fs/nfs/flexfilelayout/flexfilelayout.c

   1 /*
   2  * Module for pnfs flexfile layout driver.
   3  *
   4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   5  *
   6  * Tao Peng <bergwolf@primarydata.com>
   7  */
   8
   9 #include <linux/nfs_fs.h>
  10 #include <linux/nfs_page.h>
  11 #include <linux/module.h>
  12
  13 #include <linux/sunrpc/metrics.h>
  14
  15 #include "flexfilelayout.h"
  16 #include "../nfs4session.h"
  17 #include "../nfs4idmap.h"
  18 #include "../internal.h"
  19 #include "../delegation.h"
  20 #include "../nfs4trace.h"
  21 #include "../iostat.h"
  22 #include "../nfs.h"
  23 #include "../nfs42.h"
  24
  25 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  26
  27 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  28 #define FF_LAYOUTRETURN_MAXERR 20
  29
  30
  31 static struct group_info        *ff_zero_group;
  32
  33 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  34                 struct nfs_pgio_header *hdr);
  35 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  36                                struct nfs42_layoutstat_devinfo *devinfo,
  37                                int dev_limit);
  38 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  39                               const struct nfs42_layoutstat_devinfo *devinfo,
  40                               struct nfs4_ff_layout_mirror *mirror);
  41
  42 static struct pnfs_layout_hdr *
  43 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  44 {
  45         struct nfs4_flexfile_layout *ffl;
  46
  47         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  48         if (ffl) {
  49                 INIT_LIST_HEAD(&ffl->error_list);
  50                 INIT_LIST_HEAD(&ffl->mirrors);
  51                 ffl->last_report_time = ktime_get();
  52                 return &ffl->generic_hdr;
  53         } else
  54                 return NULL;
  55 }
  56
  57 static void
  58 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  59 {
  60         struct nfs4_ff_layout_ds_err *err, *n;
  61
  62         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  63                                  list) {
  64                 list_del(&err->list);
  65                 kfree(err);
  66         }
  67         kfree(FF_LAYOUT_FROM_HDR(lo));
  68 }
  69
  70 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  71 {
  72         __be32 *p;
  73
  74         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  75         if (unlikely(p == NULL))
  76                 return -ENOBUFS;
  77         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  78         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  79         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  80                 p[0], p[1], p[2], p[3]);
  81         return 0;
  82 }
  83
  84 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  85 {
  86         __be32 *p;
  87
  88         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  89         if (unlikely(!p))
  90                 return -ENOBUFS;
  91         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  92         nfs4_print_deviceid(devid);
  93         return 0;
  94 }
  95
  96 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  97 {
  98         __be32 *p;
  99
 100         p = xdr_inline_decode(xdr, 4);
 101         if (unlikely(!p))
 102                 return -ENOBUFS;
 103         fh->size = be32_to_cpup(p++);
 104         if (fh->size > sizeof(struct nfs_fh)) {
 105                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 106                        fh->size);
 107                 return -EOVERFLOW;
 108         }
 109         /* fh.data */
 110         p = xdr_inline_decode(xdr, fh->size);
 111         if (unlikely(!p))
 112                 return -ENOBUFS;
 113         memcpy(&fh->data, p, fh->size);
 114         dprintk("%s: fh len %d\n", __func__, fh->size);
 115
 116         return 0;
 117 }
 118
 119 /*
 120  * Currently only stringified uids and gids are accepted.
 121  * I.e., kerberos is not supported to the DSes, so no pricipals.
 122  *
 123  * That means that one common function will suffice, but when
 124  * principals are added, this should be split to accomodate
 125  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 126  */
 127 static int
 128 decode_name(struct xdr_stream *xdr, u32 *id)
 129 {
 130         __be32 *p;
 131         int len;
 132
 133         /* opaque_length(4)*/
 134         p = xdr_inline_decode(xdr, 4);
 135         if (unlikely(!p))
 136                 return -ENOBUFS;
 137         len = be32_to_cpup(p++);
 138         if (len < 0)
 139                 return -EINVAL;
 140
 141         dprintk("%s: len %u\n", __func__, len);
 142
 143         /* opaque body */
 144         p = xdr_inline_decode(xdr, len);
 145         if (unlikely(!p))
 146                 return -ENOBUFS;
 147
 148         if (!nfs_map_string_to_numeric((char *)p, len, id))
 149                 return -EINVAL;
 150
 151         return 0;
 152 }
 153
 154 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 155                 const struct nfs4_ff_layout_mirror *m2)
 156 {
 157         int i, j;
 158
 159         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 160                 return false;
 161         for (i = 0; i < m1->fh_versions_cnt; i++) {
 162                 bool found_fh = false;
 163                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 164                         if (nfs_compare_fh(&m1->fh_versions[i],
 165                                         &m2->fh_versions[j]) == 0) {
 166                                 found_fh = true;
 167                                 break;
 168                         }
 169                 }
 170                 if (!found_fh)
 171                         return false;
 172         }
 173         return true;
 174 }
 175
 176 static struct nfs4_ff_layout_mirror *
 177 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 178                 struct nfs4_ff_layout_mirror *mirror)
 179 {
 180         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 181         struct nfs4_ff_layout_mirror *pos;
 182         struct inode *inode = lo->plh_inode;
 183
 184         spin_lock(&inode->i_lock);
 185         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 186                 if (mirror->mirror_ds != pos->mirror_ds)
 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 void ff_layout_mark_devices_valid(struct nfs4_ff_layout_segment *fls)
 364 {
 365         struct nfs4_deviceid_node *node;
 366         int i;
 367
 368         if (!(fls->flags & FF_FLAGS_NO_IO_THRU_MDS))
 369                 return;
 370         for (i = 0; i < fls->mirror_array_cnt; i++) {
 371                 node = &fls->mirror_array[i]->mirror_ds->id_node;
 372                 clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
 373         }
 374 }
 375
 376 static struct pnfs_layout_segment *
 377 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 378                      struct nfs4_layoutget_res *lgr,
 379                      gfp_t gfp_flags)
 380 {
 381         struct pnfs_layout_segment *ret;
 382         struct nfs4_ff_layout_segment *fls = NULL;
 383         struct xdr_stream stream;
 384         struct xdr_buf buf;
 385         struct page *scratch;
 386         u64 stripe_unit;
 387         u32 mirror_array_cnt;
 388         __be32 *p;
 389         int i, rc;
 390
 391         dprintk("--> %s\n", __func__);
 392         scratch = alloc_page(gfp_flags);
 393         if (!scratch)
 394                 return ERR_PTR(-ENOMEM);
 395
 396         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 397                               lgr->layoutp->len);
 398         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 399
 400         /* stripe unit and mirror_array_cnt */
 401         rc = -EIO;
 402         p = xdr_inline_decode(&stream, 8 + 4);
 403         if (!p)
 404                 goto out_err_free;
 405
 406         p = xdr_decode_hyper(p, &stripe_unit);
 407         mirror_array_cnt = be32_to_cpup(p++);
 408         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 409                 stripe_unit, mirror_array_cnt);
 410
 411         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 412             mirror_array_cnt == 0)
 413                 goto out_err_free;
 414
 415         rc = -ENOMEM;
 416         fls = kzalloc(sizeof(*fls), gfp_flags);
 417         if (!fls)
 418                 goto out_err_free;
 419
 420         fls->mirror_array_cnt = mirror_array_cnt;
 421         fls->stripe_unit = stripe_unit;
 422         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 423                                     sizeof(fls->mirror_array[0]), gfp_flags);
 424         if (fls->mirror_array == NULL)
 425                 goto out_err_free;
 426
 427         for (i = 0; i < fls->mirror_array_cnt; i++) {
 428                 struct nfs4_ff_layout_mirror *mirror;
 429                 struct nfs4_deviceid devid;
 430                 struct nfs4_deviceid_node *idnode;
 431                 struct auth_cred acred = { .group_info = ff_zero_group };
 432                 struct rpc_cred __rcu *cred;
 433                 u32 ds_count, fh_count, id;
 434                 int j;
 435
 436                 rc = -EIO;
 437                 p = xdr_inline_decode(&stream, 4);
 438                 if (!p)
 439                         goto out_err_free;
 440                 ds_count = be32_to_cpup(p);
 441
 442                 /* FIXME: allow for striping? */
 443                 if (ds_count != 1)
 444                         goto out_err_free;
 445
 446                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 447                 if (fls->mirror_array[i] == NULL) {
 448                         rc = -ENOMEM;
 449                         goto out_err_free;
 450                 }
 451
 452                 fls->mirror_array[i]->ds_count = ds_count;
 453
 454                 /* deviceid */
 455                 rc = decode_deviceid(&stream, &devid);
 456                 if (rc)
 457                         goto out_err_free;
 458
 459                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 460                                                 &devid, lh->plh_lc_cred,
 461                                                 gfp_flags);
 462                 /*
 463                  * upon success, mirror_ds is allocated by previous
 464                  * getdeviceinfo, or newly by .alloc_deviceid_node
 465                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 466                  */
 467                 if (idnode)
 468                         fls->mirror_array[i]->mirror_ds =
 469                                 FF_LAYOUT_MIRROR_DS(idnode);
 470                 else
 471                         goto out_err_free;
 472
 473                 /* efficiency */
 474                 rc = -EIO;
 475                 p = xdr_inline_decode(&stream, 4);
 476                 if (!p)
 477                         goto out_err_free;
 478                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 479
