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 > NFS_MAXFHSIZE) {
 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 (atomic_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                 atomic_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 && atomic_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 (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
 925                 goto out_eagain;
 926
 927         for (i = 0; i < pgio->pg_mirror_count; i++) {
 928                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 929                 if (!ds) {
 930                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 931                                 goto out_mds;
 932                         pnfs_put_lseg(pgio->pg_lseg);
 933                         pgio->pg_lseg = NULL;
 934                         /* Sleep for 1 second before retrying */
 935                         ssleep(1);
 936                         goto retry;
 937                 }
 938                 pgm = &pgio->pg_mirrors[i];
 939                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 940                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 941         }
 942
 943         return;
 944 out_eagain:
 945         pnfs_generic_pg_cleanup(pgio);
 946         pgio->pg_error = -EAGAIN;
 947         return;
 948 out_mds:
 949         pnfs_put_lseg(pgio->pg_lseg);
 950         pgio->pg_lseg = NULL;
 951         nfs_pageio_reset_write_mds(pgio);
 952         pgio->pg_error = -EAGAIN;
 953 }
 954
 955 static unsigned int
 956 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 957                                     struct nfs_page *req)
 958 {
 959         if (!pgio->pg_lseg) {
 960                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 961                                                    req->wb_context,
 962                                                    0,
 963                                                    NFS4_MAX_UINT64,
 964                                                    IOMODE_RW,
 965                                                    false,
 966                                                    GFP_NOFS);
 967                 if (IS_ERR(pgio->pg_lseg)) {
 968                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 969                         pgio->pg_lseg = NULL;
 970                         goto out;
 971                 }
 972         }
 973         if (pgio->pg_lseg)
 974                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 975
 976         /* no lseg means that pnfs is not in use, so no mirroring here */
 977         nfs_pageio_reset_write_mds(pgio);
 978 out:
 979         return 1;
 980 }
 981
 982 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 983         .pg_init = ff_layout_pg_init_read,
 984         .pg_test = pnfs_generic_pg_test,
 985         .pg_doio = pnfs_generic_pg_readpages,
 986         .pg_cleanup = pnfs_generic_pg_cleanup,
 987 };
 988
 989 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 990         .pg_init = ff_layout_pg_init_write,
 991         .pg_test = pnfs_generic_pg_test,
 992         .pg_doio = pnfs_generic_pg_writepages,
 993         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 994         .pg_cleanup = pnfs_generic_pg_cleanup,
 995 };
 996
 997 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 998 {
 999         struct rpc_task *task = &hdr->task;
1000
1001         pnfs_layoutcommit_inode(hdr->inode, false);
1002
1003         if (retry_pnfs) {
1004                 dprintk("%s Reset task %5u for i/o through pNFS "
1005                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1006                         hdr->task.tk_pid,
1007                         hdr->inode->i_sb->s_id,
1008                         (unsigned long long)NFS_FILEID(hdr->inode),
1009                         hdr->args.count,
1010                         (unsigned long long)hdr->args.offset);
1011
1012                 hdr->completion_ops->reschedule_io(hdr);
1013                 return;
1014         }
1015
1016         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1017                 dprintk("%s Reset task %5u for i/o through MDS "
1018                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1019                         hdr->task.tk_pid,
1020                         hdr->inode->i_sb->s_id,
1021                         (unsigned long long)NFS_FILEID(hdr->inode),
1022                         hdr->args.count,
1023                         (unsigned long long)hdr->args.offset);
1024
1025                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1026         }
1027 }
1028
1029 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1030 {
1031         struct rpc_task *task = &hdr->task;
1032
1033         pnfs_layoutcommit_inode(hdr->inode, false);
1034
1035         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1036                 dprintk("%s Reset task %5u for i/o through MDS "
1037                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1038                         hdr->task.tk_pid,
1039                         hdr->inode->i_sb->s_id,
1040                         (unsigned long long)NFS_FILEID(hdr->inode),
1041                         hdr->args.count,
1042                         (unsigned long long)hdr->args.offset);
1043
1044                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1045         }
1046 }
1047
1048 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1049                                            struct nfs4_state *state,
1050                                            struct nfs_client *clp,
1051                                            struct pnfs_layout_segment *lseg,
1052                                            int idx)
1053 {
1054         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1055         struct inode *inode = lo->plh_inode;
1056         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1057         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1058
1059         switch (task->tk_status) {
1060         case -NFS4ERR_BADSESSION:
1061         case -NFS4ERR_BADSLOT:
1062         case -NFS4ERR_BAD_HIGH_SLOT:
1063         case -NFS4ERR_DEADSESSION:
1064         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1065         case -NFS4ERR_SEQ_FALSE_RETRY:
1066         case -NFS4ERR_SEQ_MISORDERED:
1067                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1068                         "flags 0x%x\n", __func__, task->tk_status,
1069                         clp->cl_exchange_flags);
1070                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1071                 break;
1072         case -NFS4ERR_DELAY:
1073         case -NFS4ERR_GRACE:
1074                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1075                 break;
1076         case -NFS4ERR_RETRY_UNCACHED_REP:
1077                 break;
1078         /* Invalidate Layout errors */
1079         case -NFS4ERR_PNFS_NO_LAYOUT:
1080         case -ESTALE:           /* mapped NFS4ERR_STALE */
1081         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1082         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1083         case -NFS4ERR_FHEXPIRED:
1084         case -NFS4ERR_WRONG_TYPE:
1085                 dprintk("%s Invalid layout error %d\n", __func__,
1086                         task->tk_status);
1087                 /*
1088                  * Destroy layout so new i/o will get a new layout.
