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_buffer(&stream, page_address(scratch), PAGE_SIZE);
 382
 383         /* stripe unit and mirror_array_cnt */
 384         rc = -EIO;
 385         p = xdr_inline_decode(&stream, 8 + 4);
 386         if (!p)
 387                 goto out_err_free;
 388
 389         p = xdr_decode_hyper(p, &stripe_unit);
 390         mirror_array_cnt = be32_to_cpup(p++);
 391         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 392                 stripe_unit, mirror_array_cnt);
 393
 394         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 395             mirror_array_cnt == 0)
 396                 goto out_err_free;
 397
 398         rc = -ENOMEM;
 399         fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
 400                         gfp_flags);
 401         if (!fls)
 402                 goto out_err_free;
 403
 404         fls->mirror_array_cnt = mirror_array_cnt;
 405         fls->stripe_unit = stripe_unit;
 406
 407         for (i = 0; i < fls->mirror_array_cnt; i++) {
 408                 struct nfs4_ff_layout_mirror *mirror;
 409                 struct cred *kcred;
 410                 const struct cred __rcu *cred;
 411                 kuid_t uid;
 412                 kgid_t gid;
 413                 u32 ds_count, fh_count, id;
 414                 int j;
 415
 416                 rc = -EIO;
 417                 p = xdr_inline_decode(&stream, 4);
 418                 if (!p)
 419                         goto out_err_free;
 420                 ds_count = be32_to_cpup(p);
 421
 422                 /* FIXME: allow for striping? */
 423                 if (ds_count != 1)
 424                         goto out_err_free;
 425
 426                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 427                 if (fls->mirror_array[i] == NULL) {
 428                         rc = -ENOMEM;
 429                         goto out_err_free;
 430                 }
 431
 432                 fls->mirror_array[i]->ds_count = ds_count;
 433
 434                 /* deviceid */
 435                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 436                 if (rc)
 437                         goto out_err_free;
 438
 439                 /* efficiency */
 440                 rc = -EIO;
 441                 p = xdr_inline_decode(&stream, 4);
 442                 if (!p)
 443                         goto out_err_free;
 444                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 445
 446                 /* stateid */
 447                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 448                 if (rc)
 449                         goto out_err_free;
 450
 451                 /* fh */
 452                 rc = -EIO;
 453                 p = xdr_inline_decode(&stream, 4);
 454                 if (!p)
 455                         goto out_err_free;
 456                 fh_count = be32_to_cpup(p);
 457
 458                 fls->mirror_array[i]->fh_versions =
 459                         kcalloc(fh_count, sizeof(struct nfs_fh),
 460                                 gfp_flags);
 461                 if (fls->mirror_array[i]->fh_versions == NULL) {
 462                         rc = -ENOMEM;
 463                         goto out_err_free;
 464                 }
 465
 466                 for (j = 0; j < fh_count; j++) {
 467                         rc = decode_nfs_fh(&stream,
 468                                            &fls->mirror_array[i]->fh_versions[j]);
 469                         if (rc)
 470                                 goto out_err_free;
 471                 }
 472
 473                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 474
 475                 /* user */
 476                 rc = decode_name(&stream, &id);
 477                 if (rc)
 478                         goto out_err_free;
 479
 480                 uid = make_kuid(&init_user_ns, id);
 481
 482                 /* group */
 483                 rc = decode_name(&stream, &id);
 484                 if (rc)
 485                         goto out_err_free;
 486
 487                 gid = make_kgid(&init_user_ns, id);
 488
 489                 if (gfp_flags & __GFP_FS)
 490                         kcred = prepare_kernel_cred(NULL);
 491                 else {
 492                         unsigned int nofs_flags = memalloc_nofs_save();
 493                         kcred = prepare_kernel_cred(NULL);
 494                         memalloc_nofs_restore(nofs_flags);
 495                 }
 496                 rc = -ENOMEM;
 497                 if (!kcred)
 498                         goto out_err_free;
 499                 kcred->fsuid = uid;
 500                 kcred->fsgid = gid;
 501                 cred = RCU_INITIALIZER(kcred);
 502
 503                 if (lgr->range.iomode == IOMODE_READ)
 504                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 505                 else
 506                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 507
 508                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 509                 if (mirror != fls->mirror_array[i]) {
 510                         /* swap cred ptrs so free_mirror will clean up old */
 511                         if (lgr->range.iomode == IOMODE_READ) {
 512                                 cred = xchg(&mirror->ro_cred, cred);
 513                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 514                         } else {
 515                                 cred = xchg(&mirror->rw_cred, cred);
 516                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 517                         }
 518                         ff_layout_free_mirror(fls->mirror_array[i]);
 519                         fls->mirror_array[i] = mirror;
 520                 }
 521
 522                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 523                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 524                         from_kuid(&init_user_ns, uid),
 525                         from_kgid(&init_user_ns, gid));
 526         }
 527
 528         p = xdr_inline_decode(&stream, 4);
 529         if (!p)
 530                 goto out_sort_mirrors;
 531         fls->flags = be32_to_cpup(p);
 532
 533         p = xdr_inline_decode(&stream, 4);
 534         if (!p)
 535                 goto out_sort_mirrors;
 536         for (i=0; i < fls->mirror_array_cnt; i++)
 537                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 538
 539 out_sort_mirrors:
 540         ff_layout_sort_mirrors(fls);
 541         ret = &fls->generic_hdr;
 542         dprintk("<-- %s (success)\n", __func__);
 543 out_free_page:
 544         __free_page(scratch);
 545         return ret;
 546 out_err_free:
 547         _ff_layout_free_lseg(fls);
 548         ret = ERR_PTR(rc);
 549         dprintk("<-- %s (%d)\n", __func__, rc);
 550         goto out_free_page;
 551 }
 552
 553 static void
 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, int 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, int 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                              int start_idx, int *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         int 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                                  int start_idx, int *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                                    int start_idx, int *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                                   int start_idx, int *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, int *best_idx)
 795 {
 796         struct pnfs_layout_segment *lseg = pgio->pg_lseg;
 797         struct nfs4_pnfs_ds *ds;
 798
 799         ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
 800                                                best_idx);
 801         if (ds || !pgio->pg_mirror_idx)
 802                 return ds;
 803         return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
 804 }
 805
 806 static void
 807 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 808                       struct nfs_page *req,
 809                       bool strict_iomode)
 810 {
 811         pnfs_put_lseg(pgio->pg_lseg);
 812         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 813                                            nfs_req_openctx(req),
 814                                            req_offset(req),
 815                                            req->wb_bytes,
 816                                            IOMODE_READ,
 817                                            strict_iomode,
 818                                            GFP_KERNEL);
 819         if (IS_ERR(pgio->pg_lseg)) {
 820                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 821                 pgio->pg_lseg = NULL;
 822         }
 823 }
 824
 825 static void
 826 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
 827                           struct nfs_page *req)
 828 {
 829         pnfs_generic_pg_check_layout(pgio);
 830         pnfs_generic_pg_check_range(pgio, req);
 831 }
 832
 833 static void
 834 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 835                         struct nfs_page *req)
 836 {
 837         struct nfs_pgio_mirror *pgm;
 838         struct nfs4_ff_layout_mirror *mirror;
 839         struct nfs4_pnfs_ds *ds;
 840         int ds_idx;
 841
 842 retry:
 843         ff_layout_pg_check_layout(pgio, req);
 844         /* Use full layout for now */
 845         if (!pgio->pg_lseg) {
 846                 ff_layout_pg_get_read(pgio, req, false);
 847                 if (!pgio->pg_lseg)
 848                         goto out_nolseg;
 849         }
 850         if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 851                 ff_layout_pg_get_read(pgio, req, true);
 852                 if (!pgio->pg_lseg)
 853                         goto out_nolseg;
 854         }
 855
 856         ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
 857         if (!ds) {
 858                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 859                         goto out_mds;
 860                 pnfs_generic_pg_cleanup(pgio);
 861                 /* Sleep for 1 second before retrying */
 862                 ssleep(1);
 863                 goto retry;
 864         }
 865
 866         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 867
 868         pgio->pg_mirror_idx = ds_idx;
 869
 870         /* read always uses only one mirror - idx 0 for pgio layer */
 871         pgm = &pgio->pg_mirrors[0];
 872         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 873
 874         if (NFS_SERVER(pgio->pg_inode)->flags &
