fs/nfs/flexfilelayout/flexfilelayout.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Module for pnfs flexfile layout driver.
   4  *
   5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   6  *
   7  * Tao Peng <[email protected]>
   8  */
   9
  10 #include <linux/nfs_fs.h>
  11 #include <linux/nfs_mount.h>
  12 #include <linux/nfs_page.h>
  13 #include <linux/module.h>
  14 #include <linux/sched/mm.h>
  15
  16 #include <linux/sunrpc/metrics.h>
  17
  18 #include "flexfilelayout.h"
  19 #include "../nfs4session.h"
  20 #include "../nfs4idmap.h"
  21 #include "../internal.h"
  22 #include "../delegation.h"
  23 #include "../nfs4trace.h"
  24 #include "../iostat.h"
  25 #include "../nfs.h"
  26 #include "../nfs42.h"
  27
  28 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  29
  30 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  31 #define FF_LAYOUTRETURN_MAXERR 20
  32
  33 static unsigned short io_maxretrans;
  34
  35 static const struct pnfs_commit_ops ff_layout_commit_ops;
  36 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
  37                 struct nfs_pgio_header *hdr);
  38 static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
  39                                struct nfs42_layoutstat_devinfo *devinfo,
  40                                int dev_limit);
  41 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
  42                               const struct nfs42_layoutstat_devinfo *devinfo,
  43                               struct nfs4_ff_layout_mirror *mirror);
  44
  45 static struct pnfs_layout_hdr *
  46 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  47 {
  48         struct nfs4_flexfile_layout *ffl;
  49
  50         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  51         if (ffl) {
  52                 pnfs_init_ds_commit_info(&ffl->commit_info);
  53                 INIT_LIST_HEAD(&ffl->error_list);
  54                 INIT_LIST_HEAD(&ffl->mirrors);
  55                 ffl->last_report_time = ktime_get();
  56                 ffl->commit_info.ops = &ff_layout_commit_ops;
  57                 return &ffl->generic_hdr;
  58         } else
  59                 return NULL;
  60 }
  61
  62 static void
  63 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  64 {
  65         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo);
  66         struct nfs4_ff_layout_ds_err *err, *n;
  67
  68         list_for_each_entry_safe(err, n, &ffl->error_list, list) {
  69                 list_del(&err->list);
  70                 kfree(err);
  71         }
  72         kfree_rcu(ffl, generic_hdr.plh_rcu);
  73 }
  74
  75 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  76 {
  77         __be32 *p;
  78
  79         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  80         if (unlikely(p == NULL))
  81                 return -ENOBUFS;
  82         stateid->type = NFS4_PNFS_DS_STATEID_TYPE;
  83         memcpy(stateid->data, p, NFS4_STATEID_SIZE);
  84         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  85                 p[0], p[1], p[2], p[3]);
  86         return 0;
  87 }
  88
  89 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  90 {
  91         __be32 *p;
  92
  93         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  94         if (unlikely(!p))
  95                 return -ENOBUFS;
  96         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  97         nfs4_print_deviceid(devid);
  98         return 0;
  99 }
 100
 101 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
 102 {
 103         __be32 *p;
 104
 105         p = xdr_inline_decode(xdr, 4);
 106         if (unlikely(!p))
 107                 return -ENOBUFS;
 108         fh->size = be32_to_cpup(p++);
 109         if (fh->size > sizeof(struct nfs_fh)) {
 110                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
 111                        fh->size);
 112                 return -EOVERFLOW;
 113         }
 114         /* fh.data */
 115         p = xdr_inline_decode(xdr, fh->size);
 116         if (unlikely(!p))
 117                 return -ENOBUFS;
 118         memcpy(&fh->data, p, fh->size);
 119         dprintk("%s: fh len %d\n", __func__, fh->size);
 120
 121         return 0;
 122 }
 123
 124 /*
 125  * Currently only stringified uids and gids are accepted.
 126  * I.e., kerberos is not supported to the DSes, so no pricipals.