1089                  * Layout will not be destroyed until all current lseg
1090                  * references are put. Mark layout as invalid to resend failed
1091                  * i/o and all i/o waiting on the slot table to the MDS until
1092                  * layout is destroyed and a new valid layout is obtained.
1093                  */
1094                 pnfs_destroy_layout(NFS_I(inode));
1095                 rpc_wake_up(&tbl->slot_tbl_waitq);
1096                 goto reset;
1097         /* RPC connection errors */
1098         case -ECONNREFUSED:
1099         case -EHOSTDOWN:
1100         case -EHOSTUNREACH:
1101         case -ENETUNREACH:
1102         case -EIO:
1103         case -ETIMEDOUT:
1104         case -EPIPE:
1105                 dprintk("%s DS connection error %d\n", __func__,
1106                         task->tk_status);
1107                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1108                                 &devid->deviceid);
1109                 rpc_wake_up(&tbl->slot_tbl_waitq);
1110                 /* fall through */
1111         default:
1112                 if (ff_layout_avoid_mds_available_ds(lseg))
1113                         return -NFS4ERR_RESET_TO_PNFS;
1114 reset:
1115                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1116                         task->tk_status);
1117                 return -NFS4ERR_RESET_TO_MDS;
1118         }
1119         task->tk_status = 0;
1120         return -EAGAIN;
1121 }
1122
1123 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1124 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1125                                            struct pnfs_layout_segment *lseg,
1126                                            int idx)
1127 {
1128         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1129
1130         switch (task->tk_status) {
1131         /* File access problems. Don't mark the device as unavailable */
1132         case -EACCES:
1133         case -ESTALE:
1134         case -EISDIR:
1135         case -EBADHANDLE:
1136         case -ELOOP:
1137         case -ENOSPC:
1138                 break;
1139         case -EJUKEBOX:
1140                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1141                 goto out_retry;
1142         default:
1143                 dprintk("%s DS connection error %d\n", __func__,
1144                         task->tk_status);
1145                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1146                                 &devid->deviceid);
1147         }
1148         /* FIXME: Need to prevent infinite looping here. */
1149         return -NFS4ERR_RESET_TO_PNFS;
1150 out_retry:
1151         task->tk_status = 0;
1152         rpc_restart_call_prepare(task);
1153         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1154         return -EAGAIN;
1155 }
1156
1157 static int ff_layout_async_handle_error(struct rpc_task *task,
1158                                         struct nfs4_state *state,
1159                                         struct nfs_client *clp,
1160                                         struct pnfs_layout_segment *lseg,
1161                                         int idx)
1162 {
1163         int vers = clp->cl_nfs_mod->rpc_vers->number;
1164
1165         if (task->tk_status >= 0)
1166                 return 0;
1167
1168         /* Handle the case of an invalid layout segment */
1169         if (!pnfs_is_valid_lseg(lseg))
1170                 return -NFS4ERR_RESET_TO_PNFS;
1171
1172         switch (vers) {
1173         case 3:
1174                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1175         case 4:
1176                 return ff_layout_async_handle_error_v4(task, state, clp,
1177                                                        lseg, idx);
1178         default:
1179                 /* should never happen */
1180                 WARN_ON_ONCE(1);
1181                 return 0;
1182         }
1183 }
1184
1185 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1186                                         int idx, u64 offset, u64 length,
1187                                         u32 status, int opnum, int error)
1188 {
1189         struct nfs4_ff_layout_mirror *mirror;
1190         int err;
1191
1192         if (status == 0) {
1193                 switch (error) {
1194                 case -ETIMEDOUT:
1195                 case -EPFNOSUPPORT:
1196                 case -EPROTONOSUPPORT:
1197                 case -EOPNOTSUPP:
1198                 case -EINVAL:
1199                 case -ECONNREFUSED:
1200                 case -ECONNRESET:
1201                 case -EHOSTDOWN:
1202                 case -EHOSTUNREACH:
1203                 case -ENETUNREACH:
1204                 case -EADDRINUSE:
1205                 case -ENOBUFS:
1206                 case -EPIPE:
1207                 case -EPERM:
1208                         status = NFS4ERR_NXIO;
1209                         break;
1210                 case -EACCES:
1211                         status = NFS4ERR_ACCESS;
1212                         break;
1213                 default:
1214                         return;
1215                 }
1216         }
1217
1218         switch (status) {
1219         case NFS4ERR_DELAY:
1220         case NFS4ERR_GRACE:
1221                 return;
1222         default:
1223                 break;
1224         }
1225
1226         mirror = FF_LAYOUT_COMP(lseg, idx);
1227         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1228                                        mirror, offset, length, status, opnum,
1229                                        GFP_NOIO);
1230         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1231         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1232 }
1233
1234 /* NFS_PROTO call done callback routines */
1235 static int ff_layout_read_done_cb(struct rpc_task *task,
1236                                 struct nfs_pgio_header *hdr)
1237 {
1238         int err;
1239
1240         trace_nfs4_pnfs_read(hdr, task->tk_status);
1241         if (task->tk_status < 0)
1242                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1243                                             hdr->args.offset, hdr->args.count,
1244                                             hdr->res.op_status, OP_READ,
1245                                             task->tk_status);
1246         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1247                                            hdr->ds_clp, hdr->lseg,
1248                                            hdr->pgio_mirror_idx);
1249
1250         switch (err) {
1251         case -NFS4ERR_RESET_TO_PNFS:
1252                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1253                                         hdr->pgio_mirror_idx + 1,
1254                                         &hdr->pgio_mirror_idx))
1255                         goto out_eagain;
1256                 ff_layout_read_record_layoutstats_done(task, hdr);
1257                 pnfs_read_resend_pnfs(hdr);
1258                 return task->tk_status;
1259         case -NFS4ERR_RESET_TO_MDS:
1260                 ff_layout_reset_read(hdr);
1261                 return task->tk_status;
1262         case -EAGAIN:
1263                 goto out_eagain;
1264         }
1265
1266         return 0;
1267 out_eagain:
1268         rpc_restart_call_prepare(task);
1269         return -EAGAIN;
1270 }
1271
1272 static bool
1273 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1274 {
1275         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1276 }
1277
1278 /*
1279  * We reference the rpc_cred of the first WRITE that triggers the need for
1280  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1281  * rfc5661 is not clear about which credential should be used.