 480                 /* stateid */
 481                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 482                 if (rc)
 483                         goto out_err_free;
 484
 485                 /* fh */
 486                 p = xdr_inline_decode(&stream, 4);
 487                 if (!p)
 488                         goto out_err_free;
 489                 fh_count = be32_to_cpup(p);
 490
 491                 fls->mirror_array[i]->fh_versions =
 492                         kzalloc(fh_count * sizeof(struct nfs_fh),
 493                                 gfp_flags);
 494                 if (fls->mirror_array[i]->fh_versions == NULL) {
 495                         rc = -ENOMEM;
 496                         goto out_err_free;
 497                 }
 498
 499                 for (j = 0; j < fh_count; j++) {
 500                         rc = decode_nfs_fh(&stream,
 501                                            &fls->mirror_array[i]->fh_versions[j]);
 502                         if (rc)
 503                                 goto out_err_free;
 504                 }
 505
 506                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 507
 508                 /* user */
 509                 rc = decode_name(&stream, &id);
 510                 if (rc)
 511                         goto out_err_free;
 512
 513                 acred.uid = make_kuid(&init_user_ns, id);
 514
 515                 /* group */
 516                 rc = decode_name(&stream, &id);
 517                 if (rc)
 518                         goto out_err_free;
 519
 520                 acred.gid = make_kgid(&init_user_ns, id);
 521
 522                 /* find the cred for it */
 523                 rcu_assign_pointer(cred, rpc_lookup_generic_cred(&acred, 0, gfp_flags));
 524                 if (IS_ERR(cred)) {
 525                         rc = PTR_ERR(cred);
 526                         goto out_err_free;
 527                 }
 528
 529                 if (lgr->range.iomode == IOMODE_READ)
 530                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 531                 else
 532                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 533
 534                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 535                 if (mirror != fls->mirror_array[i]) {
 536                         /* swap cred ptrs so free_mirror will clean up old */
 537                         if (lgr->range.iomode == IOMODE_READ) {
 538                                 cred = xchg(&mirror->ro_cred, cred);
 539                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 540                         } else {
 541                                 cred = xchg(&mirror->rw_cred, cred);
 542                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 543                         }
 544                         ff_layout_free_mirror(fls->mirror_array[i]);
 545                         fls->mirror_array[i] = mirror;
 546                 }
 547
 548                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 549                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 550                         from_kuid(&init_user_ns, acred.uid),
 551                         from_kgid(&init_user_ns, acred.gid));
 552         }
 553
 554         p = xdr_inline_decode(&stream, 4);
 555         if (!p)
 556                 goto out_sort_mirrors;
 557         fls->flags = be32_to_cpup(p);
 558
 559         p = xdr_inline_decode(&stream, 4);
 560         if (!p)
 561                 goto out_sort_mirrors;
 562         for (i=0; i < fls->mirror_array_cnt; i++)
 563                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 564
 565 out_sort_mirrors:
 566         ff_layout_sort_mirrors(fls);
 567         rc = ff_layout_check_layout(lgr);
 568         if (rc)
 569                 goto out_err_free;
 570         ff_layout_mark_devices_valid(fls);
 571
 572         ret = &fls->generic_hdr;
 573         dprintk("<-- %s (success)\n", __func__);
 574 out_free_page:
 575         __free_page(scratch);
 576         return ret;
 577 out_err_free:
 578         _ff_layout_free_lseg(fls);
 579         ret = ERR_PTR(rc);
 580         dprintk("<-- %s (%d)\n", __func__, rc);
 581         goto out_free_page;
 582 }
 583
 584 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 585 {
 586         struct pnfs_layout_segment *lseg;
 587
 588         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 589                 if (lseg->pls_range.iomode == IOMODE_RW)
 590                         return true;
 591
 592         return false;
 593 }
 594
 595 static void
 596 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 597 {
 598         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 599
 600         dprintk("--> %s\n", __func__);
 601
 602         if (lseg->pls_range.iomode == IOMODE_RW) {
 603                 struct nfs4_flexfile_layout *ffl;
 604                 struct inode *inode;
 605
 606                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 607                 inode = ffl->generic_hdr.plh_inode;
 608                 spin_lock(&inode->i_lock);
 609                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 610                         ffl->commit_info.nbuckets = 0;
 611                         kfree(ffl->commit_info.buckets);
 612                         ffl->commit_info.buckets = NULL;
 613                 }
 614                 spin_unlock(&inode->i_lock);
 615         }
 616         _ff_layout_free_lseg(fls);
 617 }
 618
 619 /* Return 1 until we have multiple lsegs support */
 620 static int
 621 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 622 {
 623         return 1;
 624 }
 625
 626 static void
 627 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 628 {
 629         /* first IO request? */
 630         if (atomic_inc_return(&timer->n_ops) == 1) {
 631                 timer->start_time = now;
 632         }
 633 }
 634
 635 static ktime_t
 636 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 637 {
 638         ktime_t start;
 639
 640         if (atomic_dec_return(&timer->n_ops) < 0)
 641                 WARN_ON_ONCE(1);
 642
 643         start = timer->start_time;
 644         timer->start_time = now;
 645         return ktime_sub(now, start);
 646 }
 647
 648 static bool
 649 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 650                             struct nfs4_ff_layoutstat *layoutstat,
 651                             ktime_t now)
 652 {
 653         static const ktime_t notime = {0};
 654         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 655         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 656
 657         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 658         if (ktime_equal(mirror->start_time, notime))
 659                 mirror->start_time = now;
 660         if (mirror->report_interval != 0)
 661                 report_interval = (s64)mirror->report_interval * 1000LL;
 662         else if (layoutstats_timer != 0)
 663                 report_interval = (s64)layoutstats_timer * 1000LL;
 664         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 665                         report_interval) {
 666                 ffl->last_report_time = now;
 667                 return true;
 668         }
 669
 670         return false;
 671 }
 672
 673 static void
 674 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 675                 __u64 requested)
 676 {
 677         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 678
 679         iostat->ops_requested++;
 680         iostat->bytes_requested += requested;
 681 }
 682
 683 static void
 684 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 685                 __u64 requested,
 686                 __u64 completed,
 687                 ktime_t time_completed,
 688                 ktime_t time_started)
 689 {
 690         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 691         ktime_t completion_time = ktime_sub(time_completed, time_started);
 692         ktime_t timer;
 693
 694         iostat->ops_completed++;
 695         iostat->bytes_completed += completed;
 696         iostat->bytes_not_delivered += requested - completed;
 697
 698         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 699         iostat->total_busy_time =
 700                         ktime_add(iostat->total_busy_time, timer);
 701         iostat->aggregate_completion_time =
 702                         ktime_add(iostat->aggregate_completion_time,
 703                                         completion_time);
 704 }
 705
 706 static void
 707 nfs4_ff_layout_stat_io_start_read(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->read_stat, now);
 715         nfs4_ff_layout_stat_io_update_requested(&mirror->read_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_KERNEL);
 721 }
 722
 723 static void
 724 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 725                 struct nfs4_ff_layout_mirror *mirror,
 726                 __u64 requested,
 727                 __u64 completed)
 728 {
 729         spin_lock(&mirror->lock);
 730         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 731                         requested, completed,
 732                         ktime_get(), task->tk_start);
 733         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 734         spin_unlock(&mirror->lock);
 735 }
 736
 737 static void
 738 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 739                 struct nfs4_ff_layout_mirror *mirror,
 740                 __u64 requested, ktime_t now)
 741 {
 742         bool report;
 743
 744         spin_lock(&mirror->lock);
 745         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 746         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 747         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 748         spin_unlock(&mirror->lock);
 749
 750         if (report)
 751                 pnfs_report_layoutstat(inode, GFP_NOIO);
 752 }
 753
 754 static void
 755 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 756                 struct nfs4_ff_layout_mirror *mirror,
 757                 __u64 requested,
 758                 __u64 completed,
 759                 enum nfs3_stable_how committed)
 760 {
 761         if (committed == NFS_UNSTABLE)
 762                 requested = completed = 0;
 763
 764         spin_lock(&mirror->lock);
 765         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 766                         requested, completed, ktime_get(), task->tk_start);
 767         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 768         spin_unlock(&mirror->lock);
 769 }
 770
 771 static int
 772 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 773                             struct nfs_commit_info *cinfo,
 774                             gfp_t gfp_flags)
 775 {
 776         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 777         struct pnfs_commit_bucket *buckets;
 778         int size;
 779
 780         if (cinfo->ds->nbuckets != 0) {
 781                 /* This assumes there is only one RW lseg per file.
