fs/nfs/flexfilelayout/flexfilelayout.c

   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 *
  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
  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
  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
  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
 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 > sizeof(struct nfs_fh)) {
 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
 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
 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 *
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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 *
 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_page(&stream, scratch);
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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
 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 *
 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 *
 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 *
 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 *
 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 *
 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
 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
 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
 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
 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
 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
 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 *
 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
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
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 + 1, &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
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
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 */
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
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
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 -ECONNREFUSED:
1244                 case -ECONNRESET:
1245                 case -EHOSTDOWN:
1246                 case -EHOSTUNREACH:
1247                 case -ENETUNREACH:
1248                 case -EADDRINUSE:
1249                 case -ENOBUFS:
1250                 case -EPIPE:
1251                 case -EPERM:
1252                         *op_status = status = NFS4ERR_NXIO;
1253                         break;
1254                 case -EACCES:
1255                         *op_status = status = NFS4ERR_ACCESS;
1256                         break;
1257                 default:
1258                         return;
1259                 }
1260         }
1261
1262         mirror = FF_LAYOUT_COMP(lseg, idx);
1263         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1264                                        mirror, offset, length, status, opnum,
1265                                        GFP_NOIO);
1266
1267         switch (status) {
1268         case NFS4ERR_DELAY:
1269         case NFS4ERR_GRACE:
1270                 break;
1271         case NFS4ERR_NXIO:
1272                 ff_layout_mark_ds_unreachable(lseg, idx);
1273                 /*
1274                  * Don't return the layout if this is a read and we still
1275                  * have layouts to try
1276                  */
1277                 if (opnum == OP_READ)
1278                         break;
1279                 fallthrough;
1280         default:
1281                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1282                                                   lseg);
1283         }
1284
1285         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1286 }
1287
1288 /* NFS_PROTO call done callback routines */
1289 static int ff_layout_read_done_cb(struct rpc_task *task,
1290                                 struct nfs_pgio_header *hdr)
1291 {
1292         int err;
1293
1294         if (task->tk_status < 0) {
1295                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1296                                             hdr->args.offset, hdr->args.count,
1297                                             &hdr->res.op_status, OP_READ,
1298                                             task->tk_status);
1299                 trace_ff_layout_read_error(hdr);
1300         }
1301
1302         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1303                                            hdr->ds_clp, hdr->lseg,
1304                                            hdr->pgio_mirror_idx);
1305
1306         trace_nfs4_pnfs_read(hdr, err);
1307         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1308         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1309         switch (err) {
1310         case -NFS4ERR_RESET_TO_PNFS:
1311                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1312                 return task->tk_status;
1313         case -NFS4ERR_RESET_TO_MDS:
1314                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1315                 return task->tk_status;
1316         case -EAGAIN:
1317                 goto out_eagain;
1318         }
1319
1320         return 0;
1321 out_eagain:
1322         rpc_restart_call_prepare(task);
1323         return -EAGAIN;
1324 }
1325
1326 static bool
1327 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1328 {
1329         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1330 }
1331
1332 /*
1333  * We reference the rpc_cred of the first WRITE that triggers the need for
1334  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1335  * rfc5661 is not clear about which credential should be used.
1336  *
1337  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1338  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1339  * we always send layoutcommit after DS writes.
1340  */
1341 static void
1342 ff_layout_set_layoutcommit(struct inode *inode,
1343                 struct pnfs_layout_segment *lseg,
1344                 loff_t end_offset)
1345 {
1346         if (!ff_layout_need_layoutcommit(lseg))
1347                 return;
1348
1349         pnfs_set_layoutcommit(inode, lseg, end_offset);
1350         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1351                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1352 }
1353
1354 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1355                 struct nfs_pgio_header *hdr)
1356 {
1357         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1358                 return;
1359         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1360                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1361                         hdr->args.count,
1362                         task->tk_start);
1363 }
1364
1365 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1366                 struct nfs_pgio_header *hdr)
1367 {
1368         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1369                 return;
1370         nfs4_ff_layout_stat_io_end_read(task,
1371                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1372                         hdr->args.count,
1373                         hdr->res.count);
1374         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1375 }
1376
1377 static int ff_layout_read_prepare_common(struct rpc_task *task,
1378                                          struct nfs_pgio_header *hdr)
1379 {
1380         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1381                 rpc_exit(task, -EIO);
1382                 return -EIO;
1383         }
1384
1385         ff_layout_read_record_layoutstats_start(task, hdr);
1386         return 0;
1387 }
1388
1389 /*
1390  * Call ops for the async read/write cases
1391  * In the case of dense layouts, the offset needs to be reset to its
1392  * original value.