1282  *
1283  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1284  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1285  * we always send layoutcommit after DS writes.
1286  */
1287 static void
1288 ff_layout_set_layoutcommit(struct inode *inode,
1289                 struct pnfs_layout_segment *lseg,
1290                 loff_t end_offset)
1291 {
1292         if (!ff_layout_need_layoutcommit(lseg))
1293                 return;
1294
1295         pnfs_set_layoutcommit(inode, lseg, end_offset);
1296         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1297                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1298 }
1299
1300 static bool
1301 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx)
1302 {
1303         /* No mirroring for now */
1304         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1305
1306         return ff_layout_test_devid_unavailable(node);
1307 }
1308
1309 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1310                 struct nfs_pgio_header *hdr)
1311 {
1312         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1313                 return;
1314         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1315                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1316                         hdr->args.count,
1317                         task->tk_start);
1318 }
1319
1320 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1321                 struct nfs_pgio_header *hdr)
1322 {
1323         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1324                 return;
1325         nfs4_ff_layout_stat_io_end_read(task,
1326                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1327                         hdr->args.count,
1328                         hdr->res.count);
1329 }
1330
1331 static int ff_layout_read_prepare_common(struct rpc_task *task,
1332                                          struct nfs_pgio_header *hdr)
1333 {
1334         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1335                 rpc_exit(task, -EIO);
1336                 return -EIO;
1337         }
1338         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1339                 rpc_exit(task, -EHOSTDOWN);
1340                 return -EAGAIN;
1341         }
1342
1343         ff_layout_read_record_layoutstats_start(task, hdr);
1344         return 0;
1345 }
1346
1347 /*
1348  * Call ops for the async read/write cases
1349  * In the case of dense layouts, the offset needs to be reset to its
1350  * original value.
1351  */
1352 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1353 {
1354         struct nfs_pgio_header *hdr = data;
1355
1356         if (ff_layout_read_prepare_common(task, hdr))
1357                 return;
1358
1359         rpc_call_start(task);
1360 }
1361
1362 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1363 {
1364         struct nfs_pgio_header *hdr = data;
1365
1366         if (nfs4_setup_sequence(hdr->ds_clp,
1367                                 &hdr->args.seq_args,
1368                                 &hdr->res.seq_res,
1369                                 task))
1370                 return;
1371
1372         ff_layout_read_prepare_common(task, hdr);
1373 }
1374
1375 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1376 {
1377         struct nfs_pgio_header *hdr = data;
1378
1379         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1380
1381         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1382             task->tk_status == 0) {
1383                 nfs4_sequence_done(task, &hdr->res.seq_res);
1384                 return;
1385         }
1386
1387         /* Note this may cause RPC to be resent */
1388         hdr->mds_ops->rpc_call_done(task, hdr);
1389 }
1390
1391 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1392 {
1393         struct nfs_pgio_header *hdr = data;
1394
1395         ff_layout_read_record_layoutstats_done(task, hdr);
1396         rpc_count_iostats_metrics(task,
1397             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1398 }
1399
1400 static void ff_layout_read_release(void *data)
1401 {
1402         struct nfs_pgio_header *hdr = data;
1403
1404         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1405         pnfs_generic_rw_release(data);
1406 }
1407
1408
1409 static int ff_layout_write_done_cb(struct rpc_task *task,
1410                                 struct nfs_pgio_header *hdr)
1411 {
1412         loff_t end_offs = 0;
1413         int err;
1414
1415         trace_nfs4_pnfs_write(hdr, task->tk_status);
1416         if (task->tk_status < 0)
1417                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1418                                             hdr->args.offset, hdr->args.count,
1419                                             hdr->res.op_status, OP_WRITE,
1420                                             task->tk_status);
1421         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1422                                            hdr->ds_clp, hdr->lseg,
1423                                            hdr->pgio_mirror_idx);
1424
1425         switch (err) {
1426         case -NFS4ERR_RESET_TO_PNFS:
1427                 ff_layout_reset_write(hdr, true);
1428                 return task->tk_status;
1429         case -NFS4ERR_RESET_TO_MDS:
1430                 ff_layout_reset_write(hdr, false);
1431                 return task->tk_status;
1432         case -EAGAIN:
1433                 return -EAGAIN;
1434         }
1435
1436         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1437             hdr->res.verf->committed == NFS_DATA_SYNC)
1438                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1439
1440         /* Note: if the write is unstable, don't set end_offs until commit */
1441         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1442
1443         /* zero out fattr since we don't care DS attr at all */
1444         hdr->fattr.valid = 0;
1445         if (task->tk_status >= 0)
1446                 nfs_writeback_update_inode(hdr);
1447
1448         return 0;
1449 }
1450
1451 static int ff_layout_commit_done_cb(struct rpc_task *task,
1452                                      struct nfs_commit_data *data)
1453 {
1454         int err;
1455
1456         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1457         if (task->tk_status < 0)
1458                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1459                                             data->args.offset, data->args.count,
1460                                             data->res.op_status, OP_COMMIT,
1461                                             task->tk_status);
1462         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1463                                            data->lseg, data->ds_commit_index);