 127  *
 128  * That means that one common function will suffice, but when
 129  * principals are added, this should be split to accomodate
 130  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 131  */
 132 static int
 133 decode_name(struct xdr_stream *xdr, u32 *id)
 134 {
 135         __be32 *p;
 136         int len;
 137
 138         /* opaque_length(4)*/
 139         p = xdr_inline_decode(xdr, 4);
 140         if (unlikely(!p))
 141                 return -ENOBUFS;
 142         len = be32_to_cpup(p++);
 143         if (len < 0)
 144                 return -EINVAL;
 145
 146         dprintk("%s: len %u\n", __func__, len);
 147
 148         /* opaque body */
 149         p = xdr_inline_decode(xdr, len);
 150         if (unlikely(!p))
 151                 return -ENOBUFS;
 152
 153         if (!nfs_map_string_to_numeric((char *)p, len, id))
 154                 return -EINVAL;
 155
 156         return 0;
 157 }
 158
 159 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 160                 const struct nfs4_ff_layout_mirror *m2)
 161 {
 162         int i, j;
 163
 164         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 165                 return false;
 166         for (i = 0; i < m1->fh_versions_cnt; i++) {
 167                 bool found_fh = false;
 168                 for (j = 0; j < m2->fh_versions_cnt; j++) {
 169                         if (nfs_compare_fh(&m1->fh_versions[i],
 170                                         &m2->fh_versions[j]) == 0) {
 171                                 found_fh = true;
 172                                 break;
 173                         }
 174                 }
 175                 if (!found_fh)
 176                         return false;
 177         }
 178         return true;
 179 }
 180
 181 static struct nfs4_ff_layout_mirror *
 182 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 183                 struct nfs4_ff_layout_mirror *mirror)
 184 {
 185         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 186         struct nfs4_ff_layout_mirror *pos;
 187         struct inode *inode = lo->plh_inode;
 188
 189         spin_lock(&inode->i_lock);
 190         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 191                 if (memcmp(&mirror->devid, &pos->devid, sizeof(pos->devid)) != 0)
 192                         continue;
 193                 if (!ff_mirror_match_fh(mirror, pos))
 194                         continue;
 195                 if (refcount_inc_not_zero(&pos->ref)) {
 196                         spin_unlock(&inode->i_lock);
 197                         return pos;
 198                 }
 199         }
 200         list_add(&mirror->mirrors, &ff_layout->mirrors);
 201         mirror->layout = lo;
 202         spin_unlock(&inode->i_lock);
 203         return mirror;
 204 }
 205
 206 static void
 207 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 208 {
 209         struct inode *inode;
 210         if (mirror->layout == NULL)
 211                 return;
 212         inode = mirror->layout->plh_inode;
 213         spin_lock(&inode->i_lock);
 214         list_del(&mirror->mirrors);
 215         spin_unlock(&inode->i_lock);
 216         mirror->layout = NULL;
 217 }
 218
 219 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 220 {
 221         struct nfs4_ff_layout_mirror *mirror;
 222
 223         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 224         if (mirror != NULL) {
 225                 spin_lock_init(&mirror->lock);
 226                 refcount_set(&mirror->ref, 1);
 227                 INIT_LIST_HEAD(&mirror->mirrors);
 228         }
 229         return mirror;
 230 }
 231
 232 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 233 {
 234         const struct cred       *cred;
 235
 236         ff_layout_remove_mirror(mirror);
 237         kfree(mirror->fh_versions);
 238         cred = rcu_access_pointer(mirror->ro_cred);
 239         put_cred(cred);
 240         cred = rcu_access_pointer(mirror->rw_cred);
 241         put_cred(cred);
 242         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 243         kfree(mirror);
 244 }
 245
 246 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 247 {
 248         if (mirror != NULL && refcount_dec_and_test(&mirror->ref))
 249                 ff_layout_free_mirror(mirror);
 250 }
 251
 252 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 253 {
 254         u32 i;
 255
 256         for (i = 0; i < fls->mirror_array_cnt; i++)
 257                 ff_layout_put_mirror(fls->mirror_array[i]);
 258 }
 259
 260 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 261 {
 262         if (fls) {
 263                 ff_layout_free_mirror_array(fls);
 264                 kfree(fls);
 265         }
 266 }
 267
 268 static bool
 269 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1,
 270                 struct pnfs_layout_segment *l2)