1393  */
1394 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1395 {
1396         struct nfs_pgio_header *hdr = data;
1397
1398         if (ff_layout_read_prepare_common(task, hdr))
1399                 return;
1400
1401         rpc_call_start(task);
1402 }
1403
1404 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1405 {
1406         struct nfs_pgio_header *hdr = data;
1407
1408         if (nfs4_setup_sequence(hdr->ds_clp,
1409                                 &hdr->args.seq_args,
1410                                 &hdr->res.seq_res,
1411                                 task))
1412                 return;
1413
1414         ff_layout_read_prepare_common(task, hdr);
1415 }
1416
1417 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1418 {
1419         struct nfs_pgio_header *hdr = data;
1420
1421         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1422
1423         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1424             task->tk_status == 0) {
1425                 nfs4_sequence_done(task, &hdr->res.seq_res);
1426                 return;
1427         }
1428
1429         /* Note this may cause RPC to be resent */
1430         hdr->mds_ops->rpc_call_done(task, hdr);
1431 }
1432
1433 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1434 {
1435         struct nfs_pgio_header *hdr = data;
1436
1437         ff_layout_read_record_layoutstats_done(task, hdr);
1438         rpc_count_iostats_metrics(task,
1439             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1440 }
1441
1442 static void ff_layout_read_release(void *data)
1443 {
1444         struct nfs_pgio_header *hdr = data;
1445
1446         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1447         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1448                 ff_layout_resend_pnfs_read(hdr);
1449         else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1450                 ff_layout_reset_read(hdr);
1451         pnfs_generic_rw_release(data);
1452 }
1453
1454
1455 static int ff_layout_write_done_cb(struct rpc_task *task,
1456                                 struct nfs_pgio_header *hdr)
1457 {
1458         loff_t end_offs = 0;
1459         int err;
1460
1461         if (task->tk_status < 0) {
1462                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1463                                             hdr->args.offset, hdr->args.count,
1464                                             &hdr->res.op_status, OP_WRITE,
1465                                             task->tk_status);
1466                 trace_ff_layout_write_error(hdr);
1467         }
1468
1469         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1470                                            hdr->ds_clp, hdr->lseg,
1471                                            hdr->pgio_mirror_idx);
1472
1473         trace_nfs4_pnfs_write(hdr, err);
1474         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1475         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1476         switch (err) {
1477         case -NFS4ERR_RESET_TO_PNFS:
1478                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1479                 return task->tk_status;
1480         case -NFS4ERR_RESET_TO_MDS:
1481                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1482                 return task->tk_status;
1483         case -EAGAIN:
1484                 return -EAGAIN;
1485         }
1486
1487         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1488             hdr->res.verf->committed == NFS_DATA_SYNC)
1489                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1490
1491         /* Note: if the write is unstable, don't set end_offs until commit */
1492         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1493
1494         /* zero out fattr since we don't care DS attr at all */
1495         hdr->fattr.valid = 0;
1496         if (task->tk_status >= 0)
1497                 nfs_writeback_update_inode(hdr);
1498
1499         return 0;
1500 }
1501
1502 static int ff_layout_commit_done_cb(struct rpc_task *task,
1503                                      struct nfs_commit_data *data)
1504 {
1505         int err;
1506
1507         if (task->tk_status < 0) {
1508                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1509                                             data->args.offset, data->args.count,
1510                                             &data->res.op_status, OP_COMMIT,
1511                                             task->tk_status);
1512                 trace_ff_layout_commit_error(data);
1513         }
1514
1515         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1516                                            data->lseg, data->ds_commit_index);
1517
1518         trace_nfs4_pnfs_commit_ds(data, err);
1519         switch (err) {
1520         case -NFS4ERR_RESET_TO_PNFS:
1521                 pnfs_generic_prepare_to_resend_writes(data);
1522                 return -EAGAIN;
1523         case -NFS4ERR_RESET_TO_MDS:
1524                 pnfs_generic_prepare_to_resend_writes(data);
1525                 return -EAGAIN;
1526         case -EAGAIN:
1527                 rpc_restart_call_prepare(task);
1528                 return -EAGAIN;
1529         }
1530
1531         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1532
1533         return 0;
1534 }
1535
1536 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1537                 struct nfs_pgio_header *hdr)
1538 {
1539         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1540                 return;
1541         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1542                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1543                         hdr->args.count,
1544                         task->tk_start);
1545 }
1546
1547 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1548                 struct nfs_pgio_header *hdr)
1549 {
1550         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1551                 return;
1552         nfs4_ff_layout_stat_io_end_write(task,
1553                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1554                         hdr->args.count, hdr->res.count,
1555                         hdr->res.verf->committed);
1556         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1557 }