 875                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 876                 pgio->pg_maxretrans = io_maxretrans;
 877         return;
 878 out_nolseg:
 879         if (pgio->pg_error < 0)
 880                 return;
 881 out_mds:
 882         trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
 883                         0, NFS4_MAX_UINT64, IOMODE_READ,
 884                         NFS_I(pgio->pg_inode)->layout,
 885                         pgio->pg_lseg);
 886         pgio->pg_maxretrans = 0;
 887         nfs_pageio_reset_read_mds(pgio);
 888 }
 889
 890 static void
 891 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 892                         struct nfs_page *req)
 893 {
 894         struct nfs4_ff_layout_mirror *mirror;
 895         struct nfs_pgio_mirror *pgm;
 896         struct nfs4_pnfs_ds *ds;
 897         int i;
 898
 899 retry:
 900         ff_layout_pg_check_layout(pgio, req);
 901         if (!pgio->pg_lseg) {
 902                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 903                                                    nfs_req_openctx(req),
 904                                                    req_offset(req),
 905                                                    req->wb_bytes,
 906                                                    IOMODE_RW,
 907                                                    false,
 908                                                    GFP_NOFS);
 909                 if (IS_ERR(pgio->pg_lseg)) {
 910                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 911                         pgio->pg_lseg = NULL;
 912                         return;
 913                 }
 914         }
 915         /* If no lseg, fall back to write through mds */
 916         if (pgio->pg_lseg == NULL)
 917                 goto out_mds;
 918
 919         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 920         if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
 921                 goto out_eagain;
 922
 923         for (i = 0; i < pgio->pg_mirror_count; i++) {
 924                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 925                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
 926                 if (!ds) {
 927                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 928                                 goto out_mds;
 929                         pnfs_generic_pg_cleanup(pgio);
 930                         /* Sleep for 1 second before retrying */
 931                         ssleep(1);
 932                         goto retry;
 933                 }
 934                 pgm = &pgio->pg_mirrors[i];
 935                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 936         }
 937
 938         if (NFS_SERVER(pgio->pg_inode)->flags &
 939                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 940                 pgio->pg_maxretrans = io_maxretrans;
 941         return;
 942 out_eagain:
 943         pnfs_generic_pg_cleanup(pgio);
 944         pgio->pg_error = -EAGAIN;
 945         return;
 946 out_mds:
 947         trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
 948                         0, NFS4_MAX_UINT64, IOMODE_RW,
 949                         NFS_I(pgio->pg_inode)->layout,
 950                         pgio->pg_lseg);
 951         pgio->pg_maxretrans = 0;
 952         nfs_pageio_reset_write_mds(pgio);
 953         pgio->pg_error = -EAGAIN;
 954 }
 955
 956 static unsigned int
 957 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 958                                     struct nfs_page *req)
 959 {
 960         if (!pgio->pg_lseg) {
 961                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 962                                                    nfs_req_openctx(req),
 963                                                    req_offset(req),
 964                                                    req->wb_bytes,
 965                                                    IOMODE_RW,
 966                                                    false,
 967                                                    GFP_NOFS);
 968                 if (IS_ERR(pgio->pg_lseg)) {
 969                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 970                         pgio->pg_lseg = NULL;
 971                         goto out;
 972                 }
 973         }
 974         if (pgio->pg_lseg)
 975                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 976
 977         trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
 978                         0, NFS4_MAX_UINT64, IOMODE_RW,
 979                         NFS_I(pgio->pg_inode)->layout,
 980                         pgio->pg_lseg);
 981         /* no lseg means that pnfs is not in use, so no mirroring here */
 982         nfs_pageio_reset_write_mds(pgio);
 983 out:
 984         return 1;
 985 }
 986
 987 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 988         .pg_init = ff_layout_pg_init_read,
 989         .pg_test = pnfs_generic_pg_test,
 990         .pg_doio = pnfs_generic_pg_readpages,
 991         .pg_cleanup = pnfs_generic_pg_cleanup,
 992 };
 993
 994 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 995         .pg_init = ff_layout_pg_init_write,
 996         .pg_test = pnfs_generic_pg_test,
 997         .pg_doio = pnfs_generic_pg_writepages,
 998         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 999         .pg_cleanup = pnfs_generic_pg_cleanup,
1000 };
1001
1002 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1003 {
1004         struct rpc_task *task = &hdr->task;
1005
1006         pnfs_layoutcommit_inode(hdr->inode, false);
1007
1008         if (retry_pnfs) {
1009                 dprintk("%s Reset task %5u for i/o through pNFS "
1010                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1011                         hdr->task.tk_pid,
1012                         hdr->inode->i_sb->s_id,
1013                         (unsigned long long)NFS_FILEID(hdr->inode),
1014                         hdr->args.count,
1015                         (unsigned long long)hdr->args.offset);
1016
1017                 hdr->completion_ops->reschedule_io(hdr);
1018                 return;
1019         }
1020
1021         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1022                 dprintk("%s Reset task %5u for i/o through MDS "
1023                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1024                         hdr->task.tk_pid,
1025                         hdr->inode->i_sb->s_id,
1026                         (unsigned long long)NFS_FILEID(hdr->inode),
1027                         hdr->args.count,
1028                         (unsigned long long)hdr->args.offset);
1029
1030                 trace_pnfs_mds_fallback_write_done(hdr->inode,
1031                                 hdr->args.offset, hdr->args.count,
1032                                 IOMODE_RW, NFS_I(hdr->inode)->layout,
1033                                 hdr->lseg);
1034                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1035         }
1036 }
1037
1038 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1039 {
1040         u32 idx = hdr->pgio_mirror_idx + 1;
1041         int new_idx = 0;
1042
1043         if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx + 1, &new_idx))
1044                 ff_layout_send_layouterror(hdr->lseg);
1045         else
1046                 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1047         pnfs_read_resend_pnfs(hdr, new_idx);
1048 }
1049
1050 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1051 {
1052         struct rpc_task *task = &hdr->task;
1053
1054         pnfs_layoutcommit_inode(hdr->inode, false);
1055         pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1056
1057         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1058                 dprintk("%s Reset task %5u for i/o through MDS "
1059                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1060                         hdr->task.tk_pid,
1061                         hdr->inode->i_sb->s_id,
1062                         (unsigned long long)NFS_FILEID(hdr->inode),
1063                         hdr->args.count,
1064                         (unsigned long long)hdr->args.offset);
1065
1066                 trace_pnfs_mds_fallback_read_done(hdr->inode,
1067                                 hdr->args.offset, hdr->args.count,
1068                                 IOMODE_READ, NFS_I(hdr->inode)->layout,
1069                                 hdr->lseg);
1070                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1071         }
1072 }
1073
1074 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1075                                            struct nfs4_state *state,
1076                                            struct nfs_client *clp,
1077                                            struct pnfs_layout_segment *lseg,
1078                                            int idx)
1079 {
1080         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1081         struct inode *inode = lo->plh_inode;
1082         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1083         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1084
1085         switch (task->tk_status) {
1086         case -NFS4ERR_BADSESSION:
1087         case -NFS4ERR_BADSLOT:
1088         case -NFS4ERR_BAD_HIGH_SLOT:
1089         case -NFS4ERR_DEADSESSION:
1090         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1091         case -NFS4ERR_SEQ_FALSE_RETRY:
1092         case -NFS4ERR_SEQ_MISORDERED:
1093                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1094                         "flags 0x%x\n", __func__, task->tk_status,
1095                         clp->cl_exchange_flags);
1096                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1097                 break;
1098         case -NFS4ERR_DELAY:
1099         case -NFS4ERR_GRACE:
1100                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1101                 break;
1102         case -NFS4ERR_RETRY_UNCACHED_REP:
1103                 break;
1104         /* Invalidate Layout errors */
1105         case -NFS4ERR_PNFS_NO_LAYOUT:
1106         case -ESTALE:           /* mapped NFS4ERR_STALE */
1107         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1108         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1109         case -NFS4ERR_FHEXPIRED:
1110         case -NFS4ERR_WRONG_TYPE:
1111                 dprintk("%s Invalid layout error %d\n", __func__,
1112                         task->tk_status);
1113                 /*
1114                  * Destroy layout so new i/o will get a new layout.