1464
1465         switch (err) {
1466         case -NFS4ERR_RESET_TO_PNFS:
1467                 pnfs_generic_prepare_to_resend_writes(data);
1468                 return -EAGAIN;
1469         case -NFS4ERR_RESET_TO_MDS:
1470                 pnfs_generic_prepare_to_resend_writes(data);
1471                 return -EAGAIN;
1472         case -EAGAIN:
1473                 rpc_restart_call_prepare(task);
1474                 return -EAGAIN;
1475         }
1476
1477         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1478
1479         return 0;
1480 }
1481
1482 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1483                 struct nfs_pgio_header *hdr)
1484 {
1485         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1486                 return;
1487         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1488                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1489                         hdr->args.count,
1490                         task->tk_start);
1491 }
1492
1493 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1494                 struct nfs_pgio_header *hdr)
1495 {
1496         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1497                 return;
1498         nfs4_ff_layout_stat_io_end_write(task,
1499                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1500                         hdr->args.count, hdr->res.count,
1501                         hdr->res.verf->committed);
1502 }
1503
1504 static int ff_layout_write_prepare_common(struct rpc_task *task,
1505                                           struct nfs_pgio_header *hdr)
1506 {
1507         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1508                 rpc_exit(task, -EIO);
1509                 return -EIO;
1510         }
1511
1512         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1513                 rpc_exit(task, -EHOSTDOWN);
1514                 return -EAGAIN;
1515         }
1516
1517         ff_layout_write_record_layoutstats_start(task, hdr);
1518         return 0;
1519 }
1520
1521 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1522 {
1523         struct nfs_pgio_header *hdr = data;
1524
1525         if (ff_layout_write_prepare_common(task, hdr))
1526                 return;
1527
1528         rpc_call_start(task);
1529 }
1530
1531 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1532 {
1533         struct nfs_pgio_header *hdr = data;
1534
1535         if (nfs4_setup_sequence(hdr->ds_clp,
1536                                 &hdr->args.seq_args,
1537                                 &hdr->res.seq_res,
1538                                 task))
1539                 return;
1540
1541         ff_layout_write_prepare_common(task, hdr);
1542 }
1543
1544 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1545 {
1546         struct nfs_pgio_header *hdr = data;
1547
1548         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1549             task->tk_status == 0) {
1550                 nfs4_sequence_done(task, &hdr->res.seq_res);
1551                 return;
1552         }
1553
1554         /* Note this may cause RPC to be resent */
1555         hdr->mds_ops->rpc_call_done(task, hdr);
1556 }
1557
1558 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1559 {
1560         struct nfs_pgio_header *hdr = data;
1561
1562         ff_layout_write_record_layoutstats_done(task, hdr);
1563         rpc_count_iostats_metrics(task,
1564             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1565 }
1566
1567 static void ff_layout_write_release(void *data)
1568 {
1569         struct nfs_pgio_header *hdr = data;
1570
1571         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1572         pnfs_generic_rw_release(data);
1573 }
1574
1575 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1576                 struct nfs_commit_data *cdata)
1577 {
1578         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1579                 return;
1580         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1581                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1582                         0, task->tk_start);
1583 }
1584
1585 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1586                 struct nfs_commit_data *cdata)
1587 {
1588         struct nfs_page *req;
1589         __u64 count = 0;
1590
1591         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1592                 return;
1593
1594         if (task->tk_status == 0) {
1595                 list_for_each_entry(req, &cdata->pages, wb_list)
1596                         count += req->wb_bytes;
1597         }
1598         nfs4_ff_layout_stat_io_end_write(task,
1599                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1600                         count, count, NFS_FILE_SYNC);
1601 }
1602
1603 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1604                 struct nfs_commit_data *cdata)
1605 {
1606         ff_layout_commit_record_layoutstats_start(task, cdata);
1607 }
1608
1609 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1610 {
1611         ff_layout_commit_prepare_common(task, data);
1612         rpc_call_start(task);
1613 }
1614
1615 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1616 {
1617         struct nfs_commit_data *wdata = data;
1618
1619         if (nfs4_setup_sequence(wdata->ds_clp,
1620                                 &wdata->args.seq_args,
1621                                 &wdata->res.seq_res,
1622                                 task))
1623                 return;
1624         ff_layout_commit_prepare_common(task, data);
1625 }
1626
1627 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1628 {
1629         pnfs_generic_write_commit_done(task, data);
1630 }
1631
1632 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1633 {
1634         struct nfs_commit_data *cdata = data;
1635
1636         ff_layout_commit_record_layoutstats_done(task, cdata);
1637         rpc_count_iostats_metrics(task,
1638             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1639 }
1640
1641 static void ff_layout_commit_release(void *data)
1642 {
1643         struct nfs_commit_data *cdata = data;
1644
1645         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1646         pnfs_generic_commit_release(data);
1647 }
1648
1649 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1650         .rpc_call_prepare = ff_layout_read_prepare_v3,
1651         .rpc_call_done = ff_layout_read_call_done,
1652         .rpc_count_stats = ff_layout_read_count_stats,
1653         .rpc_release = ff_layout_read_release,
1654 };
1655
1656 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1657         .rpc_call_prepare = ff_layout_read_prepare_v4,