 782                  * To support multiple lseg per file, we need to
 783                  * change struct pnfs_commit_bucket to allow dynamic
 784                  * increasing nbuckets.
 785                  */
 786                 return 0;
 787         }
 788
 789         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 790
 791         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 792                           gfp_flags);
 793         if (!buckets)
 794                 return -ENOMEM;
 795         else {
 796                 int i;
 797
 798                 spin_lock(&cinfo->inode->i_lock);
 799                 if (cinfo->ds->nbuckets != 0)
 800                         kfree(buckets);
 801                 else {
 802                         cinfo->ds->buckets = buckets;
 803                         cinfo->ds->nbuckets = size;
 804                         for (i = 0; i < size; i++) {
 805                                 INIT_LIST_HEAD(&buckets[i].written);
 806                                 INIT_LIST_HEAD(&buckets[i].committing);
 807                                 /* mark direct verifier as unset */
 808                                 buckets[i].direct_verf.committed =
 809                                         NFS_INVALID_STABLE_HOW;
 810                         }
 811                 }
 812                 spin_unlock(&cinfo->inode->i_lock);
 813                 return 0;
 814         }
 815 }
 816
 817 static struct nfs4_pnfs_ds *
 818 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 819                                   int start_idx,
 820                                   int *best_idx)
 821 {
 822         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 823         struct nfs4_pnfs_ds *ds;
 824         bool fail_return = false;
 825         int idx;
 826
 827         /* mirrors are sorted by efficiency */
 828         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 829                 if (idx+1 == fls->mirror_array_cnt)
 830                         fail_return = true;
 831                 ds = nfs4_ff_layout_prepare_ds(lseg, idx, fail_return);
 832                 if (ds) {
 833                         *best_idx = idx;
 834                         return ds;
 835                 }
 836         }
 837
 838         return NULL;
 839 }
 840
 841 static void
 842 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 843                       struct nfs_page *req,
 844                       bool strict_iomode)
 845 {
 846 retry_strict:
 847         pnfs_put_lseg(pgio->pg_lseg);
 848         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 849                                            req->wb_context,
 850                                            0,
 851                                            NFS4_MAX_UINT64,
 852                                            IOMODE_READ,
 853                                            strict_iomode,
 854                                            GFP_KERNEL);
 855         if (IS_ERR(pgio->pg_lseg)) {
 856                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 857                 pgio->pg_lseg = NULL;
 858         }
 859
 860         /* If we don't have checking, do get a IOMODE_RW
 861          * segment, and the server wants to avoid READs
 862          * there, then retry!
 863          */
 864         if (pgio->pg_lseg && !strict_iomode &&
 865             ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 866                 strict_iomode = true;
 867                 goto retry_strict;
 868         }
 869 }
 870
 871 static void
 872 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 873                         struct nfs_page *req)
 874 {
 875         struct nfs_pgio_mirror *pgm;
 876         struct nfs4_ff_layout_mirror *mirror;
 877         struct nfs4_pnfs_ds *ds;
 878         int ds_idx;
 879
 880 retry:
 881         /* Use full layout for now */
 882         if (!pgio->pg_lseg)
 883                 ff_layout_pg_get_read(pgio, req, false);
 884         else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg))
 885                 ff_layout_pg_get_read(pgio, req, true);
 886
 887         /* If no lseg, fall back to read through mds */
 888         if (pgio->pg_lseg == NULL)
 889                 goto out_mds;
 890
 891         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 892         if (!ds) {
 893                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 894                         goto out_mds;
 895                 pnfs_put_lseg(pgio->pg_lseg);
 896                 pgio->pg_lseg = NULL;
 897                 /* Sleep for 1 second before retrying */
 898                 ssleep(1);
 899                 goto retry;
 900         }
 901
 902         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 903
 904         pgio->pg_mirror_idx = ds_idx;
 905
 906         /* read always uses only one mirror - idx 0 for pgio layer */
 907         pgm = &pgio->pg_mirrors[0];
 908         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 909
 910         return;
 911 out_mds:
 912         pnfs_put_lseg(pgio->pg_lseg);
 913         pgio->pg_lseg = NULL;
 914         nfs_pageio_reset_read_mds(pgio);
 915 }
 916
 917 static void
 918 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 919                         struct nfs_page *req)
 920 {
 921         struct nfs4_ff_layout_mirror *mirror;
 922         struct nfs_pgio_mirror *pgm;
 923         struct nfs_commit_info cinfo;
 924         struct nfs4_pnfs_ds *ds;
 925         int i;
 926         int status;
 927
 928 retry:
 929         if (!pgio->pg_lseg) {
 930                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 931                                                    req->wb_context,
 932                                                    0,
 933                                                    NFS4_MAX_UINT64,
 934                                                    IOMODE_RW,
 935                                                    false,
 936                                                    GFP_NOFS);
 937                 if (IS_ERR(pgio->pg_lseg)) {
 938                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 939                         pgio->pg_lseg = NULL;
 940                         return;
 941                 }
 942         }
 943         /* If no lseg, fall back to write through mds */
 944         if (pgio->pg_lseg == NULL)
 945                 goto out_mds;
 946
 947         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 948         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 949         if (status < 0)
 950                 goto out_mds;
 951
 952         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 953         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 954             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 955                 goto out_mds;
 956
 957         for (i = 0; i < pgio->pg_mirror_count; i++) {
 958                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 959                 if (!ds) {
 960                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 961                                 goto out_mds;
 962                         pnfs_put_lseg(pgio->pg_lseg);
 963                         pgio->pg_lseg = NULL;
 964                         /* Sleep for 1 second before retrying */
 965                         ssleep(1);
 966                         goto retry;
 967                 }
 968                 pgm = &pgio->pg_mirrors[i];
 969                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 970                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 971         }
 972
 973         return;
 974
 975 out_mds:
 976         pnfs_put_lseg(pgio->pg_lseg);
 977         pgio->pg_lseg = NULL;
 978         nfs_pageio_reset_write_mds(pgio);
 979 }
 980
 981 static unsigned int
 982 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 983                                     struct nfs_page *req)
 984 {
 985         if (!pgio->pg_lseg) {
 986                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 987                                                    req->wb_context,
 988                                                    0,
 989                                                    NFS4_MAX_UINT64,
 990                                                    IOMODE_RW,
 991                                                    false,
 992                                                    GFP_NOFS);
 993                 if (IS_ERR(pgio->pg_lseg)) {
 994                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 995                         pgio->pg_lseg = NULL;
 996                         goto out;
 997                 }
 998         }
 999         if (pgio->pg_lseg)
1000                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
1001
1002         /* no lseg means that pnfs is not in use, so no mirroring here */
1003         nfs_pageio_reset_write_mds(pgio);
1004 out:
1005         return 1;
1006 }
1007
1008 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
1009         .pg_init = ff_layout_pg_init_read,
1010         .pg_test = pnfs_generic_pg_test,
1011         .pg_doio = pnfs_generic_pg_readpages,
1012         .pg_cleanup = pnfs_generic_pg_cleanup,
1013 };
1014
1015 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
1016         .pg_init = ff_layout_pg_init_write,
1017         .pg_test = pnfs_generic_pg_test,
1018         .pg_doio = pnfs_generic_pg_writepages,
1019         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1020         .pg_cleanup = pnfs_generic_pg_cleanup,
1021 };
1022
1023 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1024 {
1025         struct rpc_task *task = &hdr->task;
1026
1027         pnfs_layoutcommit_inode(hdr->inode, false);
1028
1029         if (retry_pnfs) {
1030                 dprintk("%s Reset task %5u for i/o through pNFS "
1031                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1032                         hdr->task.tk_pid,
1033                         hdr->inode->i_sb->s_id,
1034                         (unsigned long long)NFS_FILEID(hdr->inode),
1035                         hdr->args.count,
1036                         (unsigned long long)hdr->args.offset);
1037
1038                 hdr->completion_ops->reschedule_io(hdr);
1039                 return;
1040         }
1041
1042         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1043                 dprintk("%s Reset task %5u for i/o through MDS "
1044                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1045                         hdr->task.tk_pid,
1046                         hdr->inode->i_sb->s_id,
1047                         (unsigned long long)NFS_FILEID(hdr->inode),
1048                         hdr->args.count,
1049                         (unsigned long long)hdr->args.offset);
1050
1051                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1052         }
1053 }
1054
1055 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1056 {
1057         struct rpc_task *task = &hdr->task;
1058
1059         pnfs_layoutcommit_inode(hdr->inode, false);
1060
1061         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1062                 dprintk("%s Reset task %5u for i/o through MDS "
1063                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1064                         hdr->task.tk_pid,