1115                  * Layout will not be destroyed until all current lseg
1116                  * references are put. Mark layout as invalid to resend failed
1117                  * i/o and all i/o waiting on the slot table to the MDS until
1118                  * layout is destroyed and a new valid layout is obtained.
1119                  */
1120                 pnfs_destroy_layout(NFS_I(inode));
1121                 rpc_wake_up(&tbl->slot_tbl_waitq);
1122                 goto reset;
1123         /* RPC connection errors */
1124         case -ECONNREFUSED:
1125         case -EHOSTDOWN:
1126         case -EHOSTUNREACH:
1127         case -ENETUNREACH:
1128         case -EIO:
1129         case -ETIMEDOUT:
1130         case -EPIPE:
1131                 dprintk("%s DS connection error %d\n", __func__,
1132                         task->tk_status);
1133                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1134                                 &devid->deviceid);
1135                 rpc_wake_up(&tbl->slot_tbl_waitq);
1136                 fallthrough;
1137         default:
1138                 if (ff_layout_avoid_mds_available_ds(lseg))
1139                         return -NFS4ERR_RESET_TO_PNFS;
1140 reset:
1141                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1142                         task->tk_status);
1143                 return -NFS4ERR_RESET_TO_MDS;
1144         }
1145         task->tk_status = 0;
1146         return -EAGAIN;
1147 }
1148
1149 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1150 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1151                                            struct pnfs_layout_segment *lseg,
1152                                            int idx)
1153 {
1154         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1155
1156         switch (task->tk_status) {
1157         /* File access problems. Don't mark the device as unavailable */
1158         case -EACCES:
1159         case -ESTALE:
1160         case -EISDIR:
1161         case -EBADHANDLE:
1162         case -ELOOP:
1163         case -ENOSPC:
1164                 break;
1165         case -EJUKEBOX:
1166                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1167                 goto out_retry;
1168         default:
1169                 dprintk("%s DS connection error %d\n", __func__,
1170                         task->tk_status);
1171                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1172                                 &devid->deviceid);
1173         }
1174         /* FIXME: Need to prevent infinite looping here. */
1175         return -NFS4ERR_RESET_TO_PNFS;
1176 out_retry:
1177         task->tk_status = 0;
1178         rpc_restart_call_prepare(task);
1179         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1180         return -EAGAIN;
1181 }
1182
1183 static int ff_layout_async_handle_error(struct rpc_task *task,
1184                                         struct nfs4_state *state,
1185                                         struct nfs_client *clp,
1186                                         struct pnfs_layout_segment *lseg,
1187                                         int idx)
1188 {
1189         int vers = clp->cl_nfs_mod->rpc_vers->number;
1190
1191         if (task->tk_status >= 0) {
1192                 ff_layout_mark_ds_reachable(lseg, idx);
1193                 return 0;
1194         }
1195
1196         /* Handle the case of an invalid layout segment */
1197         if (!pnfs_is_valid_lseg(lseg))
1198                 return -NFS4ERR_RESET_TO_PNFS;
1199
1200         switch (vers) {
1201         case 3:
1202                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1203         case 4:
1204                 return ff_layout_async_handle_error_v4(task, state, clp,
1205                                                        lseg, idx);
1206         default:
1207                 /* should never happen */
1208                 WARN_ON_ONCE(1);
1209                 return 0;
1210         }
1211 }
1212
1213 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1214                                         int idx, u64 offset, u64 length,
1215                                         u32 *op_status, int opnum, int error)
1216 {
1217         struct nfs4_ff_layout_mirror *mirror;
1218         u32 status = *op_status;
1219         int err;
1220
1221         if (status == 0) {
1222                 switch (error) {
1223                 case -ETIMEDOUT:
1224                 case -EPFNOSUPPORT:
1225                 case -EPROTONOSUPPORT:
1226                 case -EOPNOTSUPP:
1227                 case -ECONNREFUSED:
1228                 case -ECONNRESET:
1229                 case -EHOSTDOWN:
1230                 case -EHOSTUNREACH:
1231                 case -ENETUNREACH:
1232                 case -EADDRINUSE:
1233                 case -ENOBUFS:
1234                 case -EPIPE:
1235                 case -EPERM:
1236                         *op_status = status = NFS4ERR_NXIO;
1237                         break;
1238                 case -EACCES:
1239                         *op_status = status = NFS4ERR_ACCESS;
1240                         break;
1241                 default:
1242                         return;
1243                 }
1244         }
1245
1246         mirror = FF_LAYOUT_COMP(lseg, idx);
1247         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1248                                        mirror, offset, length, status, opnum,
1249                                        GFP_NOIO);
1250
1251         switch (status) {
1252         case NFS4ERR_DELAY:
1253         case NFS4ERR_GRACE:
1254                 break;
1255         case NFS4ERR_NXIO:
1256                 ff_layout_mark_ds_unreachable(lseg, idx);
1257                 /*
1258                  * Don't return the layout if this is a read and we still
1259                  * have layouts to try
1260                  */
1261                 if (opnum == OP_READ)
1262                         break;
1263                 fallthrough;
1264         default:
1265                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1266                                                   lseg);
1267         }
1268
1269         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1270 }
1271
1272 /* NFS_PROTO call done callback routines */
1273 static int ff_layout_read_done_cb(struct rpc_task *task,
1274                                 struct nfs_pgio_header *hdr)
1275 {
1276         int err;
1277
1278         if (task->tk_status < 0) {
1279                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1280                                             hdr->args.offset, hdr->args.count,
1281                                             &hdr->res.op_status, OP_READ,
1282                                             task->tk_status);
1283                 trace_ff_layout_read_error(hdr);
1284         }
1285
1286         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1287                                            hdr->ds_clp, hdr->lseg,
1288                                            hdr->pgio_mirror_idx);
1289
1290         trace_nfs4_pnfs_read(hdr, err);
1291         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1292         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1293         switch (err) {
1294         case -NFS4ERR_RESET_TO_PNFS:
1295                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1296                 return task->tk_status;
1297         case -NFS4ERR_RESET_TO_MDS:
1298                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1299                 return task->tk_status;
1300         case -EAGAIN:
1301                 goto out_eagain;
1302         }
1303
1304         return 0;
1305 out_eagain:
1306         rpc_restart_call_prepare(task);
1307         return -EAGAIN;
1308 }
1309
1310 static bool
1311 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1312 {
1313         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1314 }
1315
1316 /*
1317  * We reference the rpc_cred of the first WRITE that triggers the need for
1318  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1319  * rfc5661 is not clear about which credential should be used.