1558
1559 static int ff_layout_write_prepare_common(struct rpc_task *task,
1560                                           struct nfs_pgio_header *hdr)
1561 {
1562         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1563                 rpc_exit(task, -EIO);
1564                 return -EIO;
1565         }
1566
1567         ff_layout_write_record_layoutstats_start(task, hdr);
1568         return 0;
1569 }
1570
1571 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1572 {
1573         struct nfs_pgio_header *hdr = data;
1574
1575         if (ff_layout_write_prepare_common(task, hdr))
1576                 return;
1577
1578         rpc_call_start(task);
1579 }
1580
1581 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1582 {
1583         struct nfs_pgio_header *hdr = data;
1584
1585         if (nfs4_setup_sequence(hdr->ds_clp,
1586                                 &hdr->args.seq_args,
1587                                 &hdr->res.seq_res,
1588                                 task))
1589                 return;
1590
1591         ff_layout_write_prepare_common(task, hdr);
1592 }
1593
1594 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1595 {
1596         struct nfs_pgio_header *hdr = data;
1597
1598         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1599             task->tk_status == 0) {
1600                 nfs4_sequence_done(task, &hdr->res.seq_res);
1601                 return;
1602         }
1603
1604         /* Note this may cause RPC to be resent */
1605         hdr->mds_ops->rpc_call_done(task, hdr);
1606 }
1607
1608 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1609 {
1610         struct nfs_pgio_header *hdr = data;
1611
1612         ff_layout_write_record_layoutstats_done(task, hdr);
1613         rpc_count_iostats_metrics(task,
1614             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1615 }
1616
1617 static void ff_layout_write_release(void *data)
1618 {
1619         struct nfs_pgio_header *hdr = data;
1620
1621         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1622         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1623                 ff_layout_send_layouterror(hdr->lseg);
1624                 ff_layout_reset_write(hdr, true);
1625         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1626                 ff_layout_reset_write(hdr, false);
1627         pnfs_generic_rw_release(data);
1628 }
1629
1630 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1631                 struct nfs_commit_data *cdata)
1632 {
1633         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1634                 return;
1635         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1636                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1637                         0, task->tk_start);
1638 }
1639
1640 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1641                 struct nfs_commit_data *cdata)
1642 {
1643         struct nfs_page *req;
1644         __u64 count = 0;
1645
1646         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1647                 return;
1648
1649         if (task->tk_status == 0) {
1650                 list_for_each_entry(req, &cdata->pages, wb_list)
1651                         count += req->wb_bytes;
1652         }
1653         nfs4_ff_layout_stat_io_end_write(task,
1654                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1655                         count, count, NFS_FILE_SYNC);
1656         set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1657 }
1658
1659 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1660                 struct nfs_commit_data *cdata)
1661 {
1662         ff_layout_commit_record_layoutstats_start(task, cdata);
1663 }
1664
1665 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1666 {
1667         ff_layout_commit_prepare_common(task, data);
1668         rpc_call_start(task);
1669 }
1670
1671 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1672 {
1673         struct nfs_commit_data *wdata = data;
1674
1675         if (nfs4_setup_sequence(wdata->ds_clp,
1676                                 &wdata->args.seq_args,
1677                                 &wdata->res.seq_res,
1678                                 task))
1679                 return;
1680         ff_layout_commit_prepare_common(task, data);
1681 }
1682
1683 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1684 {
1685         pnfs_generic_write_commit_done(task, data);
1686 }
1687
1688 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1689 {
1690         struct nfs_commit_data *cdata = data;
1691
1692         ff_layout_commit_record_layoutstats_done(task, cdata);
1693         rpc_count_iostats_metrics(task,
1694             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1695 }
1696
1697 static void ff_layout_commit_release(void *data)
1698 {
1699         struct nfs_commit_data *cdata = data;
1700
1701         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1702         pnfs_generic_commit_release(data);
1703 }
1704
1705 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1706         .rpc_call_prepare = ff_layout_read_prepare_v3,
1707         .rpc_call_done = ff_layout_read_call_done,
1708         .rpc_count_stats = ff_layout_read_count_stats,
1709         .rpc_release = ff_layout_read_release,
1710 };
1711
1712 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1713         .rpc_call_prepare = ff_layout_read_prepare_v4,
1714         .rpc_call_done = ff_layout_read_call_done,
1715         .rpc_count_stats = ff_layout_read_count_stats,
1716         .rpc_release = ff_layout_read_release,
1717 };
1718
1719 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1720         .rpc_call_prepare = ff_layout_write_prepare_v3,
1721         .rpc_call_done = ff_layout_write_call_done,
1722         .rpc_count_stats = ff_layout_write_count_stats,
1723         .rpc_release = ff_layout_write_release,
1724 };
1725
1726 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1727         .rpc_call_prepare = ff_layout_write_prepare_v4,
1728         .rpc_call_done = ff_layout_write_call_done,
1729         .rpc_count_stats = ff_layout_write_count_stats,
1730         .rpc_release = ff_layout_write_release,
1731 };
1732
