_free_ds_ioerr(&head);
2408 }
2409 #else
2410 void
ff_layout_send_layouterror(struct pnfs_layout_segment * lseg)2411 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2412 {
2413 }
2414 #endif
2415
2416 static int
ff_layout_ntop4(const struct sockaddr * sap,char * buf,const size_t buflen)2417 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2418 {
2419 const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2420
2421 return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2422 }
2423
2424 static size_t
ff_layout_ntop6_noscopeid(const struct sockaddr * sap,char * buf,const int buflen)2425 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2426 const int buflen)
2427 {
2428 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2429 const struct in6_addr *addr = &sin6->sin6_addr;
2430
2431 /*
2432 * RFC 4291, Section 2.2.2
2433 *
2434 * Shorthanded ANY address
2435 */
2436 if (ipv6_addr_any(addr))
2437 return snprintf(buf, buflen, "::");
2438
2439 /*
2440 * RFC 4291, Section 2.2.2
2441 *
2442 * Shorthanded loopback address
2443 */
2444 if (ipv6_addr_loopback(addr))
2445 return snprintf(buf, buflen, "::1");
2446
2447 /*
2448 * RFC 4291, Section 2.2.3
2449 *
2450 * Special presentation address format for mapped v4
2451 * addresses.
2452 */
2453 if (ipv6_addr_v4mapped(addr))
2454 return snprintf(buf, buflen, "::ffff:%pI4",
2455 &addr->s6_addr32[3]);
2456
2457 /*
2458 * RFC 4291, Section 2.2.1
2459 */
2460 return snprintf(buf, buflen, "%pI6c", addr);
2461 }
2462
2463 /* Derived from rpc_sockaddr2uaddr */
2464 static void
ff_layout_encode_netaddr(struct xdr_stream * xdr,struct nfs4_pnfs_ds_addr * da)2465 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2466 {
2467 struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2468 char portbuf[RPCBIND_MAXUADDRPLEN];
2469 char addrbuf[RPCBIND_MAXUADDRLEN];
2470 unsigned short port;
2471 int len, netid_len;
2472 __be32 *p;
2473
2474 switch (sap->sa_family) {
2475 case AF_INET:
2476 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2477 return;
2478 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2479 break;
2480 case AF_INET6:
2481 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2482 return;
2483 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2484 break;
2485 default:
2486 WARN_ON_ONCE(1);
2487 return;
2488 }
2489
2490 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2491 len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2492
2493 netid_len = strlen(da->da_netid);
2494 p = xdr_reserve_space(xdr, 4 + netid_len);
2495 xdr_encode_opaque(p, da->da_netid, netid_len);
2496
2497 p = xdr_reserve_space(xdr, 4 + len);
2498 xdr_encode_opaque(p, addrbuf, len);
2499 }
2500
2501 static void
ff_layout_encode_nfstime(struct xdr_stream * xdr,ktime_t t)2502 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2503 ktime_t t)
2504 {
2505 struct timespec64 ts;
2506 __be32 *p;
2507
2508 p = xdr_reserve_space(xdr, 12);
2509 ts = ktime_to_timespec64(t);
2510 p = xdr_encode_hyper(p, ts.tv_sec);
2511 *p++ = cpu_to_be32(ts.tv_nsec);
2512 }
2513
2514 static void
ff_layout_encode_io_latency(struct xdr_stream * xdr,struct nfs4_ff_io_stat * stat)2515 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2516 struct nfs4_ff_io_stat *stat)
2517 {
2518 __be32 *p;
2519
2520 p = xdr_reserve_space(xdr, 5 * 8);
2521 p = xdr_encode_hyper(p, stat->ops_requested);
2522 p = xdr_encode_hyper(p, stat->bytes_requested);
2523 p = xdr_encode_hyper(p, stat->ops_completed);
2524 p = xdr_encode_hyper(p, stat->bytes_completed);
2525 p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2526 ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2527 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2528 }
2529
2530 static void