1658         .rpc_call_done = ff_layout_read_call_done,
1659         .rpc_count_stats = ff_layout_read_count_stats,
1660         .rpc_release = ff_layout_read_release,
1661 };
1662
1663 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1664         .rpc_call_prepare = ff_layout_write_prepare_v3,
1665         .rpc_call_done = ff_layout_write_call_done,
1666         .rpc_count_stats = ff_layout_write_count_stats,
1667         .rpc_release = ff_layout_write_release,
1668 };
1669
1670 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1671         .rpc_call_prepare = ff_layout_write_prepare_v4,
1672         .rpc_call_done = ff_layout_write_call_done,
1673         .rpc_count_stats = ff_layout_write_count_stats,
1674         .rpc_release = ff_layout_write_release,
1675 };
1676
1677 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1678         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1679         .rpc_call_done = ff_layout_commit_done,
1680         .rpc_count_stats = ff_layout_commit_count_stats,
1681         .rpc_release = ff_layout_commit_release,
1682 };
1683
1684 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1685         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1686         .rpc_call_done = ff_layout_commit_done,
1687         .rpc_count_stats = ff_layout_commit_count_stats,
1688         .rpc_release = ff_layout_commit_release,
1689 };
1690
1691 static enum pnfs_try_status
1692 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1693 {
1694         struct pnfs_layout_segment *lseg = hdr->lseg;
1695         struct nfs4_pnfs_ds *ds;
1696         struct rpc_clnt *ds_clnt;
1697         struct rpc_cred *ds_cred;
1698         loff_t offset = hdr->args.offset;
1699         u32 idx = hdr->pgio_mirror_idx;
1700         int vers;
1701         struct nfs_fh *fh;
1702
1703         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1704                 __func__, hdr->inode->i_ino,
1705                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1706
1707         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1708         if (!ds)
1709                 goto out_failed;
1710
1711         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1712                                                    hdr->inode);
1713         if (IS_ERR(ds_clnt))
1714                 goto out_failed;
1715
1716         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1717         if (!ds_cred)
1718                 goto out_failed;
1719
1720         vers = nfs4_ff_layout_ds_version(lseg, idx);
1721
1722         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1723                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1724
1725         hdr->pgio_done_cb = ff_layout_read_done_cb;
1726         atomic_inc(&ds->ds_clp->cl_count);
1727         hdr->ds_clp = ds->ds_clp;
1728         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1729         if (fh)
1730                 hdr->args.fh = fh;
1731
1732         if (vers == 4 &&
1733                 !nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid))
1734                 goto out_failed;
1735
1736         /*
1737          * Note that if we ever decide to split across DSes,
1738          * then we may need to handle dense-like offsets.
1739          */
1740         hdr->args.offset = offset;
1741         hdr->mds_offset = offset;
1742
1743         /* Perform an asynchronous read to ds */
1744         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1745                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1746                                       &ff_layout_read_call_ops_v4,
1747                           0, RPC_TASK_SOFTCONN);
1748         put_rpccred(ds_cred);
1749         return PNFS_ATTEMPTED;
1750
1751 out_failed:
1752         if (ff_layout_avoid_mds_available_ds(lseg))
1753                 return PNFS_TRY_AGAIN;
1754         return PNFS_NOT_ATTEMPTED;
1755 }
1756
1757 /* Perform async writes. */
1758 static enum pnfs_try_status
1759 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1760 {
1761         struct pnfs_layout_segment *lseg = hdr->lseg;
1762         struct nfs4_pnfs_ds *ds;
1763         struct rpc_clnt *ds_clnt;
1764         struct rpc_cred *ds_cred;
1765         loff_t offset = hdr->args.offset;
1766         int vers;
1767         struct nfs_fh *fh;
1768         int idx = hdr->pgio_mirror_idx;
1769
1770         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1771         if (!ds)
1772                 goto out_failed;
1773
1774         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1775                                                    hdr->inode);
1776         if (IS_ERR(ds_clnt))
1777                 goto out_failed;
1778
1779         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1780         if (!ds_cred)
1781                 goto out_failed;
1782
1783         vers = nfs4_ff_layout_ds_version(lseg, idx);
1784
1785         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1786                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1787                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1788                 vers);
1789
1790         hdr->pgio_done_cb = ff_layout_write_done_cb;
1791         atomic_inc(&ds->ds_clp->cl_count);
1792         hdr->ds_clp = ds->ds_clp;
1793         hdr->ds_commit_idx = idx;
1794         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1795         if (fh)
1796                 hdr->args.fh = fh;
1797
1798         if (vers == 4 &&
1799                 !nfs4_ff_layout_select_ds_stateid(lseg, idx, &hdr->args.stateid))
1800                 goto out_failed;
1801
1802         /*
1803          * Note that if we ever decide to split across DSes,
1804          * then we may need to handle dense-like offsets.
1805          */
1806         hdr->args.offset = offset;
1807
1808         /* Perform an asynchronous write */
1809         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1810                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1811                                       &ff_layout_write_call_ops_v4,
1812                           sync, RPC_TASK_SOFTCONN);
1813         put_rpccred(ds_cred);
1814         return PNFS_ATTEMPTED;
1815
1816 out_failed:
1817         if (ff_layout_avoid_mds_available_ds(lseg))
1818                 return PNFS_TRY_AGAIN;
1819         return PNFS_NOT_ATTEMPTED;
1820 }
1821
1822 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1823 {
1824         return i;
1825 }
1826
1827 static struct nfs_fh *
1828 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1829 {
1830         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1831
1832         /* FIXME: Assume that there is only one NFS version available
1833          * for the DS.
