fs/nfs/flexfilelayout/flexfilelayout.c

   1 /*
   2  * Module for pnfs flexfile layout driver.
   3  *
   4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   5  *
   6  * Tao Peng <bergwolf@primarydata.com>
   7  */
   8
   9 #include <linux/nfs_fs.h>
  10 #include <linux/nfs_page.h>
  11 #include <linux/module.h>
  12
  13 #include <linux/sunrpc/metrics.h>
  14
  15 #include "flexfilelayout.h"
  16 #include "../nfs4session.h"
  17 #include "../nfs4idmap.h"
  18 #include "../internal.h"
  19 #include "../delegation.h"
  20 #include "../nfs4trace.h"
  21 #include "../iostat.h"
  22 #include "../nfs.h"
  23 #include "../nfs42.h"
  24
  25 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  26
  27 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  28 #define FF_LAYOUTRETURN_MAXERR 20
  29
  30
  31 static struct group_info        *ff_zero_group;
  32
  33 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  34                 struct nfs_pgio_header *hdr);
  35 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  36                                struct nfs42_layoutstat_devinfo *devinfo,
  37                                int dev_limit);
  38 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  39                               const struct nfs42_layoutstat_devinfo *devinfo,
  40                               struct nfs4_ff_layout_mirror *mirror);
  41
  42 static struct pnfs_layout_hdr *
  43 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  44 {
  45         struct nfs4_flexfile_layout *ffl;
  46
  47         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  48         if (ffl) {
  49                 INIT_LIST_HEAD(&ffl->error_list);
  50                 INIT_LIST_HEAD(&ffl->mirrors);
  51                 ffl->last_report_time = ktime_get();
  52                 return &ffl->generic_hdr;
  53         } else
  54                 return NULL;
  55 }
  56
  57 static void
  58 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  59 {
  60         struct nfs4_ff_layout_ds_err *err, *n;
  61
  62         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  63                                  list) {
  64                 list_del(&err->list);
  65                 kfree(err);
  66         }
  67         kfree(FF_LAYOUT_FROM_HDR(lo));
  68 }
  69
  70 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  71 {
  72         __be32 *p;
  73
  74         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  75         if (unlikely(p == NULL))
  76                 return -ENOBUFS;
  77         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  78         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  79         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  80                 p[0], p[1], p[2], p[3]);
  81         return 0;
  82 }
  83
  84 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  85 {
  86         __be32 *p;
  87
  88         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  89         if (unlikely(!p))
  90                 return -ENOBUFS;
  91         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  92         nfs4_print_deviceid(devid);
  93         return 0;
  94 }
  95
  96 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  97 {
  98         __be32 *p;
  99
 100         p = xdr_inline_decode(xdr, 4);
 101         if (unlikely(!p))
 102                 return -ENOBUFS;
 103         fh->size = be32_to_cpup(p++);
 104         if (fh->size > sizeof(struct nfs_fh)) {
 105                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 106                        fh->size);
 107                 return -EOVERFLOW;
 108         }
 109         /* fh.data */
 110         p = xdr_inline_decode(xdr, fh->size);
 111         if (unlikely(!p))
 112                 return -ENOBUFS;
 113         memcpy(&fh->data, p, fh->size);
 114         dprintk("%s: fh len %d\n", __func__, fh->size);
 115
 116         return 0;
 117 }
 118
 119 /*
 120  * Currently only stringified uids and gids are accepted.
 121  * I.e., kerberos is not supported to the DSes, so no pricipals.
 122  *
 123  * That means that one common function will suffice, but when
 124  * principals are added, this should be split to accomodate
 125  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 126  */
 127 static int
 128 decode_name(struct xdr_stream *xdr, u32 *id)
 129 {
 130         __be32 *p;
 131         int len;
 132
 133         /* opaque_length(4)*/
 134         p = xdr_inline_decode(xdr, 4);
 135         if (unlikely(!p))
 136                 return -ENOBUFS;
 137         len = be32_to_cpup(p++);
 138         if (len < 0)
 139                 return -EINVAL;
 140
 141         dprintk("%s: len %u\n", __func__, len);
 142
 143         /* opaque body */
 144         p = xdr_inline_decode(xdr, len);
 145         if (unlikely(!p))
 146                 return -ENOBUFS;
 147
 148         if (!nfs_map_string_to_numeric((char *)p, len, id))
 149                 return -EINVAL;
 150
 151         return 0;
 152 }
 153
 154 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 155                 const struct nfs4_ff_layout_mirror *m2)
 156 {
 157         int i, j;
 158
 159         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 160                 return false;
 161         for (i = 0; i < m1->fh_versions_cnt; i++) {
 162                 bool found_fh = false;
 163                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 164                         if (nfs_compare_fh(&m1->fh_versions[i],
 165                                         &m2->fh_versions[j]) == 0) {
 166                                 found_fh = true;
 167                                 break;
 168                         }
 169                 }
 170                 if (!found_fh)
 171                         return false;
 172         }
 173         return true;
 174 }
 175
 176 static struct nfs4_ff_layout_mirror *
 177 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 178                 struct nfs4_ff_layout_mirror *mirror)
 179 {
 180         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 181         struct nfs4_ff_layout_mirror *pos;
 182         struct inode *inode = lo->plh_inode;
 183
 184         spin_lock(&inode->i_lock);
 185         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 186                 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
 187                         continue;
 188                 if (!ff_mirror_match_fh(mirror, pos))
 189                         continue;
 190                 if (refcount_inc_not_zero(&pos->ref)) {
 191                         spin_unlock(&inode->i_lock);
 192                         return pos;
 193                 }
 194         }
 195         list_add(&mirror->mirrors, &ff_layout->mirrors);
 196         mirror->layout = lo;
 197         spin_unlock(&inode->i_lock);
 198         return mirror;
 199 }
 200
 201 static void
 202 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 203 {
 204         struct inode *inode;
 205         if (mirror->layout == NULL)
 206                 return;
 207         inode = mirror->layout->plh_inode;
 208         spin_lock(&inode->i_lock);
 209         list_del(&mirror->mirrors);
 210         spin_unlock(&inode->i_lock);
 211         mirror->layout = NULL;
 212 }
 213
 214 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 215 {
 216         struct nfs4_ff_layout_mirror *mirror;
 217
 218         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 219         if (mirror != NULL) {
 220                 spin_lock_init(&mirror->lock);
 221                 refcount_set(&mirror->ref, 1);
 222                 INIT_LIST_HEAD(&mirror->mirrors);
 223         }
 224         return mirror;
 225 }
 226
 227 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 228 {
 229         struct rpc_cred *cred;
 230
 231         ff_layout_remove_mirror(mirror);
 232         kfree(mirror->fh_versions);
 233         cred = rcu_access_pointer(mirror->ro_cred);
 234         if (cred)
 235                 put_rpccred(cred);
 236         cred = rcu_access_pointer(mirror->rw_cred);
 237         if (cred)
 238                 put_rpccred(cred);
 239         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 240         kfree(mirror);
 241 }
 242
 243 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 244 {
 245         if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
 246                 ff_layout_free_mirror(mirror);
 247 }
 248
 249 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 250 {
 251         int i;
 252
 253         if (fls->mirror_array) {
 254                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 255                         /* normally mirror_ds is freed in
 256                          * .free_deviceid_node but we still do it here
 257                          * for .alloc_lseg error path */
 258                         ff_layout_put_mirror(fls->mirror_array[i]);
 259                 }
 260                 kfree(fls->mirror_array);
 261                 fls->mirror_array = NULL;
 262         }
 263 }
 264
 265 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 266 {
 267         int ret = 0;
 268
 269         dprintk("--> %s\n", __func__);
 270
 271         /* FIXME: remove this check when layout segment support is added */
 272         if (lgr->range.offset != 0 ||
 273             lgr->range.length != NFS4_MAX_UINT64) {
 274                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 275                         __func__);
 276                 ret = -EINVAL;
 277         }
 278
 279         dprintk("--> %s returns %d\n", __func__, ret);
 280         return ret;
 281 }
 282
 283 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 284 {
 285         if (fls) {
 286                 ff_layout_free_mirror_array(fls);
 287                 kfree(fls);
 288         }
 289 }
 290
 291 static bool
 292 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 293                 const struct pnfs_layout_range *l2)
 294 {
 295         u64 end1, end2;
 296
 297         if (l1->iomode != l2->iomode)
 298                 return l1->iomode != IOMODE_READ;
 299         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 300         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 301         if (end1 < l2->offset)