1733 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1734         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1735         .rpc_call_done = ff_layout_commit_done,
1736         .rpc_count_stats = ff_layout_commit_count_stats,
1737         .rpc_release = ff_layout_commit_release,
1738 };
1739
1740 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1741         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1742         .rpc_call_done = ff_layout_commit_done,
1743         .rpc_count_stats = ff_layout_commit_count_stats,
1744         .rpc_release = ff_layout_commit_release,
1745 };
1746
1747 static enum pnfs_try_status
1748 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1749 {
1750         struct pnfs_layout_segment *lseg = hdr->lseg;
1751         struct nfs4_pnfs_ds *ds;
1752         struct rpc_clnt *ds_clnt;
1753         struct nfs4_ff_layout_mirror *mirror;
1754         const struct cred *ds_cred;
1755         loff_t offset = hdr->args.offset;
1756         u32 idx = hdr->pgio_mirror_idx;
1757         int vers;
1758         struct nfs_fh *fh;
1759
1760         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1761                 __func__, hdr->inode->i_ino,
1762                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1763
1764         mirror = FF_LAYOUT_COMP(lseg, idx);
1765         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1766         if (!ds)
1767                 goto out_failed;
1768
1769         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1770                                                    hdr->inode);
1771         if (IS_ERR(ds_clnt))
1772                 goto out_failed;
1773
1774         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1775         if (!ds_cred)
1776                 goto out_failed;
1777
1778         vers = nfs4_ff_layout_ds_version(mirror);
1779
1780         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1781                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1782
1783         hdr->pgio_done_cb = ff_layout_read_done_cb;
1784         refcount_inc(&ds->ds_clp->cl_count);
1785         hdr->ds_clp = ds->ds_clp;
1786         fh = nfs4_ff_layout_select_ds_fh(mirror);
1787         if (fh)
1788                 hdr->args.fh = fh;
1789
1790         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1791
1792         /*
1793          * Note that if we ever decide to split across DSes,
1794          * then we may need to handle dense-like offsets.
1795          */
1796         hdr->args.offset = offset;
1797         hdr->mds_offset = offset;
1798
1799         /* Perform an asynchronous read to ds */
1800         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1801                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1802                                       &ff_layout_read_call_ops_v4,
1803                           0, RPC_TASK_SOFTCONN);
1804         put_cred(ds_cred);
1805         return PNFS_ATTEMPTED;
1806
1807 out_failed:
1808         if (ff_layout_avoid_mds_available_ds(lseg))
1809                 return PNFS_TRY_AGAIN;
1810         trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1811                         hdr->args.offset, hdr->args.count,
1812                         IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1813         return PNFS_NOT_ATTEMPTED;
1814 }
1815
1816 /* Perform async writes. */
1817 static enum pnfs_try_status
1818 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1819 {
1820         struct pnfs_layout_segment *lseg = hdr->lseg;
1821         struct nfs4_pnfs_ds *ds;
1822         struct rpc_clnt *ds_clnt;
1823         struct nfs4_ff_layout_mirror *mirror;
1824         const struct cred *ds_cred;
1825         loff_t offset = hdr->args.offset;
1826         int vers;
1827         struct nfs_fh *fh;
1828         u32 idx = hdr->pgio_mirror_idx;
1829
1830         mirror = FF_LAYOUT_COMP(lseg, idx);
1831         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1832         if (!ds)
1833                 goto out_failed;
1834
1835         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1836                                                    hdr->inode);
1837         if (IS_ERR(ds_clnt))
1838                 goto out_failed;
1839
1840         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1841         if (!ds_cred)
1842                 goto out_failed;
1843
1844         vers = nfs4_ff_layout_ds_version(mirror);
1845
1846         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1847                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1848                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1849                 vers);
1850
1851         hdr->pgio_done_cb = ff_layout_write_done_cb;
1852         refcount_inc(&ds->ds_clp->cl_count);
1853         hdr->ds_clp = ds->ds_clp;
1854         hdr->ds_commit_idx = idx;
1855         fh = nfs4_ff_layout_select_ds_fh(mirror);
1856         if (fh)
1857                 hdr->args.fh = fh;
1858
1859         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1860
1861         /*
1862          * Note that if we ever decide to split across DSes,
1863          * then we may need to handle dense-like offsets.
1864          */
1865         hdr->args.offset = offset;
1866
1867         /* Perform an asynchronous write */
1868         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1869                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1870                                       &ff_layout_write_call_ops_v4,
1871                           sync, RPC_TASK_SOFTCONN);
1872         put_cred(ds_cred);
1873         return PNFS_ATTEMPTED;
1874
1875 out_failed:
1876         if (ff_layout_avoid_mds_available_ds(lseg))
1877                 return PNFS_TRY_AGAIN;
1878         trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1879                         hdr->args.offset, hdr->args.count,
1880                         IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1881         return PNFS_NOT_ATTEMPTED;
1882 }
1883
1884 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1885 {
1886         return i;
1887 }
1888
1889 static struct nfs_fh *
1890 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1891 {
1892         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1893
1894         /* FIXME: Assume that there is only one NFS version available
1895          * for the DS.