1834          */
1835         return &flseg->mirror_array[i]->fh_versions[0];
1836 }
1837
1838 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1839 {
1840         struct pnfs_layout_segment *lseg = data->lseg;
1841         struct nfs4_pnfs_ds *ds;
1842         struct rpc_clnt *ds_clnt;
1843         struct rpc_cred *ds_cred;
1844         u32 idx;
1845         int vers, ret;
1846         struct nfs_fh *fh;
1847
1848         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1849             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1850                 goto out_err;
1851
1852         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1853         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1854         if (!ds)
1855                 goto out_err;
1856
1857         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1858                                                    data->inode);
1859         if (IS_ERR(ds_clnt))
1860                 goto out_err;
1861
1862         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1863         if (!ds_cred)
1864                 goto out_err;
1865
1866         vers = nfs4_ff_layout_ds_version(lseg, idx);
1867
1868         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1869                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1870                 vers);
1871         data->commit_done_cb = ff_layout_commit_done_cb;
1872         data->cred = ds_cred;
1873         atomic_inc(&ds->ds_clp->cl_count);
1874         data->ds_clp = ds->ds_clp;
1875         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1876         if (fh)
1877                 data->args.fh = fh;
1878
1879         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1880                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1881                                                &ff_layout_commit_call_ops_v4,
1882                                    how, RPC_TASK_SOFTCONN);
1883         put_rpccred(ds_cred);
1884         return ret;
1885 out_err:
1886         pnfs_generic_prepare_to_resend_writes(data);
1887         pnfs_generic_commit_release(data);
1888         return -EAGAIN;
1889 }
1890
1891 static int
1892 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1893                            int how, struct nfs_commit_info *cinfo)
1894 {
1895         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1896                                             ff_layout_initiate_commit);
1897 }
1898
1899 static struct pnfs_ds_commit_info *
1900 ff_layout_get_ds_info(struct inode *inode)
1901 {
1902         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1903
1904         if (layout == NULL)
1905                 return NULL;
1906
1907         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1908 }
1909
1910 static void
1911 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1912 {
1913         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1914                                                   id_node));
1915 }
1916
1917 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1918                                   const struct nfs4_layoutreturn_args *args,
1919                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1920 {
1921         __be32 *start;
1922
1923         start = xdr_reserve_space(xdr, 4);
1924         if (unlikely(!start))
1925                 return -E2BIG;
1926
1927         *start = cpu_to_be32(ff_args->num_errors);
1928         /* This assume we always return _ALL_ layouts */
1929         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1930 }
1931
1932 static void
1933 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1934 {
1935         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
1936 }
1937
1938 static void
1939 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
1940                             const nfs4_stateid *stateid,
1941                             const struct nfs42_layoutstat_devinfo *devinfo)
1942 {
1943         __be32 *p;
1944
1945         p = xdr_reserve_space(xdr, 8 + 8);
1946         p = xdr_encode_hyper(p, devinfo->offset);
1947         p = xdr_encode_hyper(p, devinfo->length);
1948         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
1949         p = xdr_reserve_space(xdr, 4*8);
1950         p = xdr_encode_hyper(p, devinfo->read_count);
1951         p = xdr_encode_hyper(p, devinfo->read_bytes);
1952         p = xdr_encode_hyper(p, devinfo->write_count);
1953         p = xdr_encode_hyper(p, devinfo->write_bytes);
1954         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
1955 }
1956
1957 static void
1958 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
1959                             const nfs4_stateid *stateid,
1960                             const struct nfs42_layoutstat_devinfo *devinfo)
1961 {
1962         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
1963         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
1964                         devinfo->ld_private.data);
1965 }
1966
1967 /* report nothing for now */
1968 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
1969                 const struct nfs4_layoutreturn_args *args,
1970                 struct nfs4_flexfile_layoutreturn_args *ff_args)
1971 {
1972         __be32 *p;
1973         int i;
1974
1975         p = xdr_reserve_space(xdr, 4);
1976         *p = cpu_to_be32(ff_args->num_dev);
1977         for (i = 0; i < ff_args->num_dev; i++)
1978                 ff_layout_encode_ff_iostat(xdr,
1979                                 &args->layout->plh_stateid,
1980                                 &ff_args->devinfo[i]);
1981 }
1982
1983 static void
1984 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
1985                 unsigned int num_entries)
1986 {
1987         unsigned int i;
1988
1989         for (i = 0; i < num_entries; i++) {
1990                 if (!devinfo[i].ld_private.ops)
1991                         continue;
1992                 if (!devinfo[i].ld_private.ops->free)
1993                         continue;
1994                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
1995         }
1996 }
1997
1998 static struct nfs4_deviceid_node *
1999 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2000                               struct pnfs_device *pdev, gfp_t gfp_flags)
2001 {
2002         struct nfs4_ff_layout_ds *dsaddr;
2003
2004         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2005         if (!dsaddr)
2006                 return NULL;
2007         return &dsaddr->id_node;
2008 }
2009
2010 static void
2011 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2012                 const void *voidargs,
2013                 const struct nfs4_xdr_opaque_data *ff_opaque)
2014 {
2015         const struct nfs4_layoutreturn_args *args = voidargs;
2016         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2017         struct xdr_buf tmp_buf = {
2018                 .head = {
2019                         [0] = {
2020                                 .iov_base = page_address(ff_args->pages[0]),
2021                         },
2022                 },
2023                 .buflen = PAGE_SIZE,
2024         };
2025         struct xdr_stream tmp_xdr;