 302                 return false;
 303         if (end2 < l1->offset)
 304                 return true;
 305         return l2->offset <= l1->offset;
 306 }
 307
 308 static bool
 309 ff_lseg_merge(struct pnfs_layout_segment *new,
 310                 struct pnfs_layout_segment *old)
 311 {
 312         u64 new_end, old_end;
 313
 314         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 315                 return false;
 316         if (new->pls_range.iomode != old->pls_range.iomode)
 317                 return false;
 318         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 319                         old->pls_range.length);
 320         if (old_end < new->pls_range.offset)
 321                 return false;
 322         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 323                         new->pls_range.length);
 324         if (new_end < old->pls_range.offset)
 325                 return false;
 326
 327         /* Mergeable: copy info from 'old' to 'new' */
 328         if (new_end < old_end)
 329                 new_end = old_end;
 330         if (new->pls_range.offset < old->pls_range.offset)
 331                 new->pls_range.offset = old->pls_range.offset;
 332         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 333                         new_end);
 334         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 335                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 336         return true;
 337 }
 338
 339 static void
 340 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 341                 struct pnfs_layout_segment *lseg,
 342                 struct list_head *free_me)
 343 {
 344         pnfs_generic_layout_insert_lseg(lo, lseg,
 345                         ff_lseg_range_is_after,
 346                         ff_lseg_merge,
 347                         free_me);
 348 }
 349
 350 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 351 {
 352         int i, j;
 353
 354         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 355                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 356                         if (fls->mirror_array[i]->efficiency <
 357                             fls->mirror_array[j]->efficiency)
 358                                 swap(fls->mirror_array[i],
 359                                      fls->mirror_array[j]);
 360         }
 361 }
 362
 363 static struct pnfs_layout_segment *
 364 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 365                      struct nfs4_layoutget_res *lgr,
 366                      gfp_t gfp_flags)
 367 {
 368         struct pnfs_layout_segment *ret;
 369         struct nfs4_ff_layout_segment *fls = NULL;
 370         struct xdr_stream stream;
 371         struct xdr_buf buf;
 372         struct page *scratch;
 373         u64 stripe_unit;
 374         u32 mirror_array_cnt;
 375         __be32 *p;
 376         int i, rc;
 377
 378         dprintk("--> %s\n", __func__);
 379         scratch = alloc_page(gfp_flags);
 380         if (!scratch)
 381                 return ERR_PTR(-ENOMEM);
 382
 383         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 384                               lgr->layoutp->len);
 385         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 386
 387         /* stripe unit and mirror_array_cnt */
 388         rc = -EIO;
 389         p = xdr_inline_decode(&stream, 8 + 4);
 390         if (!p)
 391                 goto out_err_free;
 392
 393         p = xdr_decode_hyper(p, &stripe_unit);
 394         mirror_array_cnt = be32_to_cpup(p++);
 395         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 396                 stripe_unit, mirror_array_cnt);
 397
 398         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 399             mirror_array_cnt == 0)
 400                 goto out_err_free;
 401
 402         rc = -ENOMEM;
 403         fls = kzalloc(sizeof(*fls), gfp_flags);
 404         if (!fls)
 405                 goto out_err_free;
 406
 407         fls->mirror_array_cnt = mirror_array_cnt;
 408         fls->stripe_unit = stripe_unit;
 409         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 410                                     sizeof(fls->mirror_array[0]), gfp_flags);
 411         if (fls->mirror_array == NULL)
 412                 goto out_err_free;
 413
 414         for (i = 0; i < fls->mirror_array_cnt; i++) {
 415                 struct nfs4_ff_layout_mirror *mirror;
 416                 struct auth_cred acred = { .group_info = ff_zero_group };
 417                 struct rpc_cred __rcu *cred;
 418                 u32 ds_count, fh_count, id;
 419                 int j;
 420
 421                 rc = -EIO;
 422                 p = xdr_inline_decode(&stream, 4);
 423                 if (!p)
 424                         goto out_err_free;
 425                 ds_count = be32_to_cpup(p);
 426
 427                 /* FIXME: allow for striping? */
 428                 if (ds_count != 1)
 429                         goto out_err_free;
 430
 431                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 432                 if (fls->mirror_array[i] == NULL) {
 433                         rc = -ENOMEM;
 434                         goto out_err_free;
 435                 }
 436
 437                 fls->mirror_array[i]->ds_count = ds_count;
 438
 439                 /* deviceid */
 440                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 441                 if (rc)
 442                         goto out_err_free;
 443
 444                 /* efficiency */
 445                 rc = -EIO;
 446                 p = xdr_inline_decode(&stream, 4);
 447                 if (!p)
 448                         goto out_err_free;
 449                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 450
 451                 /* stateid */
 452                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 453                 if (rc)
 454                         goto out_err_free;
 455
 456                 /* fh */
 457                 rc = -EIO;
 458                 p = xdr_inline_decode(&stream, 4);
 459                 if (!p)
 460                         goto out_err_free;
 461                 fh_count = be32_to_cpup(p);
 462
 463                 fls->mirror_array[i]->fh_versions =
 464                         kzalloc(fh_count * sizeof(struct nfs_fh),
 465                                 gfp_flags);
 466                 if (fls->mirror_array[i]->fh_versions == NULL) {
 467                         rc = -ENOMEM;
 468                         goto out_err_free;
 469                 }
 470
 471                 for (j = 0; j < fh_count; j++) {
 472                         rc = decode_nfs_fh(&stream,
 473                                            &fls->mirror_array[i]->fh_versions[j]);
 474                         if (rc)
 475                                 goto out_err_free;
 476                 }
 477
 478                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 479
 480                 /* user */
 481                 rc = decode_name(&stream, &id);
 482                 if (rc)
 483                         goto out_err_free;
 484
 485                 acred.uid = make_kuid(&init_user_ns, id);
 486
 487                 /* group */
 488                 rc = decode_name(&stream, &id);
 489                 if (rc)
 490                         goto out_err_free;
 491
 492                 acred.gid = make_kgid(&init_user_ns, id);
 493
 494                 /* find the cred for it */
 495                 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags));
 496                 if (IS_ERR(cred)) {
 497                         rc = PTR_ERR(cred);
 498                         goto out_err_free;
 499                 }
 500
 501                 if (lgr->range.iomode == IOMODE_READ)
 502                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 503                 else
 504                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 505
 506                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 507                 if (mirror != fls->mirror_array[i]) {
 508                         /* swap cred ptrs so free_mirror will clean up old */
 509                         if (lgr->range.iomode == IOMODE_READ) {
 510                                 cred = xchg(&mirror->ro_cred, cred);
 511                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 512                         } else {
 513                                 cred = xchg(&mirror->rw_cred, cred);
 514                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 515                         }
 516                         ff_layout_free_mirror(fls->mirror_array[i]);
 517                         fls->mirror_array[i] = mirror;
 518                 }
 519
 520                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 521                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 522                         from_kuid(&init_user_ns, acred.uid),
 523                         from_kgid(&init_user_ns, acred.gid));
 524         }
 525
 526         p = xdr_inline_decode(&stream, 4);
 527         if (!p)
 528                 goto out_sort_mirrors;
 529         fls->flags = be32_to_cpup(p);
 530
 531         p = xdr_inline_decode(&stream, 4);
 532         if (!p)
 533                 goto out_sort_mirrors;
 534         for (i=0; i < fls->mirror_array_cnt; i++)
 535                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 536
 537 out_sort_mirrors:
 538         ff_layout_sort_mirrors(fls);
 539         rc = ff_layout_check_layout(lgr);
 540         if (rc)
 541                 goto out_err_free;
 542         ret = &fls->generic_hdr;
 543         dprintk("<-- %s (success)\n", __func__);
 544 out_free_page:
 545         __free_page(scratch);
 546         return ret;
 547 out_err_free:
 548         _ff_layout_free_lseg(fls);
 549         ret = ERR_PTR(rc);
 550         dprintk("<-- %s (%d)\n", __func__, rc);
 551         goto out_free_page;
 552 }
 553
 554 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 555 {
 556         struct pnfs_layout_segment *lseg;
 557
 558         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 559                 if (lseg->pls_range.iomode == IOMODE_RW)