ff_layout_encode_ff_layoutupdate(struct xdr_stream * xdr,const struct nfs42_layoutstat_devinfo * devinfo,struct nfs4_ff_layout_mirror * mirror)2531 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2532 const struct nfs42_layoutstat_devinfo *devinfo,
2533 struct nfs4_ff_layout_mirror *mirror)
2534 {
2535 struct nfs4_pnfs_ds_addr *da;
2536 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2537 struct nfs_fh *fh = &mirror->fh_versions[0];
2538 __be32 *p;
2539
2540 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2541 dprintk("%s: DS %s: encoding address %s\n",
2542 __func__, ds->ds_remotestr, da->da_remotestr);
2543 /* netaddr4 */
2544 ff_layout_encode_netaddr(xdr, da);
2545 /* nfs_fh4 */
2546 p = xdr_reserve_space(xdr, 4 + fh->size);
2547 xdr_encode_opaque(p, fh->data, fh->size);
2548 /* ff_io_latency4 read */
2549 spin_lock(&mirror->lock);
2550 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2551 /* ff_io_latency4 write */
2552 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2553 spin_unlock(&mirror->lock);
2554 /* nfstime4 */
2555 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2556 /* bool */
2557 p = xdr_reserve_space(xdr, 4);
2558 *p = cpu_to_be32(false);
2559 }
2560
2561 static void
ff_layout_encode_layoutstats(struct xdr_stream * xdr,const void * args,const struct nfs4_xdr_opaque_data * opaque)2562 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2563 const struct nfs4_xdr_opaque_data *opaque)
2564 {
2565 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2566 struct nfs42_layoutstat_devinfo, ld_private);
2567 __be32 *start;
2568
2569 /* layoutupdate length */
2570 start = xdr_reserve_space(xdr, 4);
2571 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2572
2573 *start = cpu_to_be32((xdr->p - start - 1) * 4);
2574 }
2575
2576 static void
ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data * opaque)2577 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2578 {
2579 struct nfs4_ff_layout_mirror *mirror = opaque->data;
2580
2581 ff_layout_put_mirror(mirror);
2582 }
2583
2584 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2585 .encode = ff_layout_encode_layoutstats,
2586 .free = ff_layout_free_layoutstats,
2587 };
2588
2589 static int
ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr * lo,struct nfs42_layoutstat_devinfo * devinfo,int dev_limit,enum nfs4_ff_op_type type)2590 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2591 struct nfs42_layoutstat_devinfo *devinfo,
2592 int dev_limit, enum nfs4_ff_op_type type)
2593 {
2594 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2595 struct nfs4_ff_layout_mirror *mirror;
2596 struct nfs4_deviceid_node *dev;
2597 int i = 0;
2598
2599 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2600 if (i >= dev_limit)
2601 break;
2602 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2603 continue;
2604 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL,
2605 &mirror->flags) &&
2606 type != NFS4_FF_OP_LAYOUTRETURN)
2607 continue;
2608 /* mirror refcount put in cleanup_layoutstats */
2609 if (!refcount_inc_not_zero(&mirror->ref))
2610 continue;
2611 dev = &mirror->mirror_ds->id_node;
2612 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2613 devinfo->offset = 0;
2614 devinfo->length = NFS4_MAX_UINT64;
2615 spin_lock(&mirror->lock);
2616 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2617 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2618 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2619 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2620 spin_unlock(&mirror->lock);
2621 devinfo->layout_type = LAYOUT_FLEX_FILES;
2622 devinfo->ld_private.ops = &layoutstat_ops;
2623 devinfo->ld_private.data = mirror;
2624
2625 devinfo++;
2626 i++;
2627 }
2628 return i;
2629 }
2630
ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args * args)2631 static int ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2632 {
2633 struct pnfs_layout_hdr *lo;
2634 struct nfs4_flexfile_layout *ff_layout;
2635 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2636
2637 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2638 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo),
2639 nfs_io_gfp_mask());
2640 if (!args->devinfo)
2641 return -ENOMEM;
2642
2643 spin_lock(&args->inode->i_lock);
2644 lo = NFS_I(args->inode)->layout;
2645 if (lo && pnfs_layout_is_valid(lo)) {
2646 ff_layout = FF_LAYOUT_FROM_HDR(lo);
2647 args->num_dev = ff_layout_mirror_prepare_stats(
2648 &ff_layout->generic_hdr, &args->devinfo[0], dev_count,
2649 NFS4_FF_OP_LAYOUTSTATS);
2650 } else
2651 args->num_dev = 0;
2652 spin_unlock(&args->inode->i_lock);
2653 if (!args->num_dev) {
2654 kfree(args->devinfo);
2655 args->devinfo = NULL;
2656 return -ENOENT;
2657 }
2658
2659 return 0;
2660 }
2661
2662 static int
ff_layout_set_layoutdriver(struct nfs_server * server,const struct nfs_fh * dummy)2663 ff_layout_set_layoutdriver(struct nfs_server *server,
2664 const struct nfs_fh *dummy)
2665 {
2666 #if IS_ENABLED(CONFIG_NFS_V4_2)
2667 server->caps |= NFS_CAP_LAYOUTSTATS | NFS_CAP_REBOOT_LAYOUTRETURN;
2668 #endif
2669 return 0;
2670 }
2671
2672 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2673 .setup_ds_info = ff_layout_setup_ds_info,
2674 .release_ds_info = ff_layout_release_ds_info,
2675 .mark_request_commit = pnfs_layout_mark_request_commit,
2676 .clear_request_commit = pnfs_generic_clear_request_commit,
2677 .scan_commit_lists = pnfs_generic_scan_commit_lists,
2678 .recover_commit_reqs = pnfs_generic_recover_commit_reqs,
2679 .commit_pagelist = ff_layout_commit_pagelist,
2680 };
2681
2682 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2683 .id = LAYOUT_FLEX_FILES,
2684 .name = "LAYOUT_FLEX_FILES",
2685 .owner = THIS_MODULE,
2686 .flags = PNFS_LAYOUTGET_ON_OPEN,
2687 .max_layoutget_response = 4096, /* 1 page or so... */
2688 .set_layoutdriver = ff_layout_set_layoutdriver,
2689 .alloc_layout_hdr = ff_layout_alloc_layout_hdr,
2690 .free_layout_hdr = ff_layout_free_layout_hdr,
2691 .alloc_lseg = ff_layout_alloc_lseg,
2692 .free_lseg = ff_layout_free_lseg,
2693 .add_lseg = ff_layout_add_lseg,
2694 .pg_read_ops = &ff_layout_pg_read_ops,
2695 .pg_write_ops = &ff_layout_pg_write_ops,
2696 .get_ds_info = ff_layout_get_ds_info,
2697 .free_deviceid_node = ff_layout_free_deviceid_node,
2698 .read_pagelist = ff_layout_read_pagelist,
2699 .write_pagelist = ff_layout_write_pagelist,
2700 .alloc_deviceid_node = ff_layout_alloc_deviceid_node,
2701 .prepare_layoutreturn = ff_layout_prepare_layoutreturn,
2702 .sync = pnfs_nfs_generic_sync,
2703 .prepare_layoutstats = ff_layout_prepare_layoutstats,
2704 .cancel_io = ff_layout_cancel_io,
2705 };
2706
nfs4flexfilelayout_init(void)2707 static int __init nfs4flexfilelayout_init(void)
2708 {
2709 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2710 __func__);
2711 return pnfs_register_layoutdriver(&flexfilelayout_type);
2712 }
2713
nfs4flexfilelayout_exit(void)2714 static void __exit nfs4flexfilelayout_exit(void)
2715 {
2716 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2717 __func__);
2718 pnfs_unregister_layoutdriver(&flexfilelayout_type);
2719 }
2720
2721 MODULE_ALIAS("nfs-layouttype4-4");
2722
2723 MODULE_LICENSE("GPL");
2724 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2725
2726 module_init(nfs4flexfilelayout_init);
2727 module_exit(nfs4flexfilelayout_exit);
2728
2729 module_param(io_maxretrans, ushort, 0644);
2730 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
2731 "retries an I/O request before returning an error. ");
2732