1320  *
1321  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1322  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1323  * we always send layoutcommit after DS writes.
1324  */
1325 static void
1326 ff_layout_set_layoutcommit(struct inode *inode,
1327                 struct pnfs_layout_segment *lseg,
1328                 loff_t end_offset)
1329 {
1330         if (!ff_layout_need_layoutcommit(lseg))
1331                 return;
1332
1333         pnfs_set_layoutcommit(inode, lseg, end_offset);
1334         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1335                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1336 }
1337
1338 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1339                 struct nfs_pgio_header *hdr)
1340 {
1341         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1342                 return;
1343         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1344                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1345                         hdr->args.count,
1346                         task->tk_start);
1347 }
1348
1349 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1350                 struct nfs_pgio_header *hdr)
1351 {
1352         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1353                 return;
1354         nfs4_ff_layout_stat_io_end_read(task,
1355                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1356                         hdr->args.count,
1357                         hdr->res.count);
1358         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1359 }
1360
1361 static int ff_layout_read_prepare_common(struct rpc_task *task,
1362                                          struct nfs_pgio_header *hdr)
1363 {
1364         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1365                 rpc_exit(task, -EIO);
1366                 return -EIO;
1367         }
1368
1369         ff_layout_read_record_layoutstats_start(task, hdr);
1370         return 0;
1371 }
1372
1373 /*
1374  * Call ops for the async read/write cases
1375  * In the case of dense layouts, the offset needs to be reset to its
1376  * original value.
1377  */
1378 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1379 {
1380         struct nfs_pgio_header *hdr = data;
1381
1382         if (ff_layout_read_prepare_common(task, hdr))
1383                 return;
1384
1385         rpc_call_start(task);
1386 }
1387
1388 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1389 {
1390         struct nfs_pgio_header *hdr = data;
1391
1392         if (nfs4_setup_sequence(hdr->ds_clp,
1393                                 &hdr->args.seq_args,
1394                                 &hdr->res.seq_res,
1395                                 task))
1396                 return;
1397
1398         ff_layout_read_prepare_common(task, hdr);
1399 }
1400
1401 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1402 {
1403         struct nfs_pgio_header *hdr = data;
1404
1405         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1406
1407         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1408             task->tk_status == 0) {
1409                 nfs4_sequence_done(task, &hdr->res.seq_res);
1410                 return;
1411         }
1412
1413         /* Note this may cause RPC to be resent */
1414         hdr->mds_ops->rpc_call_done(task, hdr);
1415 }
1416
1417 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1418 {
1419         struct nfs_pgio_header *hdr = data;
1420
1421         ff_layout_read_record_layoutstats_done(task, hdr);
1422         rpc_count_iostats_metrics(task,
1423             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1424 }
1425
1426 static void ff_layout_read_release(void *data)
1427 {
1428         struct nfs_pgio_header *hdr = data;
1429
1430         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1431         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1432                 ff_layout_resend_pnfs_read(hdr);
1433         else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1434                 ff_layout_reset_read(hdr);
1435         pnfs_generic_rw_release(data);
1436 }
1437
1438
1439 static int ff_layout_write_done_cb(struct rpc_task *task,
1440                                 struct nfs_pgio_header *hdr)
1441 {
1442         loff_t end_offs = 0;
1443         int err;
1444
1445         if (task->tk_status < 0) {
1446                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1447                                             hdr->args.offset, hdr->args.count,
1448                                             &hdr->res.op_status, OP_WRITE,
1449                                             task->tk_status);
1450                 trace_ff_layout_write_error(hdr);
1451         }
1452
1453         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1454                                            hdr->ds_clp, hdr->lseg,
1455                                            hdr->pgio_mirror_idx);
1456
1457         trace_nfs4_pnfs_write(hdr, err);
1458         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1459         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1460         switch (err) {
1461         case -NFS4ERR_RESET_TO_PNFS:
1462                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1463                 return task->tk_status;
1464         case -NFS4ERR_RESET_TO_MDS:
1465                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1466                 return task->tk_status;
1467         case -EAGAIN:
1468                 return -EAGAIN;
1469         }
1470
1471         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1472             hdr->res.verf->committed == NFS_DATA_SYNC)
1473                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1474
1475         /* Note: if the write is unstable, don't set end_offs until commit */
1476         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1477
1478         /* zero out fattr since we don't care DS attr at all */
1479         hdr->fattr.valid = 0;
1480         if (task->tk_status >= 0)
1481                 nfs_writeback_update_inode(hdr);
1482
1483         return 0;
1484 }
1485
1486 static int ff_layout_commit_done_cb(struct rpc_task *task,
1487                                      struct nfs_commit_data *data)
1488 {
1489         int err;
1490
1491         if (task->tk_status < 0) {
1492                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1493                                             data->args.offset, data->args.count,
1494                                             &data->res.op_status, OP_COMMIT,
1495                                             task->tk_status);
1496                 trace_ff_layout_commit_error(data);
1497         }
1498
1499         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1500                                            data->lseg, data->ds_commit_index);
1501
1502         trace_nfs4_pnfs_commit_ds(data, err);
1503         switch (err) {
1504         case -NFS4ERR_RESET_TO_PNFS:
1505                 pnfs_generic_prepare_to_resend_writes(data);
1506                 return -EAGAIN;
1507         case -NFS4ERR_RESET_TO_MDS:
1508                 pnfs_generic_prepare_to_resend_writes(data);
1509                 return -EAGAIN;
1510         case -EAGAIN:
1511                 rpc_restart_call_prepare(task);
1512                 return -EAGAIN;
1513         }
1514
1515         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1516
1517         return 0;
1518 }
1519
1520 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1521                 struct nfs_pgio_header *hdr)
1522 {
1523         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1524                 return;
1525         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1526                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1527                         hdr->args.count,
1528                         task->tk_start);
1529 }
1530
1531 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1532                 struct nfs_pgio_header *hdr)
1533 {
1534         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1535                 return;
1536         nfs4_ff_layout_stat_io_end_write(task,
1537                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1538                         hdr->args.count, hdr->res.count,
1539                         hdr->res.verf->committed);
1540         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1541 }
1542
1543 static int ff_layout_write_prepare_common(struct rpc_task *task,