 560                         return true;
 561
 562         return false;
 563 }
 564
 565 static void
 566 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 567 {
 568         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 569
 570         dprintk("--> %s\n", __func__);
 571
 572         if (lseg->pls_range.iomode == IOMODE_RW) {
 573                 struct nfs4_flexfile_layout *ffl;
 574                 struct inode *inode;
 575
 576                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 577                 inode = ffl->generic_hdr.plh_inode;
 578                 spin_lock(&inode->i_lock);
 579                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 580                         ffl->commit_info.nbuckets = 0;
 581                         kfree(ffl->commit_info.buckets);
 582                         ffl->commit_info.buckets = NULL;
 583                 }
 584                 spin_unlock(&inode->i_lock);
 585         }
 586         _ff_layout_free_lseg(fls);
 587 }
 588
 589 /* Return 1 until we have multiple lsegs support */
 590 static int
 591 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 592 {
 593         return 1;
 594 }
 595
 596 static void
 597 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 598 {
 599         /* first IO request? */
 600         if (atomic_inc_return(&timer->n_ops) == 1) {
 601                 timer->start_time = now;
 602         }
 603 }
 604
 605 static ktime_t
 606 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 607 {
 608         ktime_t start;
 609
 610         if (atomic_dec_return(&timer->n_ops) < 0)
 611                 WARN_ON_ONCE(1);
 612
 613         start = timer->start_time;
 614         timer->start_time = now;
 615         return ktime_sub(now, start);
 616 }
 617
 618 static bool
 619 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 620                             struct nfs4_ff_layoutstat *layoutstat,
 621                             ktime_t now)
 622 {
 623         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 624         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 625
 626         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 627         if (!mirror->start_time)
 628                 mirror->start_time = now;
 629         if (mirror->report_interval != 0)
 630                 report_interval = (s64)mirror->report_interval * 1000LL;
 631         else if (layoutstats_timer != 0)
 632                 report_interval = (s64)layoutstats_timer * 1000LL;
 633         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 634                         report_interval) {
 635                 ffl->last_report_time = now;
 636                 return true;
 637         }
 638
 639         return false;
 640 }
 641
 642 static void
 643 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 644                 __u64 requested)
 645 {
 646         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 647
 648         iostat->ops_requested++;
 649         iostat->bytes_requested += requested;
 650 }
 651
 652 static void
 653 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 654                 __u64 requested,
 655                 __u64 completed,
 656                 ktime_t time_completed,
 657                 ktime_t time_started)
 658 {
 659         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 660         ktime_t completion_time = ktime_sub(time_completed, time_started);
 661         ktime_t timer;
 662
 663         iostat->ops_completed++;
 664         iostat->bytes_completed += completed;
 665         iostat->bytes_not_delivered += requested - completed;
 666
 667         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 668         iostat->total_busy_time =
 669                         ktime_add(iostat->total_busy_time, timer);
 670         iostat->aggregate_completion_time =
 671                         ktime_add(iostat->aggregate_completion_time,
 672                                         completion_time);
 673 }
 674
 675 static void
 676 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 677                 struct nfs4_ff_layout_mirror *mirror,
 678                 __u64 requested, ktime_t now)
 679 {
 680         bool report;
 681
 682         spin_lock(&mirror->lock);
 683         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 684         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 685         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 686         spin_unlock(&mirror->lock);
 687
 688         if (report)
 689                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 690 }
 691
 692 static void
 693 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 694                 struct nfs4_ff_layout_mirror *mirror,
 695                 __u64 requested,
 696                 __u64 completed)
 697 {
 698         spin_lock(&mirror->lock);
 699         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 700                         requested, completed,
 701                         ktime_get(), task->tk_start);
 702         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 703         spin_unlock(&mirror->lock);
 704 }
 705
 706 static void
 707 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 708                 struct nfs4_ff_layout_mirror *mirror,
 709                 __u64 requested, ktime_t now)
 710 {
 711         bool report;
 712
 713         spin_lock(&mirror->lock);
 714         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 715         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 716         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 717         spin_unlock(&mirror->lock);
 718
 719         if (report)
 720                 pnfs_report_layoutstat(inode, GFP_NOIO);
 721 }
 722
 723 static void
 724 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 725                 struct nfs4_ff_layout_mirror *mirror,
 726                 __u64 requested,
 727                 __u64 completed,
 728                 enum nfs3_stable_how committed)
 729 {
 730         if (committed == NFS_UNSTABLE)
 731                 requested = completed = 0;
 732
 733         spin_lock(&mirror->lock);
 734         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 735                         requested, completed, ktime_get(), task->tk_start);
 736         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 737         spin_unlock(&mirror->lock);
 738 }
 739
 740 static int
 741 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 742                             struct nfs_commit_info *cinfo,
 743                             gfp_t gfp_flags)
 744 {
 745         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 746         struct pnfs_commit_bucket *buckets;
 747         int size;
 748
 749         if (cinfo->ds->nbuckets != 0) {
 750                 /* This assumes there is only one RW lseg per file.
 751                  * To support multiple lseg per file, we need to
 752                  * change struct pnfs_commit_bucket to allow dynamic
 753                  * increasing nbuckets.
 754                  */
 755                 return 0;
 756         }
 757
 758         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 759
 760         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 761                           gfp_flags);
 762         if (!buckets)
 763                 return -ENOMEM;
 764         else {
 765                 int i;
 766
 767                 spin_lock(&cinfo->inode->i_lock);
 768                 if (cinfo->ds->nbuckets != 0)
 769                         kfree(buckets);
 770                 else {
 771                         cinfo->ds->buckets = buckets;
 772                         cinfo->ds->nbuckets = size;
 773                         for (i = 0; i < size; i++) {
 774                                 INIT_LIST_HEAD(&buckets[i].written);
 775                                 INIT_LIST_HEAD(&buckets[i].committing);
 776                                 /* mark direct verifier as unset */
 777                                 buckets[i].direct_verf.committed =
 778                                         NFS_INVALID_STABLE_HOW;
 779                         }
 780                 }
 781                 spin_unlock(&cinfo->inode->i_lock);
 782                 return 0;
 783         }
 784 }
 785
 786 static struct nfs4_pnfs_ds *
 787 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 788                                   int start_idx,
 789                                   int *best_idx)
 790 {
 791         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 792         struct nfs4_pnfs_ds *ds;
 793         bool fail_return = false;
 794         int idx;
 795
 796         /* mirrors are sorted by efficiency */
 797         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 798                 if (idx+1 == fls->mirror_array_cnt)
 799                         fail_return = true;
 800                 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return);
 801                 if (ds) {
 802                         *best_idx = idx;
 803                         return ds;
 804                 }
 805         }
 806
 807         return NULL;
 808 }
 809
 810 static void
 811 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 812                       struct nfs_page *req,
 813                       bool strict_iomode)
 814 {
 815 retry_strict:
 816         pnfs_put_lseg(pgio->pg_lseg);
 817         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 818                                            req->wb_context,
 819                                            0,
 820                                            NFS4_MAX_UINT64,
 821                                            IOMODE_READ,
 822                                            strict_iomode,
 823                                            GFP_KERNEL);
 824         if (IS_ERR(pgio->pg_lseg)) {
 825                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 826                 pgio->pg_lseg = NULL;
 827         }
 828
 829         /* If we don't have checking, do get a IOMODE_RW
 830          * segment, and the server wants to avoid READs
 831          * there, then retry!