1896          */
1897         return &flseg->mirror_array[i]->fh_versions[0];
1898 }
1899
1900 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1901 {
1902         struct pnfs_layout_segment *lseg = data->lseg;
1903         struct nfs4_pnfs_ds *ds;
1904         struct rpc_clnt *ds_clnt;
1905         struct nfs4_ff_layout_mirror *mirror;
1906         const struct cred *ds_cred;
1907         u32 idx;
1908         int vers, ret;
1909         struct nfs_fh *fh;
1910
1911         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1912             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1913                 goto out_err;
1914
1915         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1916         mirror = FF_LAYOUT_COMP(lseg, idx);
1917         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1918         if (!ds)
1919                 goto out_err;
1920
1921         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1922                                                    data->inode);
1923         if (IS_ERR(ds_clnt))
1924                 goto out_err;
1925
1926         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1927         if (!ds_cred)
1928                 goto out_err;
1929
1930         vers = nfs4_ff_layout_ds_version(mirror);
1931
1932         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1933                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1934                 vers);
1935         data->commit_done_cb = ff_layout_commit_done_cb;
1936         data->cred = ds_cred;
1937         refcount_inc(&ds->ds_clp->cl_count);
1938         data->ds_clp = ds->ds_clp;
1939         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1940         if (fh)
1941                 data->args.fh = fh;
1942
1943         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1944                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1945                                                &ff_layout_commit_call_ops_v4,
1946                                    how, RPC_TASK_SOFTCONN);
1947         put_cred(ds_cred);
1948         return ret;
1949 out_err:
1950         pnfs_generic_prepare_to_resend_writes(data);
1951         pnfs_generic_commit_release(data);
1952         return -EAGAIN;
1953 }
1954
1955 static int
1956 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1957                            int how, struct nfs_commit_info *cinfo)
1958 {
1959         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1960                                             ff_layout_initiate_commit);
1961 }
1962
1963 static struct pnfs_ds_commit_info *
1964 ff_layout_get_ds_info(struct inode *inode)
1965 {
1966         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1967
1968         if (layout == NULL)
1969                 return NULL;
1970
1971         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1972 }
1973
1974 static void
1975 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1976                 struct pnfs_layout_segment *lseg)
1977 {
1978         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1979         struct inode *inode = lseg->pls_layout->plh_inode;
1980         struct pnfs_commit_array *array, *new;
1981
1982         new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1983         if (new) {
1984                 spin_lock(&inode->i_lock);
1985                 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1986                 spin_unlock(&inode->i_lock);
1987                 if (array != new)
1988                         pnfs_free_commit_array(new);
1989         }
1990 }
1991
1992 static void
1993 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1994                 struct inode *inode)
1995 {
1996         spin_lock(&inode->i_lock);
1997         pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1998         spin_unlock(&inode->i_lock);
1999 }
2000
2001 static void
2002 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
2003 {
2004         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
2005                                                   id_node));
2006 }
2007
2008 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
2009                                   const struct nfs4_layoutreturn_args *args,
2010                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
2011 {
2012         __be32 *start;
2013
2014         start = xdr_reserve_space(xdr, 4);
2015         if (unlikely(!start))
2016                 return -E2BIG;
2017
2018         *start = cpu_to_be32(ff_args->num_errors);
2019         /* This assume we always return _ALL_ layouts */
2020         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2021 }
2022
2023 static void
2024 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2025 {
2026         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2027 }
2028
2029 static void
2030 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2031                             const nfs4_stateid *stateid,
2032                             const struct nfs42_layoutstat_devinfo *devinfo)
2033 {
2034         __be32 *p;
2035
2036         p = xdr_reserve_space(xdr, 8 + 8);
2037         p = xdr_encode_hyper(p, devinfo->offset);
2038         p = xdr_encode_hyper(p, devinfo->length);
2039         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2040         p = xdr_reserve_space(xdr, 4*8);
2041         p = xdr_encode_hyper(p, devinfo->read_count);
2042         p = xdr_encode_hyper(p, devinfo->read_bytes);
2043         p = xdr_encode_hyper(p, devinfo->write_count);
2044         p = xdr_encode_hyper(p, devinfo->write_bytes);
2045         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2046 }
2047
2048 static void
2049 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2050                             const nfs4_stateid *stateid,
2051                             const struct nfs42_layoutstat_devinfo *devinfo)
2052 {
2053         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2054         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2055                         devinfo->ld_private.data);
2056 }
2057
2058 /* report nothing for now */
2059 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2060                 const struct nfs4_layoutreturn_args *args,
2061                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2062 {
2063         __be32 *p;
2064         int i;
2065
2066         p = xdr_reserve_space(xdr, 4);
2067         *p = cpu_to_be32(ff_args->num_dev);
2068         for (i = 0; i < ff_args->num_dev; i++)
2069                 ff_layout_encode_ff_iostat(xdr,
2070                                 &args->layout->plh_stateid,
2071                                 &ff_args->devinfo[i]);
2072 }
2073
2074 static void
2075 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2076                 unsigned int num_entries)