 271 {
 272         const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1);
 273         const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l1);
 274         u32 i;
 275
 276         if (fl1->mirror_array_cnt != fl2->mirror_array_cnt)
 277                 return false;
 278         for (i = 0; i < fl1->mirror_array_cnt; i++) {
 279                 if (fl1->mirror_array[i] != fl2->mirror_array[i])
 280                         return false;
 281         }
 282         return true;
 283 }
 284
 285 static bool
 286 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 287                 const struct pnfs_layout_range *l2)
 288 {
 289         u64 end1, end2;
 290
 291         if (l1->iomode != l2->iomode)
 292                 return l1->iomode != IOMODE_READ;
 293         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 294         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 295         if (end1 < l2->offset)
 296                 return false;
 297         if (end2 < l1->offset)
 298                 return true;
 299         return l2->offset <= l1->offset;
 300 }
 301
 302 static bool
 303 ff_lseg_merge(struct pnfs_layout_segment *new,
 304                 struct pnfs_layout_segment *old)
 305 {
 306         u64 new_end, old_end;
 307
 308         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 309                 return false;
 310         if (new->pls_range.iomode != old->pls_range.iomode)
 311                 return false;
 312         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 313                         old->pls_range.length);
 314         if (old_end < new->pls_range.offset)
 315                 return false;
 316         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 317                         new->pls_range.length);
 318         if (new_end < old->pls_range.offset)
 319                 return false;
 320         if (!ff_lseg_match_mirrors(new, old))
 321                 return false;
 322
 323         /* Mergeable: copy info from 'old' to 'new' */
 324         if (new_end < old_end)
 325                 new_end = old_end;
 326         if (new->pls_range.offset < old->pls_range.offset)
 327                 new->pls_range.offset = old->pls_range.offset;
 328         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 329                         new_end);
 330         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 331                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 332         return true;
 333 }
 334
 335 static void
 336 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 337                 struct pnfs_layout_segment *lseg,
 338                 struct list_head *free_me)
 339 {
 340         pnfs_generic_layout_insert_lseg(lo, lseg,
 341                         ff_lseg_range_is_after,
 342                         ff_lseg_merge,
 343                         free_me);
 344 }
 345
 346 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 347 {
 348         int i, j;
 349
 350         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 351                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 352                         if (fls->mirror_array[i]->efficiency <
 353                             fls->mirror_array[j]->efficiency)
 354                                 swap(fls->mirror_array[i],
 355                                      fls->mirror_array[j]);
 356         }
 357 }
 358
 359 static struct pnfs_layout_segment *
 360 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 361                      struct nfs4_layoutget_res *lgr,
 362                      gfp_t gfp_flags)
 363 {
 364         struct pnfs_layout_segment *ret;
 365         struct nfs4_ff_layout_segment *fls = NULL;
 366         struct xdr_stream stream;
 367         struct xdr_buf buf;
 368         struct page *scratch;
 369         u64 stripe_unit;
 370         u32 mirror_array_cnt;
 371         __be32 *p;
 372         int i, rc;
 373
 374         dprintk("--> %s\n", __func__);
 375         scratch = alloc_page(gfp_flags);
 376         if (!scratch)
 377                 return ERR_PTR(-ENOMEM);
 378
 379         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 380                               lgr->layoutp->len);
 381         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 382
 383         /* stripe unit and mirror_array_cnt */
 384         rc = -EIO;
 385         p = xdr_inline_decode(&stream, 8 + 4);
 386         if (!p)
 387                 goto out_err_free;
 388
 389         p = xdr_decode_hyper(p, &stripe_unit);
 390         mirror_array_cnt = be32_to_cpup(p++);
 391         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 392                 stripe_unit, mirror_array_cnt);
 393
 394         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 395             mirror_array_cnt == 0)