1544                                           struct nfs_pgio_header *hdr)
1545 {
1546         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1547                 rpc_exit(task, -EIO);
1548                 return -EIO;
1549         }
1550
1551         ff_layout_write_record_layoutstats_start(task, hdr);
1552         return 0;
1553 }
1554
1555 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1556 {
1557         struct nfs_pgio_header *hdr = data;
1558
1559         if (ff_layout_write_prepare_common(task, hdr))
1560                 return;
1561
1562         rpc_call_start(task);
1563 }
1564
1565 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1566 {
1567         struct nfs_pgio_header *hdr = data;
1568
1569         if (nfs4_setup_sequence(hdr->ds_clp,
1570                                 &hdr->args.seq_args,
1571                                 &hdr->res.seq_res,
1572                                 task))
1573                 return;
1574
1575         ff_layout_write_prepare_common(task, hdr);
1576 }
1577
1578 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1579 {
1580         struct nfs_pgio_header *hdr = data;
1581
1582         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1583             task->tk_status == 0) {
1584                 nfs4_sequence_done(task, &hdr->res.seq_res);
1585                 return;
1586         }
1587
1588         /* Note this may cause RPC to be resent */
1589         hdr->mds_ops->rpc_call_done(task, hdr);
1590 }
1591
1592 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1593 {
1594         struct nfs_pgio_header *hdr = data;
1595
1596         ff_layout_write_record_layoutstats_done(task, hdr);
1597         rpc_count_iostats_metrics(task,
1598             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1599 }
1600
1601 static void ff_layout_write_release(void *data)
1602 {
1603         struct nfs_pgio_header *hdr = data;
1604
1605         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1606         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1607                 ff_layout_send_layouterror(hdr->lseg);
1608                 ff_layout_reset_write(hdr, true);
1609         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1610                 ff_layout_reset_write(hdr, false);
1611         pnfs_generic_rw_release(data);
1612 }
1613
1614 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1615                 struct nfs_commit_data *cdata)
1616 {
1617         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1618                 return;
1619         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1620                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1621                         0, task->tk_start);
1622 }
1623
1624 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1625                 struct nfs_commit_data *cdata)
1626 {
1627         struct nfs_page *req;
1628         __u64 count = 0;
1629
1630         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1631                 return;
1632
1633         if (task->tk_status == 0) {
1634                 list_for_each_entry(req, &cdata->pages, wb_list)
1635                         count += req->wb_bytes;
1636         }
1637         nfs4_ff_layout_stat_io_end_write(task,
1638                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1639                         count, count, NFS_FILE_SYNC);
1640         set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1641 }
1642
1643 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1644                 struct nfs_commit_data *cdata)
1645 {
1646         ff_layout_commit_record_layoutstats_start(task, cdata);
1647 }
1648
1649 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1650 {
1651         ff_layout_commit_prepare_common(task, data);
1652         rpc_call_start(task);
1653 }
1654
1655 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1656 {
1657         struct nfs_commit_data *wdata = data;
1658
1659         if (nfs4_setup_sequence(wdata->ds_clp,
1660                                 &wdata->args.seq_args,
1661                                 &wdata->res.seq_res,
1662                                 task))
1663                 return;
1664         ff_layout_commit_prepare_common(task, data);
1665 }
1666
1667 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1668 {
1669         pnfs_generic_write_commit_done(task, data);
1670 }
1671
1672 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1673 {
1674         struct nfs_commit_data *cdata = data;
1675
1676         ff_layout_commit_record_layoutstats_done(task, cdata);
1677         rpc_count_iostats_metrics(task,
1678             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1679 }
1680
1681 static void ff_layout_commit_release(void *data)
1682 {
1683         struct nfs_commit_data *cdata = data;
1684
1685         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1686         pnfs_generic_commit_release(data);
1687 }
1688
1689 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1690         .rpc_call_prepare = ff_layout_read_prepare_v3,
1691         .rpc_call_done = ff_layout_read_call_done,
1692         .rpc_count_stats = ff_layout_read_count_stats,
1693         .rpc_release = ff_layout_read_release,
1694 };
1695
1696 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1697         .rpc_call_prepare = ff_layout_read_prepare_v4,
1698         .rpc_call_done = ff_layout_read_call_done,
1699         .rpc_count_stats = ff_layout_read_count_stats,
1700         .rpc_release = ff_layout_read_release,
1701 };
1702
1703 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1704         .rpc_call_prepare = ff_layout_write_prepare_v3,
1705         .rpc_call_done = ff_layout_write_call_done,
1706         .rpc_count_stats = ff_layout_write_count_stats,
1707         .rpc_release = ff_layout_write_release,
1708 };
1709
1710 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1711         .rpc_call_prepare = ff_layout_write_prepare_v4,
1712         .rpc_call_done = ff_layout_write_call_done,
1713         .rpc_count_stats = ff_layout_write_count_stats,
1714         .rpc_release = ff_layout_write_release,
1715 };
1716
1717 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1718         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1719         .rpc_call_done = ff_layout_commit_done,
1720         .rpc_count_stats = ff_layout_commit_count_stats,
1721         .rpc_release = ff_layout_commit_release,
1722 };
1723
1724 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1725         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1726         .rpc_call_done = ff_layout_commit_done,
1727         .rpc_count_stats = ff_layout_commit_count_stats,
1728         .rpc_release = ff_layout_commit_release,
1729 };
1730
1731 static enum pnfs_try_status
1732 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1733 {
1734         struct pnfs_layout_segment *lseg = hdr->lseg;
1735         struct nfs4_pnfs_ds *ds;
1736         struct rpc_clnt *ds_clnt;
1737         struct nfs4_ff_layout_mirror *mirror;
1738         const struct cred *ds_cred;
1739         loff_t offset = hdr->args.offset;
1740         u32 idx = hdr->pgio_mirror_idx;
1741         int vers;
1742         struct nfs_fh *fh;
1743
1744         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1745                 __func__, hdr->inode->i_ino,
1746                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1747
1748         mirror = FF_LAYOUT_COMP(lseg, idx);
1749         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1750         if (!ds)
1751                 goto out_failed;
1752
1753         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1754                                                    hdr->inode);
1755         if (IS_ERR(ds_clnt))
1756                 goto out_failed;
1757
1758         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1759         if (!ds_cred)
1760                 goto out_failed;
1761
1762         vers = nfs4_ff_layout_ds_version(mirror);
1763
1764         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1765                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1766
1767         hdr->pgio_done_cb = ff_layout_read_done_cb;
1768         refcount_inc(&ds->ds_clp->cl_count);
1769         hdr->ds_clp = ds->ds_clp;
1770         fh = nfs4_ff_layout_select_ds_fh(mirror);
1771         if (fh)
1772                 hdr->args.fh = fh;
1773
1774         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1775
1776         /*
1777          * Note that if we ever decide to split across DSes,
1778          * then we may need to handle dense-like offsets.