2077 {
2078         unsigned int i;
2079
2080         for (i = 0; i < num_entries; i++) {
2081                 if (!devinfo[i].ld_private.ops)
2082                         continue;
2083                 if (!devinfo[i].ld_private.ops->free)
2084                         continue;
2085                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2086         }
2087 }
2088
2089 static struct nfs4_deviceid_node *
2090 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2091                               struct pnfs_device *pdev, gfp_t gfp_flags)
2092 {
2093         struct nfs4_ff_layout_ds *dsaddr;
2094
2095         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2096         if (!dsaddr)
2097                 return NULL;
2098         return &dsaddr->id_node;
2099 }
2100
2101 static void
2102 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2103                 const void *voidargs,
2104                 const struct nfs4_xdr_opaque_data *ff_opaque)
2105 {
2106         const struct nfs4_layoutreturn_args *args = voidargs;
2107         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2108         struct xdr_buf tmp_buf = {
2109                 .head = {
2110                         [0] = {
2111                                 .iov_base = page_address(ff_args->pages[0]),
2112                         },
2113                 },
2114                 .buflen = PAGE_SIZE,
2115         };
2116         struct xdr_stream tmp_xdr;
2117         __be32 *start;
2118
2119         dprintk("%s: Begin\n", __func__);
2120
2121         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2122
2123         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2124         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2125
2126         start = xdr_reserve_space(xdr, 4);
2127         *start = cpu_to_be32(tmp_buf.len);
2128         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2129
2130         dprintk("%s: Return\n", __func__);
2131 }
2132
2133 static void
2134 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2135 {
2136         struct nfs4_flexfile_layoutreturn_args *ff_args;
2137
2138         if (!args->data)
2139                 return;
2140         ff_args = args->data;
2141         args->data = NULL;
2142
2143         ff_layout_free_ds_ioerr(&ff_args->errors);
2144         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2145
2146         put_page(ff_args->pages[0]);
2147         kfree(ff_args);
2148 }
2149
2150 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2151         .encode = ff_layout_encode_layoutreturn,
2152         .free = ff_layout_free_layoutreturn,
2153 };
2154
2155 static int
2156 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2157 {
2158         struct nfs4_flexfile_layoutreturn_args *ff_args;
2159         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2160
2161         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2162         if (!ff_args)
2163                 goto out_nomem;
2164         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2165         if (!ff_args->pages[0])
2166                 goto out_nomem_free;
2167
2168         INIT_LIST_HEAD(&ff_args->errors);
2169         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2170                         &args->range, &ff_args->errors,
2171                         FF_LAYOUTRETURN_MAXERR);
2172
2173         spin_lock(&args->inode->i_lock);
2174         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2175                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2176         spin_unlock(&args->inode->i_lock);
2177
2178         args->ld_private->ops = &layoutreturn_ops;
2179         args->ld_private->data = ff_args;
2180         return 0;
2181 out_nomem_free:
2182         kfree(ff_args);
2183 out_nomem:
2184         return -ENOMEM;
2185 }
2186
2187 #ifdef CONFIG_NFS_V4_2
2188 void
2189 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2190 {
2191         struct pnfs_layout_hdr *lo = lseg->pls_layout;
2192         struct nfs42_layout_error *errors;
2193         LIST_HEAD(head);
2194
2195         if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2196                 return;
2197         ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2198         if (list_empty(&head))
2199                 return;
2200
2201         errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2202                         sizeof(*errors), GFP_NOFS);
2203         if (errors != NULL) {
2204                 const struct nfs4_ff_layout_ds_err *pos;
2205                 size_t n = 0;
2206
2207                 list_for_each_entry(pos, &head, list) {
2208                         errors[n].offset = pos->offset;
2209                         errors[n].length = pos->length;
2210                         nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2211                         errors[n].errors[0].dev_id = pos->deviceid;
2212                         errors[n].errors[0].status = pos->status;
2213                         errors[n].errors[0].opnum = pos->opnum;
2214                         n++;
2215                         if (!list_is_last(&pos->list, &head) &&
2216                             n < NFS42_LAYOUTERROR_MAX)
2217                                 continue;
2218                         if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2219                                 break;
2220                         n = 0;
2221                 }
2222                 kfree(errors);
2223         }
2224         ff_layout_free_ds_ioerr(&head);
2225 }
2226 #else
2227 void
2228 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2229 {
2230 }
2231 #endif
2232
2233 static int
2234 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2235 {
2236         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2237
2238         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2239 }
2240
2241 static size_t
2242 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2243                           const int buflen)
2244 {
2245         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2246         const struct in6_addr *addr = &sin6->sin6_addr;
2247
2248         /*
2249          * RFC 4291, Section 2.2.2
2250          *
2251          * Shorthanded ANY address
2252          */
2253         if (ipv6_addr_any(addr))
2254                 return snprintf(buf, buflen, "::");
2255
2256         /*
2257          * RFC 4291, Section 2.2.2
2258          *
2259          * Shorthanded loopback address
2260          */
2261         if (ipv6_addr_loopback(addr))
2262                 return snprintf(buf, buflen, "::1");
2263
2264         /*
2265          * RFC 4291, Section 2.2.3
2266          *
2267          * Special presentation address format for mapped v4
2268          * addresses.