2026         __be32 *start;
2027
2028         dprintk("%s: Begin\n", __func__);
2029
2030         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL);
2031
2032         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2033         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2034
2035         start = xdr_reserve_space(xdr, 4);
2036         *start = cpu_to_be32(tmp_buf.len);
2037         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2038
2039         dprintk("%s: Return\n", __func__);
2040 }
2041
2042 static void
2043 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2044 {
2045         struct nfs4_flexfile_layoutreturn_args *ff_args;
2046
2047         if (!args->data)
2048                 return;
2049         ff_args = args->data;
2050         args->data = NULL;
2051
2052         ff_layout_free_ds_ioerr(&ff_args->errors);
2053         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2054
2055         put_page(ff_args->pages[0]);
2056         kfree(ff_args);
2057 }
2058
2059 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2060         .encode = ff_layout_encode_layoutreturn,
2061         .free = ff_layout_free_layoutreturn,
2062 };
2063
2064 static int
2065 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2066 {
2067         struct nfs4_flexfile_layoutreturn_args *ff_args;
2068         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2069
2070         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2071         if (!ff_args)
2072                 goto out_nomem;
2073         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2074         if (!ff_args->pages[0])
2075                 goto out_nomem_free;
2076
2077         INIT_LIST_HEAD(&ff_args->errors);
2078         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2079                         &args->range, &ff_args->errors,
2080                         FF_LAYOUTRETURN_MAXERR);
2081
2082         spin_lock(&args->inode->i_lock);
2083         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2084                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2085         spin_unlock(&args->inode->i_lock);
2086
2087         args->ld_private->ops = &layoutreturn_ops;
2088         args->ld_private->data = ff_args;
2089         return 0;
2090 out_nomem_free:
2091         kfree(ff_args);
2092 out_nomem:
2093         return -ENOMEM;
2094 }
2095
2096 static int
2097 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2098 {
2099         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2100
2101         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2102 }
2103
2104 static size_t
2105 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2106                           const int buflen)
2107 {
2108         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2109         const struct in6_addr *addr = &sin6->sin6_addr;
2110
2111         /*
2112          * RFC 4291, Section 2.2.2
2113          *
2114          * Shorthanded ANY address
2115          */
2116         if (ipv6_addr_any(addr))
2117                 return snprintf(buf, buflen, "::");
2118
2119         /*
2120          * RFC 4291, Section 2.2.2
2121          *
2122          * Shorthanded loopback address
2123          */
2124         if (ipv6_addr_loopback(addr))
2125                 return snprintf(buf, buflen, "::1");
2126
2127         /*
2128          * RFC 4291, Section 2.2.3
2129          *
2130          * Special presentation address format for mapped v4
2131          * addresses.
2132          */
2133         if (ipv6_addr_v4mapped(addr))
2134                 return snprintf(buf, buflen, "::ffff:%pI4",
2135                                         &addr->s6_addr32[3]);
2136
2137         /*
2138          * RFC 4291, Section 2.2.1
2139          */
2140         return snprintf(buf, buflen, "%pI6c", addr);
2141 }
2142
2143 /* Derived from rpc_sockaddr2uaddr */
2144 static void
2145 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2146 {
2147         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2148         char portbuf[RPCBIND_MAXUADDRPLEN];
2149         char addrbuf[RPCBIND_MAXUADDRLEN];
2150         char *netid;
2151         unsigned short port;
2152         int len, netid_len;
2153         __be32 *p;
2154
2155         switch (sap->sa_family) {
2156         case AF_INET:
2157                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2158                         return;
2159                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2160                 netid = "tcp";
2161                 netid_len = 3;
2162                 break;
2163         case AF_INET6:
2164                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2165                         return;
2166                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2167                 netid = "tcp6";
2168                 netid_len = 4;
2169                 break;
2170         default:
2171                 /* we only support tcp and tcp6 */
2172                 WARN_ON_ONCE(1);
2173                 return;
2174         }
2175
2176         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2177         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2178
2179         p = xdr_reserve_space(xdr, 4 + netid_len);
2180         xdr_encode_opaque(p, netid, netid_len);
2181
2182         p = xdr_reserve_space(xdr, 4 + len);
2183         xdr_encode_opaque(p, addrbuf, len);
2184 }
2185
2186 static void
2187 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2188                          ktime_t t)
2189 {
2190         struct timespec64 ts;
2191         __be32 *p;
2192
2193         p = xdr_reserve_space(xdr, 12);
2194         ts = ktime_to_timespec64(t);
2195         p = xdr_encode_hyper(p, ts.tv_sec);
2196         *p++ = cpu_to_be32(ts.tv_nsec);
2197 }
2198
2199 static void
2200 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2201                             struct nfs4_ff_io_stat *stat)
2202 {
2203         __be32 *p;
2204
2205         p = xdr_reserve_space(xdr, 5 * 8);
2206         p = xdr_encode_hyper(p, stat->ops_requested);
2207         p = xdr_encode_hyper(p, stat->bytes_requested);
2208         p = xdr_encode_hyper(p, stat->ops_completed);
2209         p = xdr_encode_hyper(p, stat->bytes_completed);
2210         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2211         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2212         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2213 }
2214
2215 static void
2216 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2217                               const struct nfs42_layoutstat_devinfo *devinfo,
2218                               struct nfs4_ff_layout_mirror *mirror)
2219 {
2220         struct nfs4_pnfs_ds_addr *da;
2221         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2222         struct nfs_fh *fh = &mirror->fh_versions[0];
2223         __be32 *p;
2224
2225         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2226         dprintk("%s: DS %s: encoding address %s\n",
2227                 __func__, ds->ds_remotestr, da->da_remotestr);
2228         /* netaddr4 */
2229         ff_layout_encode_netaddr(xdr, da);
2230         /* nfs_fh4 */