 396                 goto out_err_free;
 397
 398         rc = -ENOMEM;
 399         fls = kzalloc(struct_size(fls, mirror_array, mirror_array_cnt),
 400                         gfp_flags);
 401         if (!fls)
 402                 goto out_err_free;
 403
 404         fls->mirror_array_cnt = mirror_array_cnt;
 405         fls->stripe_unit = stripe_unit;
 406
 407         for (i = 0; i < fls->mirror_array_cnt; i++) {
 408                 struct nfs4_ff_layout_mirror *mirror;
 409                 struct cred *kcred;
 410                 const struct cred __rcu *cred;
 411                 kuid_t uid;
 412                 kgid_t gid;
 413                 u32 ds_count, fh_count, id;
 414                 int j;
 415
 416                 rc = -EIO;
 417                 p = xdr_inline_decode(&stream, 4);
 418                 if (!p)
 419                         goto out_err_free;
 420                 ds_count = be32_to_cpup(p);
 421
 422                 /* FIXME: allow for striping? */
 423                 if (ds_count != 1)
 424                         goto out_err_free;
 425
 426                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 427                 if (fls->mirror_array[i] == NULL) {
 428                         rc = -ENOMEM;
 429                         goto out_err_free;
 430                 }
 431
 432                 fls->mirror_array[i]->ds_count = ds_count;
 433
 434                 /* deviceid */
 435                 rc = decode_deviceid(&stream, &fls->mirror_array[i]->devid);
 436                 if (rc)
 437                         goto out_err_free;
 438
 439                 /* efficiency */
 440                 rc = -EIO;
 441                 p = xdr_inline_decode(&stream, 4);
 442                 if (!p)
 443                         goto out_err_free;
 444                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 445
 446                 /* stateid */
 447                 rc = decode_pnfs_stateid(&stream, &fls->mirror_array[i]->stateid);
 448                 if (rc)
 449                         goto out_err_free;
 450
 451                 /* fh */
 452                 rc = -EIO;
 453                 p = xdr_inline_decode(&stream, 4);
 454                 if (!p)
 455                         goto out_err_free;
 456                 fh_count = be32_to_cpup(p);
 457
 458                 fls->mirror_array[i]->fh_versions =
 459                         kcalloc(fh_count, sizeof(struct nfs_fh),
 460                                 gfp_flags);
 461                 if (fls->mirror_array[i]->fh_versions == NULL) {
 462                         rc = -ENOMEM;
 463                         goto out_err_free;
 464                 }
 465
 466                 for (j = 0; j < fh_count; j++) {
 467                         rc = decode_nfs_fh(&stream,
 468                                            &fls->mirror_array[i]->fh_versions[j]);
 469                         if (rc)
 470                                 goto out_err_free;
 471                 }
 472
 473                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 474
 475                 /* user */
 476                 rc = decode_name(&stream, &id);
 477                 if (rc)
 478                         goto out_err_free;
 479
 480                 uid = make_kuid(&init_user_ns, id);
 481
 482                 /* group */
 483                 rc = decode_name(&stream, &id);
 484                 if (rc)
 485                         goto out_err_free;
 486
 487                 gid = make_kgid(&init_user_ns, id);
 488
 489                 if (gfp_flags & __GFP_FS)
 490                         kcred = prepare_kernel_cred(NULL);
 491                 else {
 492                         unsigned int nofs_flags = memalloc_nofs_save();
 493                         kcred = prepare_kernel_cred(NULL);
 494                         memalloc_nofs_restore(nofs_flags);
 495                 }
 496                 rc = -ENOMEM;
 497                 if (!kcred)
 498                         goto out_err_free;
 499                 kcred->fsuid = uid;
 500                 kcred->fsgid = gid;
 501                 cred = RCU_INITIALIZER(kcred);
 502
 503                 if (lgr->range.iomode == IOMODE_READ)
 504                         rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 505                 else
 506                         rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 507
 508                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 509                 if (mirror != fls->mirror_array[i]) {
 510                         /* swap cred ptrs so free_mirror will clean up old */
 511                         if (lgr->range.iomode == IOMODE_READ) {
 512                                 cred = xchg(&mirror->ro_cred, cred);
 513                                 rcu_assign_pointer(fls->mirror_array[i]->ro_cred, cred);