 832          */
 833         if (pgio->pg_lseg && !strict_iomode &&
 834             ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 835                 strict_iomode = true;
 836                 goto retry_strict;
 837         }
 838 }
 839
 840 static void
 841 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 842                         struct nfs_page *req)
 843 {
 844         struct nfs_pgio_mirror *pgm;
 845         struct nfs4_ff_layout_mirror *mirror;
 846         struct nfs4_pnfs_ds *ds;
 847         int ds_idx;
 848
 849 retry:
 850         pnfs_generic_pg_check_layout(pgio);
 851         /* Use full layout for now */
 852         if (!pgio->pg_lseg)
 853                 ff_layout_pg_get_read(pgio, req, false);
 854         else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg))
 855                 ff_layout_pg_get_read(pgio, req, true);
 856
 857         /* If no lseg, fall back to read through mds */
 858         if (pgio->pg_lseg == NULL)
 859                 goto out_mds;
 860
 861         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 862         if (!ds) {
 863                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 864                         goto out_mds;
 865                 pnfs_put_lseg(pgio->pg_lseg);
 866                 pgio->pg_lseg = NULL;
 867                 /* Sleep for 1 second before retrying */
 868                 ssleep(1);
 869                 goto retry;
 870         }
 871
 872         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 873
 874         pgio->pg_mirror_idx = ds_idx;
 875
 876         /* read always uses only one mirror - idx 0 for pgio layer */
 877         pgm = &pgio->pg_mirrors[0];
 878         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 879
 880         return;
 881 out_mds:
 882         pnfs_put_lseg(pgio->pg_lseg);
 883         pgio->pg_lseg = NULL;
 884         nfs_pageio_reset_read_mds(pgio);
 885 }
 886
 887 static void
 888 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 889                         struct nfs_page *req)
 890 {
 891         struct nfs4_ff_layout_mirror *mirror;
 892         struct nfs_pgio_mirror *pgm;
 893         struct nfs_commit_info cinfo;
 894         struct nfs4_pnfs_ds *ds;
 895         int i;
 896         int status;
 897
 898 retry:
 899         pnfs_generic_pg_check_layout(pgio);
 900         if (!pgio->pg_lseg) {
 901                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 902                                                    req->wb_context,
 903                                                    0,
 904                                                    NFS4_MAX_UINT64,
 905                                                    IOMODE_RW,
 906                                                    false,
 907                                                    GFP_NOFS);
 908                 if (IS_ERR(pgio->pg_lseg)) {
 909                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 910                         pgio->pg_lseg = NULL;
 911                         return;
 912                 }
 913         }
 914         /* If no lseg, fall back to write through mds */
 915         if (pgio->pg_lseg == NULL)
 916                 goto out_mds;
 917
 918         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 919         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 920         if (status < 0)
 921                 goto out_mds;
 922
 923         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 924         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 925             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 926                 goto out_mds;
 927
 928         for (i = 0; i < pgio->pg_mirror_count; i++) {
 929                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 930                 if (!ds) {
 931                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 932                                 goto out_mds;
 933                         pnfs_put_lseg(pgio->pg_lseg);
 934                         pgio->pg_lseg = NULL;
 935                         /* Sleep for 1 second before retrying */
 936                         ssleep(1);
 937                         goto retry;
 938                 }
 939                 pgm = &pgio->pg_mirrors[i];
 940                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 941                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 942         }
 943
 944         return;
 945
 946 out_mds:
 947         pnfs_put_lseg(pgio->pg_lseg);
 948         pgio->pg_lseg = NULL;
 949         nfs_pageio_reset_write_mds(pgio);
 950 }
 951
 952 static unsigned int
 953 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 954                                     struct nfs_page *req)
 955 {
 956         if (!pgio->pg_lseg) {
 957                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 958                                                    req->wb_context,
 959                                                    0,
 960                                                    NFS4_MAX_UINT64,
 961                                                    IOMODE_RW,
 962                                                    false,
 963                                                    GFP_NOFS);
 964                 if (IS_ERR(pgio->pg_lseg)) {
 965                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 966                         pgio->pg_lseg = NULL;
 967                         goto out;
 968                 }
 969         }
 970         if (pgio->pg_lseg)
 971                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 972
 973         /* no lseg means that pnfs is not in use, so no mirroring here */
 974         nfs_pageio_reset_write_mds(pgio);
 975 out:
 976         return 1;
 977 }
 978
 979 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 980         .pg_init = ff_layout_pg_init_read,
 981         .pg_test = pnfs_generic_pg_test,
 982         .pg_doio = pnfs_generic_pg_readpages,
 983         .pg_cleanup = pnfs_generic_pg_cleanup,
 984 };
 985
 986 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 987         .pg_init = ff_layout_pg_init_write,
 988         .pg_test = pnfs_generic_pg_test,
 989         .pg_doio = pnfs_generic_pg_writepages,
 990         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 991         .pg_cleanup = pnfs_generic_pg_cleanup,
 992 };
 993
 994 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 995 {
 996         struct rpc_task *task = &hdr->task;
 997
 998         pnfs_layoutcommit_inode(hdr->inode, false);
 999
1000         if (retry_pnfs) {
1001                 dprintk("%s Reset task %5u for i/o through pNFS "
1002                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1003                         hdr->task.tk_pid,
1004                         hdr->inode->i_sb->s_id,
1005                         (unsigned long long)NFS_FILEID(hdr->inode),
1006                         hdr->args.count,
1007                         (unsigned long long)hdr->args.offset);
1008
1009                 hdr->completion_ops->reschedule_io(hdr);
1010                 return;
1011         }
1012
1013         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1014                 dprintk("%s Reset task %5u for i/o through MDS "
1015                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1016                         hdr->task.tk_pid,
1017                         hdr->inode->i_sb->s_id,
1018                         (unsigned long long)NFS_FILEID(hdr->inode),
1019                         hdr->args.count,
1020                         (unsigned long long)hdr->args.offset);
1021
1022                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1023         }
1024 }
1025
1026 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1027 {
1028         struct rpc_task *task = &hdr->task;
1029
1030         pnfs_layoutcommit_inode(hdr->inode, false);
1031
1032         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1033                 dprintk("%s Reset task %5u for i/o through MDS "
1034                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1035                         hdr->task.tk_pid,
1036                         hdr->inode->i_sb->s_id,
1037                         (unsigned long long)NFS_FILEID(hdr->inode),
1038                         hdr->args.count,
1039                         (unsigned long long)hdr->args.offset);
1040
1041                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1042         }
1043 }
1044
1045 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1046                                            struct nfs4_state *state,
1047                                            struct nfs_client *clp,
1048                                            struct pnfs_layout_segment *lseg,
1049                                            int idx)
1050 {
1051         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1052         struct inode *inode = lo->plh_inode;
1053         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1054         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1055
1056         switch (task->tk_status) {
1057         case -NFS4ERR_BADSESSION:
1058         case -NFS4ERR_BADSLOT:
1059         case -NFS4ERR_BAD_HIGH_SLOT:
1060         case -NFS4ERR_DEADSESSION:
1061         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1062         case -NFS4ERR_SEQ_FALSE_RETRY:
1063         case -NFS4ERR_SEQ_MISORDERED:
1064                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1065                         "flags 0x%x\n", __func__, task->tk_status,
1066                         clp->cl_exchange_flags);
1067                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1068                 break;
1069         case -NFS4ERR_DELAY:
1070         case -NFS4ERR_GRACE:
1071                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1072                 break;
1073         case -NFS4ERR_RETRY_UNCACHED_REP:
1074                 break;
1075         /* Invalidate Layout errors */
1076         case -NFS4ERR_PNFS_NO_LAYOUT:
1077         case -ESTALE:           /* mapped NFS4ERR_STALE */
1078         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1079         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1080         case -NFS4ERR_FHEXPIRED:
1081         case -NFS4ERR_WRONG_TYPE:
1082                 dprintk("%s Invalid layout error %d\n", __func__,
1083                         task->tk_status);
1084                 /*
1085                  * Destroy layout so new i/o will get a new layout.
1086                  * Layout will not be destroyed until all current lseg
1087                  * references are put. Mark layout as invalid to resend failed
1088                  * i/o and all i/o waiting on the slot table to the MDS until
1089                  * layout is destroyed and a new valid layout is obtained.
1090                  */
1091                 pnfs_destroy_layout(NFS_I(inode));
1092                 rpc_wake_up(&tbl->slot_tbl_waitq);
1093                 goto reset;
1094         /* RPC connection errors */
1095         case -ECONNREFUSED:
1096         case -EHOSTDOWN:
1097         case -EHOSTUNREACH:
1098         case -ENETUNREACH:
1099         case -EIO:
1100         case -ETIMEDOUT:
1101         case -EPIPE:
1102                 dprintk("%s DS connection error %d\n", __func__,
1103                         task->tk_status);
1104                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1105                                 &devid->deviceid);
1106                 rpc_wake_up(&tbl->slot_tbl_waitq);
1107                 /* fall through */
1108         default:
1109                 if (ff_layout_avoid_mds_available_ds(lseg))
1110                         return -NFS4ERR_RESET_TO_PNFS;
1111 reset:
1112                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1113                         task->tk_status);
1114                 return -NFS4ERR_RESET_TO_MDS;
1115         }
1116         task->tk_status = 0;
1117         return -EAGAIN;
1118 }
1119
1120 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1121 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1122                                            struct pnfs_layout_segment *lseg,
1123                                            int idx)
1124 {
1125         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1126
1127         switch (task->tk_status) {
1128         /* File access problems. Don't mark the device as unavailable */
1129         case -EACCES:
1130         case -ESTALE:
1131         case -EISDIR:
1132         case -EBADHANDLE:
1133         case -ELOOP:
1134         case -ENOSPC:
1135                 break;
1136         case -EJUKEBOX:
1137                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1138                 goto out_retry;
1139         default:
1140                 dprintk("%s DS connection error %d\n", __func__,
1141                         task->tk_status);
1142                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1143                                 &devid->deviceid);
1144         }
1145         /* FIXME: Need to prevent infinite looping here. */
1146         return -NFS4ERR_RESET_TO_PNFS;
1147 out_retry:
1148         task->tk_status = 0;
1149         rpc_restart_call_prepare(task);
1150         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1151         return -EAGAIN;
1152 }
1153
1154 static int ff_layout_async_handle_error(struct rpc_task *task,
1155                                         struct nfs4_state *state,
1156                                         struct nfs_client *clp,
1157                                         struct pnfs_layout_segment *lseg,
1158                                         int idx)
1159 {
1160         int vers = clp->cl_nfs_mod->rpc_vers->number;
1161
1162         if (task->tk_status >= 0)
1163                 return 0;
1164
1165         /* Handle the case of an invalid layout segment */
1166         if (!pnfs_is_valid_lseg(lseg))
1167                 return -NFS4ERR_RESET_TO_PNFS;
1168
1169         switch (vers) {
1170         case 3:
1171                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1172         case 4:
1173                 return ff_layout_async_handle_error_v4(task, state, clp,
1174                                                        lseg, idx);
1175         default:
1176                 /* should never happen */
1177                 WARN_ON_ONCE(1);
1178                 return 0;
1179         }
1180 }
1181
1182 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1183                                         int idx, u64 offset, u64 length,
1184                                         u32 status, int opnum, int error)
1185 {
1186         struct nfs4_ff_layout_mirror *mirror;
1187         int err;
1188
1189         if (status == 0) {
1190                 switch (error) {
1191                 case -ETIMEDOUT:
1192                 case -EPFNOSUPPORT:
1193                 case -EPROTONOSUPPORT:
1194                 case -EOPNOTSUPP:
1195                 case -ECONNREFUSED:
1196                 case -ECONNRESET:
1197                 case -EHOSTDOWN:
1198                 case -EHOSTUNREACH:
1199                 case -ENETUNREACH:
1200                 case -EADDRINUSE:
1201                 case -ENOBUFS:
1202                 case -EPIPE:
1203                 case -EPERM:
1204                         status = NFS4ERR_NXIO;
1205                         break;
1206                 case -EACCES:
1207                         status = NFS4ERR_ACCESS;
1208                         break;
1209                 default:
1210                         return;
1211                 }
1212         }
1213
1214         switch (status) {
1215         case NFS4ERR_DELAY:
1216         case NFS4ERR_GRACE:
1217                 return;
1218         default:
1219                 break;
1220         }
1221
1222         mirror = FF_LAYOUT_COMP(lseg, idx);
1223         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1224                                        mirror, offset, length, status, opnum,
1225                                        GFP_NOIO);
1226         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1227         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1228 }
1229
1230 /* NFS_PROTO call done callback routines */
1231 static int ff_layout_read_done_cb(struct rpc_task *task,
1232                                 struct nfs_pgio_header *hdr)
1233 {
1234         int err;
1235
1236         trace_nfs4_pnfs_read(hdr, task->tk_status);
1237         if (task->tk_status < 0)
1238                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1239                                             hdr->args.offset, hdr->args.count,
1240                                             hdr->res.op_status, OP_READ,
1241                                             task->tk_status);
1242         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1243                                            hdr->ds_clp, hdr->lseg,
1244                                            hdr->pgio_mirror_idx);
1245
1246         switch (err) {
1247         case -NFS4ERR_RESET_TO_PNFS:
1248                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1249                                         hdr->pgio_mirror_idx + 1,
1250                                         &hdr->pgio_mirror_idx))
1251                         goto out_eagain;
1252                 ff_layout_read_record_layoutstats_done(task, hdr);
1253                 pnfs_read_resend_pnfs(hdr);
1254                 return task->tk_status;
1255         case -NFS4ERR_RESET_TO_MDS:
1256                 ff_layout_reset_read(hdr);
1257                 return task->tk_status;
1258         case -EAGAIN:
1259                 goto out_eagain;
1260         }
1261
1262         return 0;
1263 out_eagain:
1264         rpc_restart_call_prepare(task);
1265         return -EAGAIN;
1266 }
1267
1268 static bool
1269 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1270 {
1271         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1272 }
1273
1274 /*
1275  * We reference the rpc_cred of the first WRITE that triggers the need for
1276  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1277  * rfc5661 is not clear about which credential should be used.
1278  *
1279  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1280  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1281  * we always send layoutcommit after DS writes.
1282  */
1283 static void
1284 ff_layout_set_layoutcommit(struct inode *inode,
1285                 struct pnfs_layout_segment *lseg,
1286                 loff_t end_offset)
1287 {
1288         if (!ff_layout_need_layoutcommit(lseg))
1289                 return;
1290
1291         pnfs_set_layoutcommit(inode, lseg, end_offset);
1292         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1293                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1294 }
1295
1296 static bool
1297 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx)
1298 {
1299         /* No mirroring for now */
1300         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1301
1302         return ff_layout_test_devid_unavailable(node);
1303 }
1304
1305 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1306                 struct nfs_pgio_header *hdr)
1307 {
1308         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1309                 return;
1310         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1311                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1312                         hdr->args.count,
1313                         task->tk_start);
1314 }
1315
1316 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1317                 struct nfs_pgio_header *hdr)
1318 {
1319         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1320                 return;
1321         nfs4_ff_layout_stat_io_end_read(task,
1322                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1323                         hdr->args.count,
1324                         hdr->res.count);
1325 }
1326
1327 static int ff_layout_read_prepare_common(struct rpc_task *task,
1328                                          struct nfs_pgio_header *hdr)
1329 {
1330         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1331                 rpc_exit(task, -EIO);
1332                 return -EIO;
1333         }
1334         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1335                 rpc_exit(task, -EHOSTDOWN);
1336                 return -EAGAIN;
1337         }
1338
1339         ff_layout_read_record_layoutstats_start(task, hdr);
1340         return 0;
1341 }
1342
1343 /*
1344  * Call ops for the async read/write cases
1345  * In the case of dense layouts, the offset needs to be reset to its
1346  * original value.
1347  */
1348 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1349 {
1350         struct nfs_pgio_header *hdr = data;
1351
1352         if (ff_layout_read_prepare_common(task, hdr))
1353                 return;
1354
1355         rpc_call_start(task);
1356 }
1357
1358 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1359 {
1360         struct nfs_pgio_header *hdr = data;
1361
1362         if (nfs4_setup_sequence(hdr->ds_clp,
1363                                 &hdr->args.seq_args,
1364                                 &hdr->res.seq_res,
1365                                 task))
1366                 return;
1367
1368         if (ff_layout_read_prepare_common(task, hdr))
1369                 return;
1370
1371         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1372                         hdr->args.lock_context, FMODE_READ) == -EIO)
1373                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1374 }
1375
1376 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1377 {
1378         struct nfs_pgio_header *hdr = data;
1379
1380         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1381
1382         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1383             task->tk_status == 0) {
1384                 nfs4_sequence_done(task, &hdr->res.seq_res);
1385                 return;
1386         }
1387
1388         /* Note this may cause RPC to be resent */
1389         hdr->mds_ops->rpc_call_done(task, hdr);
1390 }
1391
1392 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1393 {
1394         struct nfs_pgio_header *hdr = data;
1395
1396         ff_layout_read_record_layoutstats_done(task, hdr);
1397         rpc_count_iostats_metrics(task,
1398             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1399 }
1400
1401 static void ff_layout_read_release(void *data)
1402 {
1403         struct nfs_pgio_header *hdr = data;
1404
1405         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1406         pnfs_generic_rw_release(data);
1407 }
1408
1409
1410 static int ff_layout_write_done_cb(struct rpc_task *task,
1411                                 struct nfs_pgio_header *hdr)
1412 {
1413         loff_t end_offs = 0;
1414         int err;
1415
1416         trace_nfs4_pnfs_write(hdr, task->tk_status);
1417         if (task->tk_status < 0)
1418                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1419                                             hdr->args.offset, hdr->args.count,
1420                                             hdr->res.op_status, OP_WRITE,
1421                                             task->tk_status);
1422         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1423                                            hdr->ds_clp, hdr->lseg,
1424                                            hdr->pgio_mirror_idx);
1425
1426         switch (err) {
1427         case -NFS4ERR_RESET_TO_PNFS:
1428                 ff_layout_reset_write(hdr, true);
1429                 return task->tk_status;
1430         case -NFS4ERR_RESET_TO_MDS:
1431                 ff_layout_reset_write(hdr, false);
1432                 return task->tk_status;
1433         case -EAGAIN:
1434                 return -EAGAIN;
1435         }
1436
1437         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1438             hdr->res.verf->committed == NFS_DATA_SYNC)
1439                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1440
1441         /* Note: if the write is unstable, don't set end_offs until commit */
1442         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1443
1444         /* zero out fattr since we don't care DS attr at all */
1445         hdr->fattr.valid = 0;
1446         if (task->tk_status >= 0)
1447                 nfs_writeback_update_inode(hdr);
1448
1449         return 0;
1450 }
1451
1452 static int ff_layout_commit_done_cb(struct rpc_task *task,
1453                                      struct nfs_commit_data *data)
1454 {
1455         int err;
1456
1457         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1458         if (task->tk_status < 0)
1459                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1460                                             data->args.offset, data->args.count,
1461                                             data->res.op_status, OP_COMMIT,