1779          */
1780         hdr->args.offset = offset;
1781         hdr->mds_offset = offset;
1782
1783         /* Perform an asynchronous read to ds */
1784         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1785                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1786                                       &ff_layout_read_call_ops_v4,
1787                           0, RPC_TASK_SOFTCONN);
1788         put_cred(ds_cred);
1789         return PNFS_ATTEMPTED;
1790
1791 out_failed:
1792         if (ff_layout_avoid_mds_available_ds(lseg))
1793                 return PNFS_TRY_AGAIN;
1794         trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1795                         hdr->args.offset, hdr->args.count,
1796                         IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1797         return PNFS_NOT_ATTEMPTED;
1798 }
1799
1800 /* Perform async writes. */
1801 static enum pnfs_try_status
1802 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1803 {
1804         struct pnfs_layout_segment *lseg = hdr->lseg;
1805         struct nfs4_pnfs_ds *ds;
1806         struct rpc_clnt *ds_clnt;
1807         struct nfs4_ff_layout_mirror *mirror;
1808         const struct cred *ds_cred;
1809         loff_t offset = hdr->args.offset;
1810         int vers;
1811         struct nfs_fh *fh;
1812         int idx = hdr->pgio_mirror_idx;
1813
1814         mirror = FF_LAYOUT_COMP(lseg, idx);
1815         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1816         if (!ds)
1817                 goto out_failed;
1818
1819         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1820                                                    hdr->inode);
1821         if (IS_ERR(ds_clnt))
1822                 goto out_failed;
1823
1824         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1825         if (!ds_cred)
1826                 goto out_failed;
1827
1828         vers = nfs4_ff_layout_ds_version(mirror);
1829
1830         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1831                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1832                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1833                 vers);
1834
1835         hdr->pgio_done_cb = ff_layout_write_done_cb;
1836         refcount_inc(&ds->ds_clp->cl_count);
1837         hdr->ds_clp = ds->ds_clp;
1838         hdr->ds_commit_idx = idx;
1839         fh = nfs4_ff_layout_select_ds_fh(mirror);
1840         if (fh)
1841                 hdr->args.fh = fh;
1842
1843         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1844
1845         /*
1846          * Note that if we ever decide to split across DSes,
1847          * then we may need to handle dense-like offsets.
1848          */
1849         hdr->args.offset = offset;
1850
1851         /* Perform an asynchronous write */
1852         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1853                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1854                                       &ff_layout_write_call_ops_v4,
1855                           sync, RPC_TASK_SOFTCONN);
1856         put_cred(ds_cred);
1857         return PNFS_ATTEMPTED;
1858
1859 out_failed:
1860         if (ff_layout_avoid_mds_available_ds(lseg))
1861                 return PNFS_TRY_AGAIN;
1862         trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1863                         hdr->args.offset, hdr->args.count,
1864                         IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1865         return PNFS_NOT_ATTEMPTED;
1866 }
1867
1868 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1869 {
1870         return i;
1871 }
1872
1873 static struct nfs_fh *
1874 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1875 {
1876         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1877
1878         /* FIXME: Assume that there is only one NFS version available
1879          * for the DS.
1880          */
1881         return &flseg->mirror_array[i]->fh_versions[0];
1882 }
1883
1884 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1885 {
1886         struct pnfs_layout_segment *lseg = data->lseg;
1887         struct nfs4_pnfs_ds *ds;
1888         struct rpc_clnt *ds_clnt;
1889         struct nfs4_ff_layout_mirror *mirror;
1890         const struct cred *ds_cred;
1891         u32 idx;
1892         int vers, ret;
1893         struct nfs_fh *fh;
1894
1895         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1896             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1897                 goto out_err;
1898
1899         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1900         mirror = FF_LAYOUT_COMP(lseg, idx);
1901         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1902         if (!ds)
1903                 goto out_err;
1904
1905         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1906                                                    data->inode);
1907         if (IS_ERR(ds_clnt))
1908                 goto out_err;
1909
1910         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1911         if (!ds_cred)
1912                 goto out_err;
1913
1914         vers = nfs4_ff_layout_ds_version(mirror);
1915
1916         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1917                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1918                 vers);
1919         data->commit_done_cb = ff_layout_commit_done_cb;
1920         data->cred = ds_cred;
1921         refcount_inc(&ds->ds_clp->cl_count);
1922         data->ds_clp = ds->ds_clp;
1923         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1924         if (fh)
1925                 data->args.fh = fh;
1926
1927         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1928                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1929                                                &ff_layout_commit_call_ops_v4,
1930                                    how, RPC_TASK_SOFTCONN);
1931         put_cred(ds_cred);
1932         return ret;
1933 out_err:
1934         pnfs_generic_prepare_to_resend_writes(data);
1935         pnfs_generic_commit_release(data);
1936         return -EAGAIN;
1937 }
1938
1939 static int
1940 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1941                            int how, struct nfs_commit_info *cinfo)
1942 {
1943         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1944                                             ff_layout_initiate_commit);
1945 }
1946
1947 static struct pnfs_ds_commit_info *
1948 ff_layout_get_ds_info(struct inode *inode)
1949 {
1950         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1951
1952         if (layout == NULL)
1953                 return NULL;
1954
1955         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1956 }
1957
1958 static void
1959 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1960                 struct pnfs_layout_segment *lseg)
1961 {
1962         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1963         struct inode *inode = lseg->pls_layout->plh_inode;
1964         struct pnfs_commit_array *array, *new;
1965
1966         new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1967         if (new) {
1968                 spin_lock(&inode->i_lock);
1969                 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1970                 spin_unlock(&inode->i_lock);
1971                 if (array != new)
1972                         pnfs_free_commit_array(new);
1973         }
1974 }
1975
1976 static void
1977 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1978                 struct inode *inode)
1979 {
1980         spin_lock(&inode->i_lock);
1981         pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1982         spin_unlock(&inode->i_lock);
1983 }
1984
1985 static void
1986 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1987 {
1988         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1989                                                   id_node));
1990 }
1991
1992 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1993                                   const struct nfs4_layoutreturn_args *args,
1994                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1995 {
1996         __be32 *start;
1997
1998         start = xdr_reserve_space(xdr, 4);
1999         if (unlikely(!start))
2000                 return -E2BIG;
2001
2002         *start = cpu_to_be32(ff_args->num_errors);
2003         /* This assume we always return _ALL_ layouts */
2004         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2005 }
2006
2007 static void
2008 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2009 {
2010         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2011 }