2269          */
2270         if (ipv6_addr_v4mapped(addr))
2271                 return snprintf(buf, buflen, "::ffff:%pI4",
2272                                         &addr->s6_addr32[3]);
2273
2274         /*
2275          * RFC 4291, Section 2.2.1
2276          */
2277         return snprintf(buf, buflen, "%pI6c", addr);
2278 }
2279
2280 /* Derived from rpc_sockaddr2uaddr */
2281 static void
2282 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2283 {
2284         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2285         char portbuf[RPCBIND_MAXUADDRPLEN];
2286         char addrbuf[RPCBIND_MAXUADDRLEN];
2287         char *netid;
2288         unsigned short port;
2289         int len, netid_len;
2290         __be32 *p;
2291
2292         switch (sap->sa_family) {
2293         case AF_INET:
2294                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2295                         return;
2296                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2297                 netid = "tcp";
2298                 netid_len = 3;
2299                 break;
2300         case AF_INET6:
2301                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2302                         return;
2303                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2304                 netid = "tcp6";
2305                 netid_len = 4;
2306                 break;
2307         default:
2308                 /* we only support tcp and tcp6 */
2309                 WARN_ON_ONCE(1);
2310                 return;
2311         }
2312
2313         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2314         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2315
2316         p = xdr_reserve_space(xdr, 4 + netid_len);
2317         xdr_encode_opaque(p, netid, netid_len);
2318
2319         p = xdr_reserve_space(xdr, 4 + len);
2320         xdr_encode_opaque(p, addrbuf, len);
2321 }
2322
2323 static void
2324 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2325                          ktime_t t)
2326 {
2327         struct timespec64 ts;
2328         __be32 *p;
2329
2330         p = xdr_reserve_space(xdr, 12);
2331         ts = ktime_to_timespec64(t);
2332         p = xdr_encode_hyper(p, ts.tv_sec);
2333         *p++ = cpu_to_be32(ts.tv_nsec);
2334 }
2335
2336 static void
2337 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2338                             struct nfs4_ff_io_stat *stat)
2339 {
2340         __be32 *p;
2341
2342         p = xdr_reserve_space(xdr, 5 * 8);
2343         p = xdr_encode_hyper(p, stat->ops_requested);
2344         p = xdr_encode_hyper(p, stat->bytes_requested);
2345         p = xdr_encode_hyper(p, stat->ops_completed);
2346         p = xdr_encode_hyper(p, stat->bytes_completed);
2347         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2348         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2349         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2350 }
2351
2352 static void
2353 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2354                               const struct nfs42_layoutstat_devinfo *devinfo,
2355                               struct nfs4_ff_layout_mirror *mirror)
2356 {
2357         struct nfs4_pnfs_ds_addr *da;
2358         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2359         struct nfs_fh *fh = &mirror->fh_versions[0];
2360         __be32 *p;
2361
2362         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2363         dprintk("%s: DS %s: encoding address %s\n",
2364                 __func__, ds->ds_remotestr, da->da_remotestr);
2365         /* netaddr4 */
2366         ff_layout_encode_netaddr(xdr, da);
2367         /* nfs_fh4 */
2368         p = xdr_reserve_space(xdr, 4 + fh->size);
2369         xdr_encode_opaque(p, fh->data, fh->size);
2370         /* ff_io_latency4 read */
2371         spin_lock(&mirror->lock);
2372         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2373         /* ff_io_latency4 write */
2374         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2375         spin_unlock(&mirror->lock);
2376         /* nfstime4 */
2377         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2378         /* bool */
2379         p = xdr_reserve_space(xdr, 4);
2380         *p = cpu_to_be32(false);
2381 }
2382
2383 static void
2384 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2385                              const struct nfs4_xdr_opaque_data *opaque)
2386 {
2387         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2388                         struct nfs42_layoutstat_devinfo, ld_private);
2389         __be32 *start;
2390
2391         /* layoutupdate length */
2392         start = xdr_reserve_space(xdr, 4);
2393         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2394
2395         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2396 }
2397
2398 static void
2399 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2400 {
2401         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2402
2403         ff_layout_put_mirror(mirror);
2404 }
2405
2406 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2407         .encode = ff_layout_encode_layoutstats,
2408         .free   = ff_layout_free_layoutstats,
2409 };
2410
2411 static int
2412 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2413                                struct nfs42_layoutstat_devinfo *devinfo,
2414                                int dev_limit)
2415 {
2416         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2417         struct nfs4_ff_layout_mirror *mirror;