1462                                             task->tk_status);
1463         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1464                                            data->lseg, data->ds_commit_index);
1465
1466         switch (err) {
1467         case -NFS4ERR_RESET_TO_PNFS:
1468                 pnfs_generic_prepare_to_resend_writes(data);
1469                 return -EAGAIN;
1470         case -NFS4ERR_RESET_TO_MDS:
1471                 pnfs_generic_prepare_to_resend_writes(data);
1472                 return -EAGAIN;
1473         case -EAGAIN:
1474                 rpc_restart_call_prepare(task);
1475                 return -EAGAIN;
1476         }
1477
1478         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1479
1480         return 0;
1481 }
1482
1483 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1484                 struct nfs_pgio_header *hdr)
1485 {
1486         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1487                 return;
1488         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1489                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1490                         hdr->args.count,
1491                         task->tk_start);
1492 }
1493
1494 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1495                 struct nfs_pgio_header *hdr)
1496 {
1497         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1498                 return;
1499         nfs4_ff_layout_stat_io_end_write(task,
1500                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1501                         hdr->args.count, hdr->res.count,
1502                         hdr->res.verf->committed);
1503 }
1504
1505 static int ff_layout_write_prepare_common(struct rpc_task *task,
1506                                           struct nfs_pgio_header *hdr)
1507 {
1508         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1509                 rpc_exit(task, -EIO);
1510                 return -EIO;
1511         }
1512
1513         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1514                 rpc_exit(task, -EHOSTDOWN);
1515                 return -EAGAIN;
1516         }
1517
1518         ff_layout_write_record_layoutstats_start(task, hdr);
1519         return 0;
1520 }
1521
1522 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1523 {
1524         struct nfs_pgio_header *hdr = data;
1525
1526         if (ff_layout_write_prepare_common(task, hdr))
1527                 return;
1528
1529         rpc_call_start(task);
1530 }
1531
1532 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1533 {
1534         struct nfs_pgio_header *hdr = data;
1535
1536         if (nfs4_setup_sequence(hdr->ds_clp,
1537                                 &hdr->args.seq_args,
1538                                 &hdr->res.seq_res,
1539                                 task))
1540                 return;
1541
1542         if (ff_layout_write_prepare_common(task, hdr))
1543                 return;
1544
1545         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1546                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1547                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1548 }
1549
1550 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1551 {
1552         struct nfs_pgio_header *hdr = data;
1553
1554         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1555             task->tk_status == 0) {
1556                 nfs4_sequence_done(task, &hdr->res.seq_res);
1557                 return;
1558         }
1559
1560         /* Note this may cause RPC to be resent */
1561         hdr->mds_ops->rpc_call_done(task, hdr);
1562 }
1563
1564 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1565 {
1566         struct nfs_pgio_header *hdr = data;
1567
1568         ff_layout_write_record_layoutstats_done(task, hdr);
1569         rpc_count_iostats_metrics(task,
1570             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1571 }
1572
1573 static void ff_layout_write_release(void *data)
1574 {
1575         struct nfs_pgio_header *hdr = data;
1576
1577         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1578         pnfs_generic_rw_release(data);
1579 }
1580
1581 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1582                 struct nfs_commit_data *cdata)
1583 {
1584         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1585                 return;
1586         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1587                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1588                         0, task->tk_start);
1589 }
1590
1591 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1592                 struct nfs_commit_data *cdata)
1593 {
1594         struct nfs_page *req;
1595         __u64 count = 0;
1596
1597         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1598                 return;
1599
1600         if (task->tk_status == 0) {
1601                 list_for_each_entry(req, &cdata->pages, wb_list)
1602                         count += req->wb_bytes;
1603         }
1604         nfs4_ff_layout_stat_io_end_write(task,
1605                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1606                         count, count, NFS_FILE_SYNC);
1607 }
1608
1609 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1610                 struct nfs_commit_data *cdata)
1611 {
1612         ff_layout_commit_record_layoutstats_start(task, cdata);
1613 }
1614
1615 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1616 {
1617         ff_layout_commit_prepare_common(task, data);
1618         rpc_call_start(task);
1619 }
1620
1621 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1622 {
1623         struct nfs_commit_data *wdata = data;
1624
1625         if (nfs4_setup_sequence(wdata->ds_clp,
1626                                 &wdata->args.seq_args,
1627                                 &wdata->res.seq_res,
1628                                 task))
1629                 return;
1630         ff_layout_commit_prepare_common(task, data);
1631 }
1632
1633 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1634 {
1635         pnfs_generic_write_commit_done(task, data);
1636 }
1637
1638 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1639 {
1640         struct nfs_commit_data *cdata = data;
1641
1642         ff_layout_commit_record_layoutstats_done(task, cdata);
1643         rpc_count_iostats_metrics(task,
1644             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1645 }
1646
1647 static void ff_layout_commit_release(void *data)
1648 {
1649         struct nfs_commit_data *cdata = data;
1650
1651         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1652         pnfs_generic_commit_release(data);
1653 }
1654
1655 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1656         .rpc_call_prepare = ff_layout_read_prepare_v3,
1657         .rpc_call_done = ff_layout_read_call_done,
1658         .rpc_count_stats = ff_layout_read_count_stats,
1659         .rpc_release = ff_layout_read_release,
1660 };
1661
1662 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1663         .rpc_call_prepare = ff_layout_read_prepare_v4,
1664         .rpc_call_done = ff_layout_read_call_done,
1665         .rpc_count_stats = ff_layout_read_count_stats,
1666         .rpc_release = ff_layout_read_release,
1667 };
1668
1669 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1670         .rpc_call_prepare = ff_layout_write_prepare_v3,
1671         .rpc_call_done = ff_layout_write_call_done,
1672         .rpc_count_stats = ff_layout_write_count_stats,
1673         .rpc_release = ff_layout_write_release,
1674 };
1675
1676 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1677         .rpc_call_prepare = ff_layout_write_prepare_v4,
1678         .rpc_call_done = ff_layout_write_call_done,
1679         .rpc_count_stats = ff_layout_write_count_stats,
1680         .rpc_release = ff_layout_write_release,
1681 };
1682
1683 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1684         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1685         .rpc_call_done = ff_layout_commit_done,
1686         .rpc_count_stats = ff_layout_commit_count_stats,
1687         .rpc_release = ff_layout_commit_release,
1688 };
1689
1690 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1691         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1692         .rpc_call_done = ff_layout_commit_done,
1693         .rpc_count_stats = ff_layout_commit_count_stats,
1694         .rpc_release = ff_layout_commit_release,
1695 };
1696
1697 static enum pnfs_try_status
1698 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1699 {
1700         struct pnfs_layout_segment *lseg = hdr->lseg;
1701         struct nfs4_pnfs_ds *ds;
1702         struct rpc_clnt *ds_clnt;
1703         struct rpc_cred *ds_cred;
1704         loff_t offset = hdr->args.offset;
1705         u32 idx = hdr->pgio_mirror_idx;
1706         int vers;
1707         struct nfs_fh *fh;
1708
1709         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1710                 __func__, hdr->inode->i_ino,
1711                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1712
1713         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1714         if (!ds)
1715                 goto out_failed;
1716
1717         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1718                                                    hdr->inode);
1719         if (IS_ERR(ds_clnt))
1720                 goto out_failed;
1721
1722         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1723         if (!ds_cred)
1724                 goto out_failed;
1725
1726         vers = nfs4_ff_layout_ds_version(lseg, idx);
1727
1728         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1729                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1730
1731         hdr->pgio_done_cb = ff_layout_read_done_cb;
1732         refcount_inc(&ds->ds_clp->cl_count);
1733         hdr->ds_clp = ds->ds_clp;
1734         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1735         if (fh)
1736                 hdr->args.fh = fh;
1737         /*
1738          * Note that if we ever decide to split across DSes,
1739          * then we may need to handle dense-like offsets.
1740          */
1741         hdr->args.offset = offset;
1742         hdr->mds_offset = offset;
1743
1744         /* Perform an asynchronous read to ds */
1745         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1746                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1747                                       &ff_layout_read_call_ops_v4,
1748                           0, RPC_TASK_SOFTCONN);
1749         put_rpccred(ds_cred);
1750         return PNFS_ATTEMPTED;
1751
1752 out_failed:
1753         if (ff_layout_avoid_mds_available_ds(lseg))
1754                 return PNFS_TRY_AGAIN;
1755         return PNFS_NOT_ATTEMPTED;
1756 }
1757
1758 /* Perform async writes. */
1759 static enum pnfs_try_status
1760 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1761 {
1762         struct pnfs_layout_segment *lseg = hdr->lseg;
1763         struct nfs4_pnfs_ds *ds;
1764         struct rpc_clnt *ds_clnt;
1765         struct rpc_cred *ds_cred;
1766         loff_t offset = hdr->args.offset;
1767         int vers;
1768         struct nfs_fh *fh;
1769         int idx = hdr->pgio_mirror_idx;
1770
1771         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1772         if (!ds)
1773                 goto out_failed;
1774
1775         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1776                                                    hdr->inode);
1777         if (IS_ERR(ds_clnt))
1778                 goto out_failed;
1779
1780         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1781         if (!ds_cred)
1782                 goto out_failed;
1783
1784         vers = nfs4_ff_layout_ds_version(lseg, idx);
1785
1786         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1787                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1788                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1789                 vers);
1790
1791         hdr->pgio_done_cb = ff_layout_write_done_cb;
1792         refcount_inc(&ds->ds_clp->cl_count);
1793         hdr->ds_clp = ds->ds_clp;
1794         hdr->ds_commit_idx = idx;
1795         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1796         if (fh)
1797                 hdr->args.fh = fh;
1798
1799         /*
1800          * Note that if we ever decide to split across DSes,
1801          * then we may need to handle dense-like offsets.