2012
2013 static void
2014 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2015                             const nfs4_stateid *stateid,
2016                             const struct nfs42_layoutstat_devinfo *devinfo)
2017 {
2018         __be32 *p;
2019
2020         p = xdr_reserve_space(xdr, 8 + 8);
2021         p = xdr_encode_hyper(p, devinfo->offset);
2022         p = xdr_encode_hyper(p, devinfo->length);
2023         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2024         p = xdr_reserve_space(xdr, 4*8);
2025         p = xdr_encode_hyper(p, devinfo->read_count);
2026         p = xdr_encode_hyper(p, devinfo->read_bytes);
2027         p = xdr_encode_hyper(p, devinfo->write_count);
2028         p = xdr_encode_hyper(p, devinfo->write_bytes);
2029         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2030 }
2031
2032 static void
2033 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2034                             const nfs4_stateid *stateid,
2035                             const struct nfs42_layoutstat_devinfo *devinfo)
2036 {
2037         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2038         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2039                         devinfo->ld_private.data);
2040 }
2041
2042 /* report nothing for now */
2043 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2044                 const struct nfs4_layoutreturn_args *args,
2045                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2046 {
2047         __be32 *p;
2048         int i;
2049
2050         p = xdr_reserve_space(xdr, 4);
2051         *p = cpu_to_be32(ff_args->num_dev);
2052         for (i = 0; i < ff_args->num_dev; i++)
2053                 ff_layout_encode_ff_iostat(xdr,
2054                                 &args->layout->plh_stateid,
2055                                 &ff_args->devinfo[i]);
2056 }
2057
2058 static void
2059 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2060                 unsigned int num_entries)
2061 {
2062         unsigned int i;
2063
2064         for (i = 0; i < num_entries; i++) {
2065                 if (!devinfo[i].ld_private.ops)
2066                         continue;
2067                 if (!devinfo[i].ld_private.ops->free)
2068                         continue;
2069                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2070         }
2071 }
2072
2073 static struct nfs4_deviceid_node *
2074 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2075                               struct pnfs_device *pdev, gfp_t gfp_flags)
2076 {
2077         struct nfs4_ff_layout_ds *dsaddr;
2078
2079         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2080         if (!dsaddr)
2081                 return NULL;
2082         return &dsaddr->id_node;
2083 }
2084
2085 static void
2086 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2087                 const void *voidargs,
2088                 const struct nfs4_xdr_opaque_data *ff_opaque)
2089 {
2090         const struct nfs4_layoutreturn_args *args = voidargs;
2091         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2092         struct xdr_buf tmp_buf = {
2093                 .head = {
2094                         [0] = {
2095                                 .iov_base = page_address(ff_args->pages[0]),
2096                         },
2097                 },
2098                 .buflen = PAGE_SIZE,
2099         };
2100         struct xdr_stream tmp_xdr;
2101         __be32 *start;
2102
2103         dprintk("%s: Begin\n", __func__);
2104
2105         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2106
2107         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2108         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2109
2110         start = xdr_reserve_space(xdr, 4);
2111         *start = cpu_to_be32(tmp_buf.len);
2112         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2113
2114         dprintk("%s: Return\n", __func__);
2115 }
2116
2117 static void
2118 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2119 {
2120         struct nfs4_flexfile_layoutreturn_args *ff_args;
2121
2122         if (!args->data)
2123                 return;
2124         ff_args = args->data;
2125         args->data = NULL;
2126
2127         ff_layout_free_ds_ioerr(&ff_args->errors);
2128         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2129
2130         put_page(ff_args->pages[0]);
2131         kfree(ff_args);
2132 }
2133
2134 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2135         .encode = ff_layout_encode_layoutreturn,
2136         .free = ff_layout_free_layoutreturn,
2137 };
2138
2139 static int
2140 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2141 {
2142         struct nfs4_flexfile_layoutreturn_args *ff_args;
2143         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2144
2145         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2146         if (!ff_args)
2147                 goto out_nomem;
2148         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2149         if (!ff_args->pages[0])
2150                 goto out_nomem_free;
2151
2152         INIT_LIST_HEAD(&ff_args->errors);
2153         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2154                         &args->range, &ff_args->errors,
2155                         FF_LAYOUTRETURN_MAXERR);
2156
2157         spin_lock(&args->inode->i_lock);
2158         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2159                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2160         spin_unlock(&args->inode->i_lock);
2161
2162         args->ld_private->ops = &layoutreturn_ops;
2163         args->ld_private->data = ff_args;
2164         return 0;
2165 out_nomem_free:
2166         kfree(ff_args);
2167 out_nomem:
2168         return -ENOMEM;
2169 }
2170
2171 #ifdef CONFIG_NFS_V4_2
2172 void
2173 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2174 {
2175         struct pnfs_layout_hdr *lo = lseg->pls_layout;
2176         struct nfs42_layout_error *errors;
2177         LIST_HEAD(head);
2178
2179         if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2180                 return;
2181         ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2182         if (list_empty(&head))
2183                 return;
2184
2185         errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2186                         sizeof(*errors), GFP_NOFS);
2187         if (errors != NULL) {
2188                 const struct nfs4_ff_layout_ds_err *pos;
2189                 size_t n = 0;
2190
2191                 list_for_each_entry(pos, &head, list) {
2192                         errors[n].offset = pos->offset;
2193                         errors[n].length = pos->length;
2194                         nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2195                         errors[n].errors[0].dev_id = pos->deviceid;
2196                         errors[n].errors[0].status = pos->status;
2197                         errors[n].errors[0].opnum = pos->opnum;
2198                         n++;
2199                         if (!list_is_last(&pos->list, &head) &&
2200                             n < NFS42_LAYOUTERROR_MAX)
2201                                 continue;
2202                         if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2203                                 break;
2204                         n = 0;
2205                 }
2206                 kfree(errors);
2207         }
2208         ff_layout_free_ds_ioerr(&head);
2209 }
2210 #else
2211 void
2212 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2213 {
2214 }
2215 #endif
2216
2217 static int
2218 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2219 {
2220         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2221
2222         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2223 }
2224
2225 static size_t
2226 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2227                           const int buflen)
2228 {
2229         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2230         const struct in6_addr *addr = &sin6->sin6_addr;
2231
2232         /*
2233          * RFC 4291, Section 2.2.2
2234          *
2235          * Shorthanded ANY address
2236          */
2237         if (ipv6_addr_any(addr))
2238                 return snprintf(buf, buflen, "::");
2239
2240         /*
2241          * RFC 4291, Section 2.2.2
2242          *
2243          * Shorthanded loopback address
2244          */
2245         if (ipv6_addr_loopback(addr))
2246                 return snprintf(buf, buflen, "::1");
2247
2248         /*
2249          * RFC 4291, Section 2.2.3
2250          *
2251          * Special presentation address format for mapped v4
2252          * addresses.