2231         p = xdr_reserve_space(xdr, 4 + fh->size);
2232         xdr_encode_opaque(p, fh->data, fh->size);
2233         /* ff_io_latency4 read */
2234         spin_lock(&mirror->lock);
2235         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2236         /* ff_io_latency4 write */
2237         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2238         spin_unlock(&mirror->lock);
2239         /* nfstime4 */
2240         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2241         /* bool */
2242         p = xdr_reserve_space(xdr, 4);
2243         *p = cpu_to_be32(false);
2244 }
2245
2246 static void
2247 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2248                              const struct nfs4_xdr_opaque_data *opaque)
2249 {
2250         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2251                         struct nfs42_layoutstat_devinfo, ld_private);
2252         __be32 *start;
2253
2254         /* layoutupdate length */
2255         start = xdr_reserve_space(xdr, 4);
2256         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2257
2258         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2259 }
2260
2261 static void
2262 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2263 {
2264         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2265
2266         ff_layout_put_mirror(mirror);
2267 }
2268
2269 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2270         .encode = ff_layout_encode_layoutstats,
2271         .free   = ff_layout_free_layoutstats,
2272 };
2273
2274 static int
2275 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2276                                struct nfs42_layoutstat_devinfo *devinfo,
2277                                int dev_limit)
2278 {
2279         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2280         struct nfs4_ff_layout_mirror *mirror;
2281         struct nfs4_deviceid_node *dev;
2282         int i = 0;
2283
2284         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2285                 if (i >= dev_limit)
2286                         break;
2287                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2288                         continue;
2289                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2290                         continue;
2291                 /* mirror refcount put in cleanup_layoutstats */
2292                 if (!atomic_inc_not_zero(&mirror->ref))
2293                         continue;
2294                 dev = &mirror->mirror_ds->id_node;
2295                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2296                 devinfo->offset = 0;
2297                 devinfo->length = NFS4_MAX_UINT64;
2298                 spin_lock(&mirror->lock);
2299                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2300                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2301                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2302                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2303                 spin_unlock(&mirror->lock);
2304                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2305                 devinfo->ld_private.ops = &layoutstat_ops;
2306                 devinfo->ld_private.data = mirror;
2307
2308                 devinfo++;
2309                 i++;
2310         }
2311         return i;
2312 }
2313
2314 static int
2315 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2316 {
2317         struct nfs4_flexfile_layout *ff_layout;
2318         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2319
2320         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2321         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2322         if (!args->devinfo)
2323                 return -ENOMEM;
2324
2325         spin_lock(&args->inode->i_lock);
2326         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2327         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2328                         &args->devinfo[0], dev_count);
2329         spin_unlock(&args->inode->i_lock);
2330         if (!args->num_dev) {
2331                 kfree(args->devinfo);
2332                 args->devinfo = NULL;
2333                 return -ENOENT;
2334         }
2335
2336         return 0;
2337 }
2338
2339 static int
2340 ff_layout_set_layoutdriver(struct nfs_server *server,
2341                 const struct nfs_fh *dummy)
2342 {
2343 #if IS_ENABLED(CONFIG_NFS_V4_2)
2344         server->caps |= NFS_CAP_LAYOUTSTATS;
2345 #endif
2346         return 0;
2347 }
2348
2349 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2350         .id                     = LAYOUT_FLEX_FILES,
2351         .name                   = "LAYOUT_FLEX_FILES",
2352         .owner                  = THIS_MODULE,
2353         .set_layoutdriver       = ff_layout_set_layoutdriver,
2354         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2355         .free_layout_hdr        = ff_layout_free_layout_hdr,
2356         .alloc_lseg             = ff_layout_alloc_lseg,
2357         .free_lseg              = ff_layout_free_lseg,
2358         .add_lseg               = ff_layout_add_lseg,
2359         .pg_read_ops            = &ff_layout_pg_read_ops,
2360         .pg_write_ops           = &ff_layout_pg_write_ops,
2361         .get_ds_info            = ff_layout_get_ds_info,
2362         .free_deviceid_node     = ff_layout_free_deviceid_node,
2363         .mark_request_commit    = pnfs_layout_mark_request_commit,
2364         .clear_request_commit   = pnfs_generic_clear_request_commit,
2365         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2366         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2367         .commit_pagelist        = ff_layout_commit_pagelist,
2368         .read_pagelist          = ff_layout_read_pagelist,
2369         .write_pagelist         = ff_layout_write_pagelist,
2370         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2371         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2372         .sync                   = pnfs_nfs_generic_sync,
2373         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2374 };
2375
2376 static int __init nfs4flexfilelayout_init(void)
2377 {
2378         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2379                __func__);
2380         if (!ff_zero_group) {
2381                 ff_zero_group = groups_alloc(0);
2382                 if (!ff_zero_group)
2383                         return -ENOMEM;
2384         }
2385         return pnfs_register_layoutdriver(&flexfilelayout_type);
2386 }
2387
2388 static void __exit nfs4flexfilelayout_exit(void)
2389 {
2390         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2391                __func__);
2392         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2393         if (ff_zero_group) {
2394                 put_group_info(ff_zero_group);
2395                 ff_zero_group = NULL;
2396         }
2397 }
2398
2399 MODULE_ALIAS("nfs-layouttype4-4");
2400
2401 MODULE_LICENSE("GPL");
2402 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2403
2404 module_init(nfs4flexfilelayout_init);
2405 module_exit(nfs4flexfilelayout_exit);