1802          */
1803         hdr->args.offset = offset;
1804
1805         /* Perform an asynchronous write */
1806         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1807                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1808                                       &ff_layout_write_call_ops_v4,
1809                           sync, RPC_TASK_SOFTCONN);
1810         put_rpccred(ds_cred);
1811         return PNFS_ATTEMPTED;
1812
1813 out_failed:
1814         if (ff_layout_avoid_mds_available_ds(lseg))
1815                 return PNFS_TRY_AGAIN;
1816         return PNFS_NOT_ATTEMPTED;
1817 }
1818
1819 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1820 {
1821         return i;
1822 }
1823
1824 static struct nfs_fh *
1825 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1826 {
1827         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1828
1829         /* FIXME: Assume that there is only one NFS version available
1830          * for the DS.
1831          */
1832         return &flseg->mirror_array[i]->fh_versions[0];
1833 }
1834
1835 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1836 {
1837         struct pnfs_layout_segment *lseg = data->lseg;
1838         struct nfs4_pnfs_ds *ds;
1839         struct rpc_clnt *ds_clnt;
1840         struct rpc_cred *ds_cred;
1841         u32 idx;
1842         int vers, ret;
1843         struct nfs_fh *fh;
1844
1845         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1846             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1847                 goto out_err;
1848
1849         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1850         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1851         if (!ds)
1852                 goto out_err;
1853
1854         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1855                                                    data->inode);
1856         if (IS_ERR(ds_clnt))
1857                 goto out_err;
1858
1859         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1860         if (!ds_cred)
1861                 goto out_err;
1862
1863         vers = nfs4_ff_layout_ds_version(lseg, idx);
1864
1865         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1866                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1867                 vers);
1868         data->commit_done_cb = ff_layout_commit_done_cb;
1869         data->cred = ds_cred;
1870         refcount_inc(&ds->ds_clp->cl_count);
1871         data->ds_clp = ds->ds_clp;
1872         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1873         if (fh)
1874                 data->args.fh = fh;
1875
1876         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1877                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1878                                                &ff_layout_commit_call_ops_v4,
1879                                    how, RPC_TASK_SOFTCONN);
1880         put_rpccred(ds_cred);
1881         return ret;
1882 out_err:
1883         pnfs_generic_prepare_to_resend_writes(data);
1884         pnfs_generic_commit_release(data);
1885         return -EAGAIN;
1886 }
1887
1888 static int
1889 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1890                            int how, struct nfs_commit_info *cinfo)
1891 {
1892         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1893                                             ff_layout_initiate_commit);
1894 }
1895
1896 static struct pnfs_ds_commit_info *
1897 ff_layout_get_ds_info(struct inode *inode)
1898 {
1899         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1900
1901         if (layout == NULL)
1902                 return NULL;
1903
1904         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1905 }
1906
1907 static void
1908 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1909 {
1910         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1911                                                   id_node));
1912 }
1913
1914 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1915                                   const struct nfs4_layoutreturn_args *args,
1916                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1917 {
1918         __be32 *start;
1919
1920         start = xdr_reserve_space(xdr, 4);
1921         if (unlikely(!start))
1922                 return -E2BIG;
1923
1924         *start = cpu_to_be32(ff_args->num_errors);
1925         /* This assume we always return _ALL_ layouts */
1926         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1927 }
1928
1929 static void
1930 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1931 {
1932         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
1933 }
1934
1935 static void
1936 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1937                             const nfs4_stateid *stateid,
1938                             const struct nfs42_layoutstat_devinfo *devinfo)
1939 {
1940         __be32 *p;
1941
1942         p = xdr_reserve_space(xdr, 8 + 8);
1943         p = xdr_encode_hyper(p, devinfo->offset);
1944         p = xdr_encode_hyper(p, devinfo->length);
1945         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
1946         p = xdr_reserve_space(xdr, 4*8);
1947         p = xdr_encode_hyper(p, devinfo->read_count);
1948         p = xdr_encode_hyper(p, devinfo->read_bytes);
1949         p = xdr_encode_hyper(p, devinfo->write_count);
1950         p = xdr_encode_hyper(p, devinfo->write_bytes);
1951         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
1952 }
1953
1954 static void
1955 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
1956                             const nfs4_stateid *stateid,
1957                             const struct nfs42_layoutstat_devinfo *devinfo)
1958 {
1959         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
1960         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
1961                         devinfo->ld_private.data);
1962 }
1963
1964 /* report nothing for now */
1965 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
1966                 const struct nfs4_layoutreturn_args *args,
1967                 struct nfs4_flexfile_layoutreturn_args *ff_args)
1968 {
1969         __be32 *p;
1970         int i;
1971
1972         p = xdr_reserve_space(xdr, 4);
1973         *p = cpu_to_be32(ff_args->num_dev);
1974         for (i = 0; i < ff_args->num_dev; i++)
1975                 ff_layout_encode_ff_iostat(xdr,
1976                                 &args->layout->plh_stateid,
1977                                 &ff_args->devinfo[i]);
1978 }
1979
1980 static void
1981 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
1982                 unsigned int num_entries)
1983 {
1984         unsigned int i;
1985
1986         for (i = 0; i < num_entries; i++) {
1987                 if (!devinfo[i].ld_private.ops)
1988                         continue;
1989                 if (!devinfo[i].ld_private.ops->free)
1990                         continue;
1991                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
1992         }
1993 }
1994
1995 static struct nfs4_deviceid_node *
1996 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1997                               struct pnfs_device *pdev, gfp_t gfp_flags)
1998 {
1999         struct nfs4_ff_layout_ds *dsaddr;
2000
2001         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2002         if (!dsaddr)
2003                 return NULL;
2004         return &dsaddr->id_node;
2005 }
2006
2007 static void
2008 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2009                 const void *voidargs,
2010                 const struct nfs4_xdr_opaque_data *ff_opaque)
2011 {
2012         const struct nfs4_layoutreturn_args *args = voidargs;
2013         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2014         struct xdr_buf tmp_buf = {
2015                 .head = {
2016                         [0] = {
2017                                 .iov_base = page_address(ff_args->pages[0]),
2018                         },
2019                 },
2020                 .buflen = PAGE_SIZE,
2021         };
2022         struct xdr_stream tmp_xdr;
2023         __be32 *start;
2024
2025         dprintk("%s: Begin\n", __func__);
2026
2027         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL);
2028
2029         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2030         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2031
2032         start = xdr_reserve_space(xdr, 4);
2033         *start = cpu_to_be32(tmp_buf.len);
2034         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2035
2036         dprintk("%s: Return\n", __func__);
2037 }
2038
2039 static void
2040 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2041 {
2042         struct nfs4_flexfile_layoutreturn_args *ff_args;
2043
2044         if (!args->data)
2045                 return;
2046         ff_args = args->data;
2047         args->data = NULL;
2048
2049         ff_layout_free_ds_ioerr(&ff_args->errors);
2050         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2051
2052         put_page(ff_args->pages[0]);
2053         kfree(ff_args);
2054 }
2055
2056 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2057         .encode = ff_layout_encode_layoutreturn,
2058         .free = ff_layout_free_layoutreturn,
2059 };
2060
2061 static int
2062 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2063 {
2064         struct nfs4_flexfile_layoutreturn_args *ff_args;
2065         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2066
2067         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2068         if (!ff_args)
2069                 goto out_nomem;
2070         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2071         if (!ff_args->pages[0])
2072                 goto out_nomem_free;
2073
2074         INIT_LIST_HEAD(&ff_args->errors);
2075         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2076                         &args->range, &ff_args->errors,
2077                         FF_LAYOUTRETURN_MAXERR);
2078
2079         spin_lock(&args->inode->i_lock);
2080         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2081                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2082         spin_unlock(&args->inode->i_lock);
2083
2084         args->ld_private->ops = &layoutreturn_ops;
2085         args->ld_private->data = ff_args;
2086         return 0;
2087 out_nomem_free:
2088         kfree(ff_args);
2089 out_nomem:
2090         return -ENOMEM;
2091 }
2092
2093 static int
2094 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2095 {
2096         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2097
2098         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2099 }
2100
2101 static size_t
2102 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2103                           const int buflen)
2104 {
2105         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2106         const struct in6_addr *addr = &sin6->sin6_addr;
2107
2108         /*
2109          * RFC 4291, Section 2.2.2
2110          *
2111          * Shorthanded ANY address
2112          */
2113         if (ipv6_addr_any(addr))
2114                 return snprintf(buf, buflen, "::");
2115
2116         /*
2117          * RFC 4291, Section 2.2.2
2118          *
2119          * Shorthanded loopback address
2120          */
2121         if (ipv6_addr_loopback(addr))
2122                 return snprintf(buf, buflen, "::1");
2123
2124         /*
2125          * RFC 4291, Section 2.2.3
2126          *
2127          * Special presentation address format for mapped v4
2128          * addresses.