2253          */
2254         if (ipv6_addr_v4mapped(addr))
2255                 return snprintf(buf, buflen, "::ffff:%pI4",
2256                                         &addr->s6_addr32[3]);
2257
2258         /*
2259          * RFC 4291, Section 2.2.1
2260          */
2261         return snprintf(buf, buflen, "%pI6c", addr);
2262 }
2263
2264 /* Derived from rpc_sockaddr2uaddr */
2265 static void
2266 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2267 {
2268         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2269         char portbuf[RPCBIND_MAXUADDRPLEN];
2270         char addrbuf[RPCBIND_MAXUADDRLEN];
2271         char *netid;
2272         unsigned short port;
2273         int len, netid_len;
2274         __be32 *p;
2275
2276         switch (sap->sa_family) {
2277         case AF_INET:
2278                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2279                         return;
2280                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2281                 netid = "tcp";
2282                 netid_len = 3;
2283                 break;
2284         case AF_INET6:
2285                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2286                         return;
2287                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2288                 netid = "tcp6";
2289                 netid_len = 4;
2290                 break;
2291         default:
2292                 /* we only support tcp and tcp6 */
2293                 WARN_ON_ONCE(1);
2294                 return;
2295         }
2296
2297         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2298         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2299
2300         p = xdr_reserve_space(xdr, 4 + netid_len);
2301         xdr_encode_opaque(p, netid, netid_len);
2302
2303         p = xdr_reserve_space(xdr, 4 + len);
2304         xdr_encode_opaque(p, addrbuf, len);
2305 }
2306
2307 static void
2308 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2309                          ktime_t t)
2310 {
2311         struct timespec64 ts;
2312         __be32 *p;
2313
2314         p = xdr_reserve_space(xdr, 12);
2315         ts = ktime_to_timespec64(t);
2316         p = xdr_encode_hyper(p, ts.tv_sec);
2317         *p++ = cpu_to_be32(ts.tv_nsec);
2318 }
2319
2320 static void
2321 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2322                             struct nfs4_ff_io_stat *stat)
2323 {
2324         __be32 *p;
2325
2326         p = xdr_reserve_space(xdr, 5 * 8);
2327         p = xdr_encode_hyper(p, stat->ops_requested);
2328         p = xdr_encode_hyper(p, stat->bytes_requested);
2329         p = xdr_encode_hyper(p, stat->ops_completed);
2330         p = xdr_encode_hyper(p, stat->bytes_completed);
2331         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2332         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2333         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2334 }
2335
2336 static void
2337 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2338                               const struct nfs42_layoutstat_devinfo *devinfo,
2339                               struct nfs4_ff_layout_mirror *mirror)
2340 {
2341         struct nfs4_pnfs_ds_addr *da;
2342         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2343         struct nfs_fh *fh = &mirror->fh_versions[0];
2344         __be32 *p;
2345
2346         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2347         dprintk("%s: DS %s: encoding address %s\n",
2348                 __func__, ds->ds_remotestr, da->da_remotestr);
2349         /* netaddr4 */
2350         ff_layout_encode_netaddr(xdr, da);
2351         /* nfs_fh4 */
2352         p = xdr_reserve_space(xdr, 4 + fh->size);
2353         xdr_encode_opaque(p, fh->data, fh->size);
2354         /* ff_io_latency4 read */
2355         spin_lock(&mirror->lock);
2356         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2357         /* ff_io_latency4 write */
2358         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2359         spin_unlock(&mirror->lock);
2360         /* nfstime4 */
2361         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2362         /* bool */
2363         p = xdr_reserve_space(xdr, 4);
2364         *p = cpu_to_be32(false);
2365 }
2366
2367 static void
2368 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2369                              const struct nfs4_xdr_opaque_data *opaque)
2370 {
2371         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2372                         struct nfs42_layoutstat_devinfo, ld_private);
2373         __be32 *start;
2374
2375         /* layoutupdate length */
2376         start = xdr_reserve_space(xdr, 4);
2377         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2378
2379         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2380 }
2381
2382 static void
2383 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2384 {
2385         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2386
2387         ff_layout_put_mirror(mirror);
2388 }
2389
2390 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2391         .encode = ff_layout_encode_layoutstats,
2392         .free   = ff_layout_free_layoutstats,
2393 };
2394
2395 static int
2396 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2397                                struct nfs42_layoutstat_devinfo *devinfo,
2398                                int dev_limit)
2399 {
2400         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2401         struct nfs4_ff_layout_mirror *mirror;
2402         struct nfs4_deviceid_node *dev;
2403         int i = 0;
2404
2405         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2406                 if (i >= dev_limit)
2407                         break;
2408                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2409                         continue;
2410                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2411                         continue;
2412                 /* mirror refcount put in cleanup_layoutstats */
2413                 if (!refcount_inc_not_zero(&mirror->ref))
2414                         continue;
2415                 dev = &mirror->mirror_ds->id_node;
2416                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2417                 devinfo->offset = 0;
2418                 devinfo->length = NFS4_MAX_UINT64;
2419                 spin_lock(&mirror->lock);
2420                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2421                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2422                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2423                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2424                 spin_unlock(&mirror->lock);
2425                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2426                 devinfo->ld_private.ops = &layoutstat_ops;
2427                 devinfo->ld_private.data = mirror;
2428
2429                 devinfo++;
2430                 i++;
2431         }
2432         return i;
2433 }
2434
2435 static int
2436 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2437 {
2438         struct nfs4_flexfile_layout *ff_layout;
2439         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2440
2441         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2442         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2443         if (!args->devinfo)
2444                 return -ENOMEM;
2445
2446         spin_lock(&args->inode->i_lock);
2447         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2448         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2449                         &args->devinfo[0], dev_count);
2450         spin_unlock(&args->inode->i_lock);
2451         if (!args->num_dev) {
2452                 kfree(args->devinfo);
2453                 args->devinfo = NULL;
2454                 return -ENOENT;
2455         }
2456
2457         return 0;
2458 }
2459
2460 static int
2461 ff_layout_set_layoutdriver(struct nfs_server *server,
2462                 const struct nfs_fh *dummy)
2463 {
2464 #if IS_ENABLED(CONFIG_NFS_V4_2)
2465         server->caps |= NFS_CAP_LAYOUTSTATS;
2466 #endif
2467         return 0;
2468 }
2469
2470 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2471         .setup_ds_info          = ff_layout_setup_ds_info,
2472         .release_ds_info        = ff_layout_release_ds_info,
2473         .mark_request_commit    = pnfs_layout_mark_request_commit,
2474         .clear_request_commit   = pnfs_generic_clear_request_commit,
2475         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2476         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2477         .commit_pagelist        = ff_layout_commit_pagelist,
2478 };
2479
2480 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2481         .id                     = LAYOUT_FLEX_FILES,
2482         .name                   = "LAYOUT_FLEX_FILES",
2483         .owner                  = THIS_MODULE,
2484         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2485         .max_layoutget_response = 4096, /* 1 page or so... */
2486         .set_layoutdriver       = ff_layout_set_layoutdriver,
2487         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2488         .free_layout_hdr        = ff_layout_free_layout_hdr,
2489         .alloc_lseg             = ff_layout_alloc_lseg,
2490         .free_lseg              = ff_layout_free_lseg,
2491         .add_lseg               = ff_layout_add_lseg,
2492         .pg_read_ops            = &ff_layout_pg_read_ops,
2493         .pg_write_ops           = &ff_layout_pg_write_ops,
2494         .get_ds_info            = ff_layout_get_ds_info,
2495         .free_deviceid_node     = ff_layout_free_deviceid_node,
2496         .read_pagelist          = ff_layout_read_pagelist,
2497         .write_pagelist         = ff_layout_write_pagelist,
2498         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2499         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2500         .sync                   = pnfs_nfs_generic_sync,
2501         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2502 };
2503
2504 static int __init nfs4flexfilelayout_init(void)
2505 {
2506         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2507                __func__);
2508         return pnfs_register_layoutdriver(&flexfilelayout_type);
2509 }
2510
2511 static void __exit nfs4flexfilelayout_exit(void)
2512 {
2513         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2514                __func__);
2515         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2516 }
2517
2518 MODULE_ALIAS("nfs-layouttype4-4");
2519
2520 MODULE_LICENSE("GPL");
2521 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2522
2523 module_init(nfs4flexfilelayout_init);
2524 module_exit(nfs4flexfilelayout_exit);
2525
2526 module_param(io_maxretrans, ushort, 0644);
2527 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2528                         "retries an I/O request before returning an error. ");