1065                         hdr->inode->i_sb->s_id,
1066                         (unsigned long long)NFS_FILEID(hdr->inode),
1067                         hdr->args.count,
1068                         (unsigned long long)hdr->args.offset);
1069
1070                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1071         }
1072 }
1073
1074 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1075                                            struct nfs4_state *state,
1076                                            struct nfs_client *clp,
1077                                            struct pnfs_layout_segment *lseg,
1078                                            int idx)
1079 {
1080         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1081         struct inode *inode = lo->plh_inode;
1082         struct nfs_server *mds_server = NFS_SERVER(inode);
1083
1084         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1085         struct nfs_client *mds_client = mds_server->nfs_client;
1086         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1087
1088         if (task->tk_status >= 0)
1089                 return 0;
1090
1091         switch (task->tk_status) {
1092         /* MDS state errors */
1093         case -NFS4ERR_DELEG_REVOKED:
1094         case -NFS4ERR_ADMIN_REVOKED:
1095         case -NFS4ERR_BAD_STATEID:
1096                 if (state == NULL)
1097                         break;
1098                 nfs_remove_bad_delegation(state->inode, NULL);
1099         case -NFS4ERR_OPENMODE:
1100                 if (state == NULL)
1101                         break;
1102                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1103                         goto out_bad_stateid;
1104                 goto wait_on_recovery;
1105         case -NFS4ERR_EXPIRED:
1106                 if (state != NULL) {
1107                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1108                                 goto out_bad_stateid;
1109                 }
1110                 nfs4_schedule_lease_recovery(mds_client);
1111                 goto wait_on_recovery;
1112         /* DS session errors */
1113         case -NFS4ERR_BADSESSION:
1114         case -NFS4ERR_BADSLOT:
1115         case -NFS4ERR_BAD_HIGH_SLOT:
1116         case -NFS4ERR_DEADSESSION:
1117         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1118         case -NFS4ERR_SEQ_FALSE_RETRY:
1119         case -NFS4ERR_SEQ_MISORDERED:
1120                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1121                         "flags 0x%x\n", __func__, task->tk_status,
1122                         clp->cl_exchange_flags);
1123                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1124                 break;
1125         case -NFS4ERR_DELAY:
1126         case -NFS4ERR_GRACE:
1127                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1128                 break;
1129         case -NFS4ERR_RETRY_UNCACHED_REP:
1130                 break;
1131         /* Invalidate Layout errors */
1132         case -NFS4ERR_PNFS_NO_LAYOUT:
1133         case -ESTALE:           /* mapped NFS4ERR_STALE */
1134         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1135         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1136         case -NFS4ERR_FHEXPIRED:
1137         case -NFS4ERR_WRONG_TYPE:
1138                 dprintk("%s Invalid layout error %d\n", __func__,
1139                         task->tk_status);
1140                 /*
1141                  * Destroy layout so new i/o will get a new layout.
1142                  * Layout will not be destroyed until all current lseg
1143                  * references are put. Mark layout as invalid to resend failed
1144                  * i/o and all i/o waiting on the slot table to the MDS until
1145                  * layout is destroyed and a new valid layout is obtained.
1146                  */
1147                 pnfs_destroy_layout(NFS_I(inode));
1148                 rpc_wake_up(&tbl->slot_tbl_waitq);
1149                 goto reset;
1150         /* RPC connection errors */
1151         case -ECONNREFUSED:
1152         case -EHOSTDOWN:
1153         case -EHOSTUNREACH:
1154         case -ENETUNREACH:
1155         case -EIO:
1156         case -ETIMEDOUT:
1157         case -EPIPE:
1158                 dprintk("%s DS connection error %d\n", __func__,
1159                         task->tk_status);
1160                 nfs4_mark_deviceid_unavailable(devid);
1161                 rpc_wake_up(&tbl->slot_tbl_waitq);
1162                 /* fall through */
1163         default:
1164                 if (ff_layout_avoid_mds_available_ds(lseg))
1165                         return -NFS4ERR_RESET_TO_PNFS;
1166 reset:
1167                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1168                         task->tk_status);
1169                 return -NFS4ERR_RESET_TO_MDS;
1170         }
1171 out:
1172         task->tk_status = 0;
1173         return -EAGAIN;
1174 out_bad_stateid:
1175         task->tk_status = -EIO;
1176         return 0;
1177 wait_on_recovery:
1178         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
1179         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
1180                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
1181         goto out;
1182 }
1183
1184 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1185 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1186                                            struct pnfs_layout_segment *lseg,
1187                                            int idx)
1188 {
1189         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1190
1191         if (task->tk_status >= 0)
1192                 return 0;
1193
1194         switch (task->tk_status) {
1195         /* File access problems. Don't mark the device as unavailable */
1196         case -EACCES:
1197         case -ESTALE:
1198         case -EISDIR:
1199         case -EBADHANDLE:
1200         case -ELOOP:
1201         case -ENOSPC:
1202                 break;
1203         case -EJUKEBOX:
1204                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1205                 goto out_retry;
1206         default:
1207                 dprintk("%s DS connection error %d\n", __func__,
1208                         task->tk_status);
1209                 nfs4_mark_deviceid_unavailable(devid);
1210         }
1211         /* FIXME: Need to prevent infinite looping here. */
1212         return -NFS4ERR_RESET_TO_PNFS;
1213 out_retry:
1214         task->tk_status = 0;
1215         rpc_restart_call_prepare(task);
1216         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1217         return -EAGAIN;
1218 }
1219
1220 static int ff_layout_async_handle_error(struct rpc_task *task,
1221                                         struct nfs4_state *state,
1222                                         struct nfs_client *clp,
1223                                         struct pnfs_layout_segment *lseg,
1224                                         int idx)
1225 {
1226         int vers = clp->cl_nfs_mod->rpc_vers->number;
1227
1228         switch (vers) {
1229         case 3:
1230                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1231         case 4:
1232                 return ff_layout_async_handle_error_v4(task, state, clp,
1233                                                        lseg, idx);
1234         default:
1235                 /* should never happen */
1236                 WARN_ON_ONCE(1);
1237                 return 0;
1238         }
1239 }
1240
1241 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1242                                         int idx, u64 offset, u64 length,
1243                                         u32 status, int opnum, int error)
1244 {
1245         struct nfs4_ff_layout_mirror *mirror;
1246         int err;
1247
1248         if (status == 0) {
1249                 switch (error) {
1250                 case -ETIMEDOUT:
1251                 case -EPFNOSUPPORT:
1252                 case -EPROTONOSUPPORT:
1253                 case -EOPNOTSUPP:
1254                 case -ECONNREFUSED:
1255                 case -ECONNRESET:
1256                 case -EHOSTDOWN:
1257                 case -EHOSTUNREACH:
1258                 case -ENETUNREACH:
1259                 case -EADDRINUSE:
1260                 case -ENOBUFS:
1261                 case -EPIPE:
1262                 case -EPERM:
1263                         status = NFS4ERR_NXIO;
1264                         break;
1265                 case -EACCES:
1266                         status = NFS4ERR_ACCESS;
1267                         break;
1268                 default:
1269                         return;
1270                 }
1271         }
1272
1273         switch (status) {
1274         case NFS4ERR_DELAY:
1275         case NFS4ERR_GRACE:
1276                 return;
1277         default:
1278                 break;
1279         }
1280
1281         mirror = FF_LAYOUT_COMP(lseg, idx);
1282         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1283                                        mirror, offset, length, status, opnum,
1284                                        GFP_NOIO);
1285         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1286         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1287 }
1288
1289 /* NFS_PROTO call done callback routines */
1290 static int ff_layout_read_done_cb(struct rpc_task *task,
1291                                 struct nfs_pgio_header *hdr)
1292 {
1293         int err;
1294
1295         trace_nfs4_pnfs_read(hdr, task->tk_status);
1296         if (task->tk_status < 0)
1297                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1298                                             hdr->args.offset, hdr->args.count,
1299                                             hdr->res.op_status, OP_READ,
1300                                             task->tk_status);
1301         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1302                                            hdr->ds_clp, hdr->lseg,
1303                                            hdr->pgio_mirror_idx);
1304
1305         switch (err) {
1306         case -NFS4ERR_RESET_TO_PNFS:
1307                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1308                                         hdr->pgio_mirror_idx + 1,
1309                                         &hdr->pgio_mirror_idx))
1310                         goto out_eagain;
1311                 ff_layout_read_record_layoutstats_done(task, hdr);
1312                 pnfs_read_resend_pnfs(hdr);
1313                 return task->tk_status;
1314         case -NFS4ERR_RESET_TO_MDS:
1315                 ff_layout_reset_read(hdr);
1316                 return task->tk_status;
1317         case -EAGAIN:
1318                 goto out_eagain;
1319         }
1320
1321         return 0;
1322 out_eagain:
1323         rpc_restart_call_prepare(task);
1324         return -EAGAIN;
1325 }
1326
1327 static bool
1328 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1329 {
1330         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1331 }
1332
1333 /*
1334  * We reference the rpc_cred of the first WRITE that triggers the need for
1335  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1336  * rfc5661 is not clear about which credential should be used.