2129          */
2130         if (ipv6_addr_v4mapped(addr))
2131                 return snprintf(buf, buflen, "::ffff:%pI4",
2132                                         &addr->s6_addr32[3]);
2133
2134         /*
2135          * RFC 4291, Section 2.2.1
2136          */
2137         return snprintf(buf, buflen, "%pI6c", addr);
2138 }
2139
2140 /* Derived from rpc_sockaddr2uaddr */
2141 static void
2142 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2143 {
2144         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2145         char portbuf[RPCBIND_MAXUADDRPLEN];
2146         char addrbuf[RPCBIND_MAXUADDRLEN];
2147         char *netid;
2148         unsigned short port;
2149         int len, netid_len;
2150         __be32 *p;
2151
2152         switch (sap->sa_family) {
2153         case AF_INET:
2154                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2155                         return;
2156                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2157                 netid = "tcp";
2158                 netid_len = 3;
2159                 break;
2160         case AF_INET6:
2161                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2162                         return;
2163                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2164                 netid = "tcp6";
2165                 netid_len = 4;
2166                 break;
2167         default:
2168                 /* we only support tcp and tcp6 */
2169                 WARN_ON_ONCE(1);
2170                 return;
2171         }
2172
2173         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2174         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2175
2176         p = xdr_reserve_space(xdr, 4 + netid_len);
2177         xdr_encode_opaque(p, netid, netid_len);
2178
2179         p = xdr_reserve_space(xdr, 4 + len);
2180         xdr_encode_opaque(p, addrbuf, len);
2181 }
2182
2183 static void
2184 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2185                          ktime_t t)
2186 {
2187         struct timespec64 ts;
2188         __be32 *p;
2189
2190         p = xdr_reserve_space(xdr, 12);
2191         ts = ktime_to_timespec64(t);
2192         p = xdr_encode_hyper(p, ts.tv_sec);
2193         *p++ = cpu_to_be32(ts.tv_nsec);
2194 }
2195
2196 static void
2197 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2198                             struct nfs4_ff_io_stat *stat)
2199 {
2200         __be32 *p;
2201
2202         p = xdr_reserve_space(xdr, 5 * 8);
2203         p = xdr_encode_hyper(p, stat->ops_requested);
2204         p = xdr_encode_hyper(p, stat->bytes_requested);
2205         p = xdr_encode_hyper(p, stat->ops_completed);
2206         p = xdr_encode_hyper(p, stat->bytes_completed);
2207         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2208         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2209         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2210 }
2211
2212 static void
2213 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2214                               const struct nfs42_layoutstat_devinfo *devinfo,
2215                               struct nfs4_ff_layout_mirror *mirror)
2216 {
2217         struct nfs4_pnfs_ds_addr *da;
2218         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2219         struct nfs_fh *fh = &mirror->fh_versions[0];
2220         __be32 *p;
2221
2222         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2223         dprintk("%s: DS %s: encoding address %s\n",
2224                 __func__, ds->ds_remotestr, da->da_remotestr);
2225         /* netaddr4 */
2226         ff_layout_encode_netaddr(xdr, da);
2227         /* nfs_fh4 */
2228         p = xdr_reserve_space(xdr, 4 + fh->size);
2229         xdr_encode_opaque(p, fh->data, fh->size);
2230         /* ff_io_latency4 read */
2231         spin_lock(&mirror->lock);
2232         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2233         /* ff_io_latency4 write */
2234         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2235         spin_unlock(&mirror->lock);
2236         /* nfstime4 */
2237         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2238         /* bool */
2239         p = xdr_reserve_space(xdr, 4);
2240         *p = cpu_to_be32(false);
2241 }
2242
2243 static void
2244 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2245                              const struct nfs4_xdr_opaque_data *opaque)
2246 {
2247         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2248                         struct nfs42_layoutstat_devinfo, ld_private);
2249         __be32 *start;
2250
2251         /* layoutupdate length */
2252         start = xdr_reserve_space(xdr, 4);
2253         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2254
2255         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2256 }
2257
2258 static void
2259 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2260 {
2261         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2262
2263         ff_layout_put_mirror(mirror);
2264 }
2265
2266 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2267         .encode = ff_layout_encode_layoutstats,
2268         .free   = ff_layout_free_layoutstats,
2269 };
2270
2271 static int
2272 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2273                                struct nfs42_layoutstat_devinfo *devinfo,
2274                                int dev_limit)
2275 {
2276         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2277         struct nfs4_ff_layout_mirror *mirror;
2278         struct nfs4_deviceid_node *dev;
2279         int i = 0;
2280
2281         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2282                 if (i >= dev_limit)
2283                         break;
2284                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2285                         continue;
2286                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2287                         continue;
2288                 /* mirror refcount put in cleanup_layoutstats */
2289                 if (!refcount_inc_not_zero(&mirror->ref))
2290                         continue;
2291                 dev = &mirror->mirror_ds->id_node;
2292                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2293                 devinfo->offset = 0;
2294                 devinfo->length = NFS4_MAX_UINT64;
2295                 spin_lock(&mirror->lock);
2296                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2297                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2298                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2299                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2300                 spin_unlock(&mirror->lock);
2301                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2302                 devinfo->ld_private.ops = &layoutstat_ops;
2303                 devinfo->ld_private.data = mirror;
2304
2305                 devinfo++;
2306                 i++;
2307         }
2308         return i;
2309 }
2310
2311 static int
2312 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2313 {
2314         struct nfs4_flexfile_layout *ff_layout;
2315         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2316
2317         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2318         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2319         if (!args->devinfo)
2320                 return -ENOMEM;
2321
2322         spin_lock(&args->inode->i_lock);
2323         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2324         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2325                         &args->devinfo[0], dev_count);
2326         spin_unlock(&args->inode->i_lock);
2327         if (!args->num_dev) {
2328                 kfree(args->devinfo);
2329                 args->devinfo = NULL;
2330                 return -ENOENT;
2331         }
2332
2333         return 0;
2334 }
2335
2336 static int
2337 ff_layout_set_layoutdriver(struct nfs_server *server,
2338                 const struct nfs_fh *dummy)
2339 {
2340 #if IS_ENABLED(CONFIG_NFS_V4_2)
2341         server->caps |= NFS_CAP_LAYOUTSTATS;
2342 #endif
2343         return 0;
2344 }
2345
2346 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2347         .id                     = LAYOUT_FLEX_FILES,
2348         .name                   = "LAYOUT_FLEX_FILES",
2349         .owner                  = THIS_MODULE,
2350         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2351         .set_layoutdriver       = ff_layout_set_layoutdriver,
2352         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2353         .free_layout_hdr        = ff_layout_free_layout_hdr,
2354         .alloc_lseg             = ff_layout_alloc_lseg,
2355         .free_lseg              = ff_layout_free_lseg,
2356         .add_lseg               = ff_layout_add_lseg,
2357         .pg_read_ops            = &ff_layout_pg_read_ops,
2358         .pg_write_ops           = &ff_layout_pg_write_ops,
2359         .get_ds_info            = ff_layout_get_ds_info,
2360         .free_deviceid_node     = ff_layout_free_deviceid_node,
2361         .mark_request_commit    = pnfs_layout_mark_request_commit,
2362         .clear_request_commit   = pnfs_generic_clear_request_commit,
2363         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2364         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2365         .commit_pagelist        = ff_layout_commit_pagelist,
2366         .read_pagelist          = ff_layout_read_pagelist,
2367         .write_pagelist         = ff_layout_write_pagelist,
2368         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2369         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2370         .sync                   = pnfs_nfs_generic_sync,
2371         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2372 };
2373
2374 static int __init nfs4flexfilelayout_init(void)
2375 {
2376         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2377                __func__);
2378         if (!ff_zero_group) {
2379                 ff_zero_group = groups_alloc(0);
2380                 if (!ff_zero_group)
2381                         return -ENOMEM;
2382         }
2383         return pnfs_register_layoutdriver(&flexfilelayout_type);
2384 }
2385
2386 static void __exit nfs4flexfilelayout_exit(void)
2387 {
2388         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2389                __func__);
2390         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2391         if (ff_zero_group) {
2392                 put_group_info(ff_zero_group);
2393                 ff_zero_group = NULL;
2394         }
2395 }
2396
2397 MODULE_ALIAS("nfs-layouttype4-4");
2398
2399 MODULE_LICENSE("GPL");
2400 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2401
2402 module_init(nfs4flexfilelayout_init);
2403 module_exit(nfs4flexfilelayout_exit);