2418         struct nfs4_deviceid_node *dev;
2419         int i = 0;
2420
2421         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2422                 if (i >= dev_limit)
2423                         break;
2424                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2425                         continue;
2426                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2427                         continue;
2428                 /* mirror refcount put in cleanup_layoutstats */
2429                 if (!refcount_inc_not_zero(&mirror->ref))
2430                         continue;
2431                 dev = &mirror->mirror_ds->id_node;
2432                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2433                 devinfo->offset = 0;
2434                 devinfo->length = NFS4_MAX_UINT64;
2435                 spin_lock(&mirror->lock);
2436                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2437                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2438                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2439                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2440                 spin_unlock(&mirror->lock);
2441                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2442                 devinfo->ld_private.ops = &layoutstat_ops;
2443                 devinfo->ld_private.data = mirror;
2444
2445                 devinfo++;
2446                 i++;
2447         }
2448         return i;
2449 }
2450
2451 static int
2452 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2453 {
2454         struct nfs4_flexfile_layout *ff_layout;
2455         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2456
2457         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2458         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2459         if (!args->devinfo)
2460                 return -ENOMEM;
2461
2462         spin_lock(&args->inode->i_lock);
2463         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2464         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2465                         &args->devinfo[0], dev_count);
2466         spin_unlock(&args->inode->i_lock);
2467         if (!args->num_dev) {
2468                 kfree(args->devinfo);
2469                 args->devinfo = NULL;
2470                 return -ENOENT;
2471         }
2472
2473         return 0;
2474 }
2475
2476 static int
2477 ff_layout_set_layoutdriver(struct nfs_server *server,
2478                 const struct nfs_fh *dummy)
2479 {
2480 #if IS_ENABLED(CONFIG_NFS_V4_2)
2481         server->caps |= NFS_CAP_LAYOUTSTATS;
2482 #endif
2483         return 0;
2484 }
2485
2486 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2487         .setup_ds_info          = ff_layout_setup_ds_info,
2488         .release_ds_info        = ff_layout_release_ds_info,
2489         .mark_request_commit    = pnfs_layout_mark_request_commit,
2490         .clear_request_commit   = pnfs_generic_clear_request_commit,
2491         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2492         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2493         .commit_pagelist        = ff_layout_commit_pagelist,
2494 };
2495
2496 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2497         .id                     = LAYOUT_FLEX_FILES,
2498         .name                   = "LAYOUT_FLEX_FILES",
2499         .owner                  = THIS_MODULE,
2500         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2501         .max_layoutget_response = 4096, /* 1 page or so... */
2502         .set_layoutdriver       = ff_layout_set_layoutdriver,
2503         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2504         .free_layout_hdr        = ff_layout_free_layout_hdr,
2505         .alloc_lseg             = ff_layout_alloc_lseg,
2506         .free_lseg              = ff_layout_free_lseg,
2507         .add_lseg               = ff_layout_add_lseg,
2508         .pg_read_ops            = &ff_layout_pg_read_ops,
2509         .pg_write_ops           = &ff_layout_pg_write_ops,
2510         .get_ds_info            = ff_layout_get_ds_info,
2511         .free_deviceid_node     = ff_layout_free_deviceid_node,
2512         .read_pagelist          = ff_layout_read_pagelist,
2513         .write_pagelist         = ff_layout_write_pagelist,
2514         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2515         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2516         .sync                   = pnfs_nfs_generic_sync,
2517         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2518 };
2519
2520 static int __init nfs4flexfilelayout_init(void)
2521 {
2522         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2523                __func__);
2524         return pnfs_register_layoutdriver(&flexfilelayout_type);
2525 }
2526
2527 static void __exit nfs4flexfilelayout_exit(void)
2528 {
2529         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2530                __func__);
2531         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2532 }
2533
2534 MODULE_ALIAS("nfs-layouttype4-4");
2535
2536 MODULE_LICENSE("GPL");
2537 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2538
2539 module_init(nfs4flexfilelayout_init);
2540 module_exit(nfs4flexfilelayout_exit);
2541
2542 module_param(io_maxretrans, ushort, 0644);
2543 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2544                         "retries an I/O request before returning an error. ");