1337  *
1338  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1339  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1340  * we always send layoutcommit after DS writes.
1341  */
1342 static void
1343 ff_layout_set_layoutcommit(struct inode *inode,
1344                 struct pnfs_layout_segment *lseg,
1345                 loff_t end_offset)
1346 {
1347         if (!ff_layout_need_layoutcommit(lseg))
1348                 return;
1349
1350         pnfs_set_layoutcommit(inode, lseg, end_offset);
1351         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1352                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1353 }
1354
1355 static bool
1356 ff_layout_device_unavailable(struct pnfs_layout_segment *lseg, int idx)
1357 {
1358         /* No mirroring for now */
1359         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1360
1361         return ff_layout_test_devid_unavailable(node);
1362 }
1363
1364 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1365                 struct nfs_pgio_header *hdr)
1366 {
1367         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1368                 return;
1369         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1370                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1371                         hdr->args.count,
1372                         task->tk_start);
1373 }
1374
1375 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1376                 struct nfs_pgio_header *hdr)
1377 {
1378         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1379                 return;
1380         nfs4_ff_layout_stat_io_end_read(task,
1381                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1382                         hdr->args.count,
1383                         hdr->res.count);
1384 }
1385
1386 static int ff_layout_read_prepare_common(struct rpc_task *task,
1387                                          struct nfs_pgio_header *hdr)
1388 {
1389         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1390                 rpc_exit(task, -EIO);
1391                 return -EIO;
1392         }
1393         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1394                 rpc_exit(task, -EHOSTDOWN);
1395                 return -EAGAIN;
1396         }
1397
1398         ff_layout_read_record_layoutstats_start(task, hdr);
1399         return 0;
1400 }
1401
1402 /*
1403  * Call ops for the async read/write cases
1404  * In the case of dense layouts, the offset needs to be reset to its
1405  * original value.
1406  */
1407 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1408 {
1409         struct nfs_pgio_header *hdr = data;
1410
1411         if (ff_layout_read_prepare_common(task, hdr))
1412                 return;
1413
1414         rpc_call_start(task);
1415 }
1416
1417 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1418                                     struct nfs4_sequence_args *args,
1419                                     struct nfs4_sequence_res *res,
1420                                     struct rpc_task *task)
1421 {
1422         if (ds_clp->cl_session)
1423                 return nfs41_setup_sequence(ds_clp->cl_session,
1424                                            args,
1425                                            res,
1426                                            task);
1427         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1428                                    args,
1429                                    res,
1430                                    task);
1431 }
1432
1433 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1434 {
1435         struct nfs_pgio_header *hdr = data;
1436
1437         if (ff_layout_setup_sequence(hdr->ds_clp,
1438                                      &hdr->args.seq_args,
1439                                      &hdr->res.seq_res,
1440                                      task))
1441                 return;
1442
1443         if (ff_layout_read_prepare_common(task, hdr))
1444                 return;
1445
1446         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1447                         hdr->args.lock_context, FMODE_READ) == -EIO)
1448                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1449 }
1450
1451 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1452 {
1453         struct nfs_pgio_header *hdr = data;
1454
1455         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1456
1457         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1458             task->tk_status == 0) {
1459                 nfs4_sequence_done(task, &hdr->res.seq_res);
1460                 return;
1461         }
1462
1463         /* Note this may cause RPC to be resent */
1464         hdr->mds_ops->rpc_call_done(task, hdr);
1465 }
1466
1467 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1468 {
1469         struct nfs_pgio_header *hdr = data;
1470
1471         ff_layout_read_record_layoutstats_done(task, hdr);
1472         rpc_count_iostats_metrics(task,
1473             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1474 }
1475
1476 static void ff_layout_read_release(void *data)
1477 {
1478         struct nfs_pgio_header *hdr = data;
1479
1480         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1481         pnfs_generic_rw_release(data);
1482 }
1483
1484
1485 static int ff_layout_write_done_cb(struct rpc_task *task,
1486                                 struct nfs_pgio_header *hdr)
1487 {
1488         loff_t end_offs = 0;
1489         int err;
1490
1491         trace_nfs4_pnfs_write(hdr, task->tk_status);
1492         if (task->tk_status < 0)
1493                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1494                                             hdr->args.offset, hdr->args.count,
1495                                             hdr->res.op_status, OP_WRITE,
1496                                             task->tk_status);
1497         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1498                                            hdr->ds_clp, hdr->lseg,
1499                                            hdr->pgio_mirror_idx);
1500
1501         switch (err) {
1502         case -NFS4ERR_RESET_TO_PNFS:
1503                 ff_layout_reset_write(hdr, true);
1504                 return task->tk_status;
1505         case -NFS4ERR_RESET_TO_MDS:
1506                 ff_layout_reset_write(hdr, false);
1507                 return task->tk_status;
1508         case -EAGAIN:
1509                 return -EAGAIN;
1510         }
1511
1512         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1513             hdr->res.verf->committed == NFS_DATA_SYNC)
1514                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1515
1516         /* Note: if the write is unstable, don't set end_offs until commit */
1517         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1518
1519         /* zero out fattr since we don't care DS attr at all */
1520         hdr->fattr.valid = 0;
1521         if (task->tk_status >= 0)
1522                 nfs_writeback_update_inode(hdr);
1523
1524         return 0;
1525 }
1526
1527 static int ff_layout_commit_done_cb(struct rpc_task *task,
1528                                      struct nfs_commit_data *data)
1529 {
1530         int err;
1531
1532         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1533         if (task->tk_status < 0)
1534                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1535                                             data->args.offset, data->args.count,
1536                                             data->res.op_status, OP_COMMIT,
1537                                             task->tk_status);
1538         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1539                                            data->lseg, data->ds_commit_index);
1540
1541         switch (err) {
1542         case -NFS4ERR_RESET_TO_PNFS:
1543                 pnfs_generic_prepare_to_resend_writes(data);
1544                 return -EAGAIN;
1545         case -NFS4ERR_RESET_TO_MDS:
1546                 pnfs_generic_prepare_to_resend_writes(data);
1547                 return -EAGAIN;
1548         case -EAGAIN:
1549                 rpc_restart_call_prepare(task);
1550                 return -EAGAIN;
1551         }
1552
1553         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1554
1555         return 0;
1556 }
1557
1558 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1559                 struct nfs_pgio_header *hdr)
1560 {
1561         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1562                 return;
1563         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1564                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1565                         hdr->args.count,
1566                         task->tk_start);
1567 }
1568
1569 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1570                 struct nfs_pgio_header *hdr)
1571 {
1572         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1573                 return;
1574         nfs4_ff_layout_stat_io_end_write(task,
1575                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1576                         hdr->args.count, hdr->res.count,
1577                         hdr->res.verf->committed);
1578 }
1579
1580 static int ff_layout_write_prepare_common(struct rpc_task *task,
1581                                           struct nfs_pgio_header *hdr)
1582 {
1583         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1584                 rpc_exit(task, -EIO);
1585                 return -EIO;
1586         }
1587
1588         if (ff_layout_device_unavailable(hdr->lseg, hdr->pgio_mirror_idx)) {
1589                 rpc_exit(task, -EHOSTDOWN);
1590                 return -EAGAIN;
1591         }
1592
1593         ff_layout_write_record_layoutstats_start(task, hdr);
1594         return 0;
1595 }
1596
1597 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1598 {
1599         struct nfs_pgio_header *hdr = data;
1600
1601         if (ff_layout_write_prepare_common(task, hdr))
1602                 return;
1603
1604         rpc_call_start(task);
1605 }
1606
1607 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1608 {
1609         struct nfs_pgio_header *hdr = data;
1610
1611         if (ff_layout_setup_sequence(hdr->ds_clp,
1612                                      &hdr->args.seq_args,
1613                                      &hdr->res.seq_res,
1614                                      task))
1615                 return;
1616
1617         if (ff_layout_write_prepare_common(task, hdr))
1618                 return;
1619
1620         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1621                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1622                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1623 }
1624
1625 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1626 {
1627         struct nfs_pgio_header *hdr = data;
1628
1629         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1630             task->tk_status == 0) {
1631                 nfs4_sequence_done(task, &hdr->res.seq_res);
1632                 return;
1633         }
1634
1635         /* Note this may cause RPC to be resent */
1636         hdr->mds_ops->rpc_call_done(task, hdr);
1637 }
1638
1639 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1640 {
1641         struct nfs_pgio_header *hdr = data;
1642
1643         ff_layout_write_record_layoutstats_done(task, hdr);
1644         rpc_count_iostats_metrics(task,
1645             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1646 }
1647
1648 static void ff_layout_write_release(void *data)
1649 {
1650         struct nfs_pgio_header *hdr = data;
1651
1652         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1653         pnfs_generic_rw_release(data);
1654 }
1655
1656 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1657                 struct nfs_commit_data *cdata)
1658 {
1659         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1660                 return;
1661         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1662                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1663                         0, task->tk_start);
1664 }
1665
1666 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1667                 struct nfs_commit_data *cdata)
1668 {
1669         struct nfs_page *req;
1670         __u64 count = 0;
1671
1672         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1673                 return;
1674
1675         if (task->tk_status == 0) {
1676                 list_for_each_entry(req, &cdata->pages, wb_list)
1677                         count += req->wb_bytes;
1678         }
1679         nfs4_ff_layout_stat_io_end_write(task,
1680                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1681                         count, count, NFS_FILE_SYNC);
1682 }
1683
1684 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1685                 struct nfs_commit_data *cdata)
1686 {
1687         ff_layout_commit_record_layoutstats_start(task, cdata);
1688 }
1689
1690 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1691 {
1692         ff_layout_commit_prepare_common(task, data);
1693         rpc_call_start(task);
1694 }
1695
1696 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1697 {
1698         struct nfs_commit_data *wdata = data;
1699
1700         if (ff_layout_setup_sequence(wdata->ds_clp,
1701                                  &wdata->args.seq_args,
1702                                  &wdata->res.seq_res,
1703                                  task))
1704                 return;
1705         ff_layout_commit_prepare_common(task, data);
1706 }
1707
1708 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1709 {
1710         pnfs_generic_write_commit_done(task, data);
1711 }
1712
1713 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1714 {
1715         struct nfs_commit_data *cdata = data;
1716
1717         ff_layout_commit_record_layoutstats_done(task, cdata);
1718         rpc_count_iostats_metrics(task,
1719             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1720 }
1721
1722 static void ff_layout_commit_release(void *data)
1723 {
1724         struct nfs_commit_data *cdata = data;
1725
1726         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1727         pnfs_generic_commit_release(data);
1728 }
1729
1730 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1731         .rpc_call_prepare = ff_layout_read_prepare_v3,
1732         .rpc_call_done = ff_layout_read_call_done,
1733         .rpc_count_stats = ff_layout_read_count_stats,
1734         .rpc_release = ff_layout_read_release,
1735 };
1736
1737 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1738         .rpc_call_prepare = ff_layout_read_prepare_v4,
1739         .rpc_call_done = ff_layout_read_call_done,
1740         .rpc_count_stats = ff_layout_read_count_stats,
1741         .rpc_release = ff_layout_read_release,
1742 };
1743
1744 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1745         .rpc_call_prepare = ff_layout_write_prepare_v3,
1746         .rpc_call_done = ff_layout_write_call_done,
1747         .rpc_count_stats = ff_layout_write_count_stats,
1748         .rpc_release = ff_layout_write_release,
1749 };
1750
1751 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1752         .rpc_call_prepare = ff_layout_write_prepare_v4,
1753         .rpc_call_done = ff_layout_write_call_done,
1754         .rpc_count_stats = ff_layout_write_count_stats,
1755         .rpc_release = ff_layout_write_release,
1756 };
1757
1758 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1759         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1760         .rpc_call_done = ff_layout_commit_done,
1761         .rpc_count_stats = ff_layout_commit_count_stats,
1762         .rpc_release = ff_layout_commit_release,
1763 };
1764
1765 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1766         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1767         .rpc_call_done = ff_layout_commit_done,
1768         .rpc_count_stats = ff_layout_commit_count_stats,
1769         .rpc_release = ff_layout_commit_release,
1770 };
1771
1772 static enum pnfs_try_status
1773 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1774 {
1775         struct pnfs_layout_segment *lseg = hdr->lseg;
1776         struct nfs4_pnfs_ds *ds;
1777         struct rpc_clnt *ds_clnt;
1778         struct rpc_cred *ds_cred;
1779         loff_t offset = hdr->args.offset;
1780         u32 idx = hdr->pgio_mirror_idx;
1781         int vers;
1782         struct nfs_fh *fh;
1783
1784         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1785                 __func__, hdr->inode->i_ino,
1786                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1787
1788         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1789         if (!ds)
1790                 goto out_failed;
1791
1792         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1793                                                    hdr->inode);
1794         if (IS_ERR(ds_clnt))
1795                 goto out_failed;
1796
1797         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1798         if (!ds_cred)
1799                 goto out_failed;
1800
1801         vers = nfs4_ff_layout_ds_version(lseg, idx);
1802
1803         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1804                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1805
1806         hdr->pgio_done_cb = ff_layout_read_done_cb;
1807         atomic_inc(&ds->ds_clp->cl_count);
1808         hdr->ds_clp = ds->ds_clp;
1809         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1810         if (fh)
1811                 hdr->args.fh = fh;
1812         /*
1813          * Note that if we ever decide to split across DSes,
1814          * then we may need to handle dense-like offsets.
1815          */
1816         hdr->args.offset = offset;
1817         hdr->mds_offset = offset;
1818
1819         /* Perform an asynchronous read to ds */
1820         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1821                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1822                                       &ff_layout_read_call_ops_v4,
1823                           0, RPC_TASK_SOFTCONN);
1824         put_rpccred(ds_cred);
1825         return PNFS_ATTEMPTED;
1826
1827 out_failed:
1828         if (ff_layout_avoid_mds_available_ds(lseg))
1829                 return PNFS_TRY_AGAIN;
1830         return PNFS_NOT_ATTEMPTED;
1831 }
1832
1833 /* Perform async writes. */
1834 static enum pnfs_try_status
1835 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1836 {
1837         struct pnfs_layout_segment *lseg = hdr->lseg;
1838         struct nfs4_pnfs_ds *ds;
1839         struct rpc_clnt *ds_clnt;
1840         struct rpc_cred *ds_cred;
1841         loff_t offset = hdr->args.offset;
1842         int vers;
1843         struct nfs_fh *fh;
1844         int idx = hdr->pgio_mirror_idx;
1845
1846         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1847         if (!ds)
1848                 return PNFS_NOT_ATTEMPTED;
1849
1850         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1851                                                    hdr->inode);
1852         if (IS_ERR(ds_clnt))
1853                 return PNFS_NOT_ATTEMPTED;
1854
1855         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1856         if (!ds_cred)
1857                 return PNFS_NOT_ATTEMPTED;
1858
1859         vers = nfs4_ff_layout_ds_version(lseg, idx);
1860
1861         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1862                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1863                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1864                 vers);
1865
1866         hdr->pgio_done_cb = ff_layout_write_done_cb;
1867         atomic_inc(&ds->ds_clp->cl_count);
1868         hdr->ds_clp = ds->ds_clp;
1869         hdr->ds_commit_idx = idx;
1870         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1871         if (fh)
1872                 hdr->args.fh = fh;
1873
1874         /*
1875          * Note that if we ever decide to split across DSes,
1876          * then we may need to handle dense-like offsets.
1877          */
1878         hdr->args.offset = offset;
1879
1880         /* Perform an asynchronous write */
1881         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1882                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1883                                       &ff_layout_write_call_ops_v4,
1884                           sync, RPC_TASK_SOFTCONN);
1885         put_rpccred(ds_cred);
1886         return PNFS_ATTEMPTED;
1887 }
1888
1889 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1890 {
1891         return i;
1892 }
1893
1894 static struct nfs_fh *
1895 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1896 {
1897         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1898
1899         /* FIXME: Assume that there is only one NFS version available
1900          * for the DS.
1901          */
1902         return &flseg->mirror_array[i]->fh_versions[0];
1903 }
1904
1905 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1906 {
1907         struct pnfs_layout_segment *lseg = data->lseg;
1908         struct nfs4_pnfs_ds *ds;
1909         struct rpc_clnt *ds_clnt;
1910         struct rpc_cred *ds_cred;
1911         u32 idx;
1912         int vers, ret;
1913         struct nfs_fh *fh;
1914
1915         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1916         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1917         if (!ds)
1918                 goto out_err;
1919
1920         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1921                                                    data->inode);
1922         if (IS_ERR(ds_clnt))
1923                 goto out_err;
1924
1925         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1926         if (!ds_cred)
1927                 goto out_err;
1928
1929         vers = nfs4_ff_layout_ds_version(lseg, idx);
1930
1931         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1932                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1933                 vers);
1934         data->commit_done_cb = ff_layout_commit_done_cb;
1935         data->cred = ds_cred;
1936         atomic_inc(&ds->ds_clp->cl_count);
1937         data->ds_clp = ds->ds_clp;
1938         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1939         if (fh)
1940                 data->args.fh = fh;
1941
1942         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1943                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1944                                                &ff_layout_commit_call_ops_v4,
1945                                    how, RPC_TASK_SOFTCONN);
1946         put_rpccred(ds_cred);
1947         return ret;
1948 out_err:
1949         pnfs_generic_prepare_to_resend_writes(data);
1950         pnfs_generic_commit_release(data);
1951         return -EAGAIN;
1952 }
1953
1954 static int
1955 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1956                            int how, struct nfs_commit_info *cinfo)
1957 {
1958         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1959                                             ff_layout_initiate_commit);
1960 }
1961
1962 static struct pnfs_ds_commit_info *
1963 ff_layout_get_ds_info(struct inode *inode)
1964 {
1965         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1966
1967         if (layout == NULL)
1968                 return NULL;
1969
1970         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1971 }
1972
1973 static void
1974 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1975 {
1976         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1977                                                   id_node));
1978 }
1979
1980 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1981                                   struct xdr_stream *xdr,
1982                                   const struct nfs4_layoutreturn_args *args,
1983                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1984 {
1985         __be32 *start;
1986
1987         start = xdr_reserve_space(xdr, 4);
1988         if (unlikely(!start))
1989                 return -E2BIG;
1990
1991         *start = cpu_to_be32(ff_args->num_errors);
1992         /* This assume we always return _ALL_ layouts */
1993         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
1994 }
1995
1996 static void
1997 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
1998 {
1999         __be32 *p;
2000
2001         p = xdr_reserve_space(xdr, len);
2002         xdr_encode_opaque_fixed(p, buf, len);
2003 }
2004
2005 static void
2006 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2007                             const nfs4_stateid *stateid,
2008                             const struct nfs42_layoutstat_devinfo *devinfo)
2009 {
2010         __be32 *p;
2011
2012         p = xdr_reserve_space(xdr, 8 + 8);
2013         p = xdr_encode_hyper(p, devinfo->offset);
2014         p = xdr_encode_hyper(p, devinfo->length);
2015         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2016         p = xdr_reserve_space(xdr, 4*8);
2017         p = xdr_encode_hyper(p, devinfo->read_count);
2018         p = xdr_encode_hyper(p, devinfo->read_bytes);
2019         p = xdr_encode_hyper(p, devinfo->write_count);
2020         p = xdr_encode_hyper(p, devinfo->write_bytes);
2021         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2022 }
2023
2024 static void
2025 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2026                             const nfs4_stateid *stateid,
2027                             const struct nfs42_layoutstat_devinfo *devinfo)
2028 {
2029         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2030         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2031                         devinfo->ld_private.data);
2032 }
2033
2034 /* report nothing for now */
2035 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2036                 const struct nfs4_layoutreturn_args *args,
2037                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2038 {
2039         __be32 *p;
2040         int i;
2041
2042         p = xdr_reserve_space(xdr, 4);
2043         *p = cpu_to_be32(ff_args->num_dev);
2044         for (i = 0; i < ff_args->num_dev; i++)
2045                 ff_layout_encode_ff_iostat(xdr,
2046                                 &args->layout->plh_stateid,
2047                                 &ff_args->devinfo[i]);
2048 }
2049
2050 static void
2051 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2052                 unsigned int num_entries)
2053 {
2054         unsigned int i;
2055
2056         for (i = 0; i < num_entries; i++) {
2057                 if (!devinfo[i].ld_private.ops)
2058                         continue;
2059                 if (!devinfo[i].ld_private.ops->free)
2060                         continue;
2061                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2062         }
2063 }
2064
2065 static struct nfs4_deviceid_node *
2066 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2067                               struct pnfs_device *pdev, gfp_t gfp_flags)
2068 {
2069         struct nfs4_ff_layout_ds *dsaddr;
2070
2071         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2072         if (!dsaddr)
2073                 return NULL;
2074         return &dsaddr->id_node;
2075 }
2076
2077 static void
2078 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2079                 const void *voidargs,
2080                 const struct nfs4_xdr_opaque_data *ff_opaque)
2081 {
2082         const struct nfs4_layoutreturn_args *args = voidargs;
2083         struct pnfs_layout_hdr *lo = args->layout;
2084         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
2085         __be32 *start;
2086
2087         dprintk("%s: Begin\n", __func__);
2088         start = xdr_reserve_space(xdr, 4);
2089         BUG_ON(!start);
2090
2091         ff_layout_encode_ioerr(flo, xdr, args, ff_opaque->data);
2092         ff_layout_encode_iostats_array(xdr, args, ff_opaque->data);
2093
2094         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2095         dprintk("%s: Return\n", __func__);
2096 }
2097
2098 static void
2099 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2100 {
2101         struct nfs4_flexfile_layoutreturn_args *ff_args;
2102
2103         if (!args->data)
2104                 return;
2105         ff_args = args->data;
2106         args->data = NULL;
2107
2108         ff_layout_free_ds_ioerr(&ff_args->errors);
2109         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2110
2111         kfree(ff_args);
2112 }
2113
2114 const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2115         .encode = ff_layout_encode_layoutreturn,
2116         .free = ff_layout_free_layoutreturn,
2117 };
2118
2119 static int
2120 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2121 {
2122         struct nfs4_flexfile_layoutreturn_args *ff_args;
2123         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2124
2125         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2126         if (!ff_args)
2127                 return -ENOMEM;
2128
2129         INIT_LIST_HEAD(&ff_args->errors);
2130         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2131                         &args->range, &ff_args->errors,
2132                         FF_LAYOUTRETURN_MAXERR);
2133
2134         spin_lock(&args->inode->i_lock);
2135         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2136                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2137         spin_unlock(&args->inode->i_lock);
2138
2139         args->ld_private->ops = &layoutreturn_ops;
2140         args->ld_private->data = ff_args;
2141         return 0;
2142 }
2143
2144 static int
2145 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2146 {
2147         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2148
2149         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2150 }
2151
2152 static size_t
2153 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2154                           const int buflen)
2155 {
2156         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2157         const struct in6_addr *addr = &sin6->sin6_addr;
2158
2159         /*
2160          * RFC 4291, Section 2.2.2
2161          *
2162          * Shorthanded ANY address
2163          */
2164         if (ipv6_addr_any(addr))
2165                 return snprintf(buf, buflen, "::");
2166
2167         /*
2168          * RFC 4291, Section 2.2.2
2169          *
2170          * Shorthanded loopback address
2171          */
2172         if (ipv6_addr_loopback(addr))
2173                 return snprintf(buf, buflen, "::1");
2174
2175         /*
2176          * RFC 4291, Section 2.2.3
2177          *
2178          * Special presentation address format for mapped v4
2179          * addresses.
2180          */
2181         if (ipv6_addr_v4mapped(addr))
2182                 return snprintf(buf, buflen, "::ffff:%pI4",
2183                                         &addr->s6_addr32[3]);
2184
2185         /*
2186          * RFC 4291, Section 2.2.1
2187          */
2188         return snprintf(buf, buflen, "%pI6c", addr);
2189 }
2190
2191 /* Derived from rpc_sockaddr2uaddr */
2192 static void
2193 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2194 {
2195         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2196         char portbuf[RPCBIND_MAXUADDRPLEN];
2197         char addrbuf[RPCBIND_MAXUADDRLEN];
2198         char *netid;
2199         unsigned short port;
2200         int len, netid_len;
2201         __be32 *p;
2202
2203         switch (sap->sa_family) {
2204         case AF_INET:
2205                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2206                         return;
2207                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2208                 netid = "tcp";
2209                 netid_len = 3;
2210                 break;
2211         case AF_INET6:
2212                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2213                         return;
2214                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2215                 netid = "tcp6";
2216                 netid_len = 4;
2217                 break;
2218         default:
2219                 /* we only support tcp and tcp6 */
2220                 WARN_ON_ONCE(1);
2221                 return;
2222         }
2223
2224         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2225         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2226
2227         p = xdr_reserve_space(xdr, 4 + netid_len);
2228         xdr_encode_opaque(p, netid, netid_len);
2229
2230         p = xdr_reserve_space(xdr, 4 + len);
2231         xdr_encode_opaque(p, addrbuf, len);
2232 }
2233
2234 static void
2235 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2236                          ktime_t t)
2237 {
2238         struct timespec64 ts;
2239         __be32 *p;
2240
2241         p = xdr_reserve_space(xdr, 12);
2242         ts = ktime_to_timespec64(t);
2243         p = xdr_encode_hyper(p, ts.tv_sec);
2244         *p++ = cpu_to_be32(ts.tv_nsec);
2245 }
2246
2247 static void
2248 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2249                             struct nfs4_ff_io_stat *stat)
2250 {
2251         __be32 *p;
2252
2253         p = xdr_reserve_space(xdr, 5 * 8);
2254         p = xdr_encode_hyper(p, stat->ops_requested);
2255         p = xdr_encode_hyper(p, stat->bytes_requested);
2256         p = xdr_encode_hyper(p, stat->ops_completed);
2257         p = xdr_encode_hyper(p, stat->bytes_completed);
2258         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2259         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2260         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2261 }
2262
2263 static void
2264 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2265                               const struct nfs42_layoutstat_devinfo *devinfo,
2266                               struct nfs4_ff_layout_mirror *mirror)
2267 {
2268         struct nfs4_pnfs_ds_addr *da;
2269         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2270         struct nfs_fh *fh = &mirror->fh_versions[0];
2271         __be32 *p;
2272
2273         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2274         dprintk("%s: DS %s: encoding address %s\n",
2275                 __func__, ds->ds_remotestr, da->da_remotestr);
2276         /* netaddr4 */
2277         ff_layout_encode_netaddr(xdr, da);
2278         /* nfs_fh4 */
2279         p = xdr_reserve_space(xdr, 4 + fh->size);
2280         xdr_encode_opaque(p, fh->data, fh->size);
2281         /* ff_io_latency4 read */
2282         spin_lock(&mirror->lock);
2283         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2284         /* ff_io_latency4 write */
2285         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2286         spin_unlock(&mirror->lock);
2287         /* nfstime4 */
2288         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2289         /* bool */
2290         p = xdr_reserve_space(xdr, 4);
2291         *p = cpu_to_be32(false);
2292 }
2293
2294 static void
2295 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2296                              const struct nfs4_xdr_opaque_data *opaque)
2297 {
2298         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2299                         struct nfs42_layoutstat_devinfo, ld_private);
2300         __be32 *start;
2301
2302         /* layoutupdate length */
2303         start = xdr_reserve_space(xdr, 4);
2304         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2305
2306         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2307 }
2308
2309 static void
2310 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2311 {
2312         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2313
2314         ff_layout_put_mirror(mirror);
2315 }
2316
2317 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2318         .encode = ff_layout_encode_layoutstats,
2319         .free   = ff_layout_free_layoutstats,
2320 };
2321
2322 static int
2323 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2324                                struct nfs42_layoutstat_devinfo *devinfo,
2325                                int dev_limit)
2326 {
2327         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2328         struct nfs4_ff_layout_mirror *mirror;
2329         struct nfs4_deviceid_node *dev;
2330         int i = 0;
2331
2332         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2333                 if (i >= dev_limit)
2334                         break;
2335                 if (!mirror->mirror_ds)
2336                         continue;
2337                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2338                         continue;
2339                 /* mirror refcount put in cleanup_layoutstats */
2340                 if (!atomic_inc_not_zero(&mirror->ref))
2341                         continue;
2342                 dev = &mirror->mirror_ds->id_node;
2343                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2344                 devinfo->offset = 0;
2345                 devinfo->length = NFS4_MAX_UINT64;
2346                 spin_lock(&mirror->lock);
2347                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2348                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2349                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2350                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2351                 spin_unlock(&mirror->lock);
2352                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2353                 devinfo->ld_private.ops = &layoutstat_ops;
2354                 devinfo->ld_private.data = mirror;
2355
2356                 devinfo++;
2357                 i++;
2358         }
2359         return i;
2360 }
2361
2362 static int
2363 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2364 {
2365         struct nfs4_flexfile_layout *ff_layout;
2366         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2367
2368         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2369         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2370         if (!args->devinfo)
2371                 return -ENOMEM;
2372
2373         spin_lock(&args->inode->i_lock);
2374         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2375         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2376                         &args->devinfo[0], dev_count);
2377         spin_unlock(&args->inode->i_lock);
2378         if (!args->num_dev) {
2379                 kfree(args->devinfo);
2380                 args->devinfo = NULL;
2381                 return -ENOENT;
2382         }
2383
2384         return 0;
2385 }
2386
2387 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2388         .id                     = LAYOUT_FLEX_FILES,
2389         .name                   = "LAYOUT_FLEX_FILES",
2390         .owner                  = THIS_MODULE,
2391         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2392         .free_layout_hdr        = ff_layout_free_layout_hdr,
2393         .alloc_lseg             = ff_layout_alloc_lseg,
2394         .free_lseg              = ff_layout_free_lseg,
2395         .add_lseg               = ff_layout_add_lseg,
2396         .pg_read_ops            = &ff_layout_pg_read_ops,
2397         .pg_write_ops           = &ff_layout_pg_write_ops,
2398         .get_ds_info            = ff_layout_get_ds_info,
2399         .free_deviceid_node     = ff_layout_free_deviceid_node,
2400         .mark_request_commit    = pnfs_layout_mark_request_commit,
2401         .clear_request_commit   = pnfs_generic_clear_request_commit,
2402         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2403         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2404         .commit_pagelist        = ff_layout_commit_pagelist,
2405         .read_pagelist          = ff_layout_read_pagelist,
2406         .write_pagelist         = ff_layout_write_pagelist,
2407         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2408         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2409         .sync                   = pnfs_nfs_generic_sync,
2410         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2411 };
2412
2413 static int __init nfs4flexfilelayout_init(void)
2414 {
2415         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2416                __func__);
2417         if (!ff_zero_group) {
2418                 ff_zero_group = groups_alloc(0);
2419                 if (!ff_zero_group)
2420                         return -ENOMEM;
2421         }
2422         return pnfs_register_layoutdriver(&flexfilelayout_type);
2423 }
2424
2425 static void __exit nfs4flexfilelayout_exit(void)
2426 {
2427         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2428                __func__);
2429         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2430         if (ff_zero_group) {
2431                 put_group_info(ff_zero_group);
2432                 ff_zero_group = NULL;
2433         }
2434 }
2435
2436 MODULE_ALIAS("nfs-layouttype4-4");
2437
2438 MODULE_LICENSE("GPL");
2439 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2440
2441 module_init(nfs4flexfilelayout_init);
2442 module_exit(nfs4flexfilelayout_exit);