 514                         } else {
 515                                 cred = xchg(&mirror->rw_cred, cred);
 516                                 rcu_assign_pointer(fls->mirror_array[i]->rw_cred, cred);
 517                         }
 518                         ff_layout_free_mirror(fls->mirror_array[i]);
 519                         fls->mirror_array[i] = mirror;
 520                 }
 521
 522                 dprintk("%s: iomode %s uid %u gid %u\n", __func__,
 523                         lgr->range.iomode == IOMODE_READ ? "READ" : "RW",
 524                         from_kuid(&init_user_ns, uid),
 525                         from_kgid(&init_user_ns, gid));
 526         }
 527
 528         p = xdr_inline_decode(&stream, 4);
 529         if (!p)
 530                 goto out_sort_mirrors;
 531         fls->flags = be32_to_cpup(p);
 532
 533         p = xdr_inline_decode(&stream, 4);
 534         if (!p)
 535                 goto out_sort_mirrors;
 536         for (i=0; i < fls->mirror_array_cnt; i++)
 537                 fls->mirror_array[i]->report_interval = be32_to_cpup(p);
 538
 539 out_sort_mirrors:
 540         ff_layout_sort_mirrors(fls);
 541         ret = &fls->generic_hdr;
 542         dprintk("<-- %s (success)\n", __func__);
 543 out_free_page:
 544         __free_page(scratch);
 545         return ret;
 546 out_err_free:
 547         _ff_layout_free_lseg(fls);
 548         ret = ERR_PTR(rc);
 549         dprintk("<-- %s (%d)\n", __func__, rc);
 550         goto out_free_page;
 551 }
 552
 553 static void
 554 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 555 {
 556         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 557
 558         dprintk("--> %s\n", __func__);
 559
 560         if (lseg->pls_range.iomode == IOMODE_RW) {
 561                 struct nfs4_flexfile_layout *ffl;
 562                 struct inode *inode;
 563
 564                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 565                 inode = ffl->generic_hdr.plh_inode;
 566                 spin_lock(&inode->i_lock);
 567                 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg);
 568                 spin_unlock(&inode->i_lock);
 569         }
 570         _ff_layout_free_lseg(fls);
 571 }
 572
 573 static void
 574 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 575 {
 576         /* first IO request? */
 577         if (atomic_inc_return(&timer->n_ops) == 1) {
 578                 timer->start_time = now;
 579         }
 580 }
 581
 582 static ktime_t
 583 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 584 {
 585         ktime_t start;
 586
 587         if (atomic_dec_return(&timer->n_ops) < 0)
 588                 WARN_ON_ONCE(1);
 589
 590         start = timer->start_time;
 591         timer->start_time = now;
 592         return ktime_sub(now, start);
 593 }
 594
 595 static bool
 596 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 597                             struct nfs4_ff_layoutstat *layoutstat,
 598                             ktime_t now)
 599 {
 600         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 601         struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout);
 602
 603         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 604         if (!mirror->start_time)
 605                 mirror->start_time = now;
 606         if (mirror->report_interval != 0)
 607                 report_interval = (s64)mirror->report_interval * 1000LL;
 608         else if (layoutstats_timer != 0)
 609                 report_interval = (s64)layoutstats_timer * 1000LL;
 610         if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >=
 611                         report_interval) {
 612                 ffl->last_report_time = now;
 613                 return true;
 614         }
 615
 616         return false;
 617 }
 618
 619 static void
 620 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 621                 __u64 requested)
 622 {
 623         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 624
 625         iostat->ops_requested++;
 626         iostat->bytes_requested += requested;
 627 }
 628
 629 static void
 630 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 631                 __u64 requested,
 632                 __u64 completed,
 633                 ktime_t time_completed,
 634                 ktime_t time_started)
 635 {
 636         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 637         ktime_t completion_time = ktime_sub(time_completed, time_started);
 638         ktime_t timer;
 639
 640         iostat->ops_completed++;
 641         iostat->bytes_completed += completed;
 642         iostat->bytes_not_delivered += requested - completed;
 643
 644         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 645         iostat->total_busy_time =
 646                         ktime_add(iostat->total_busy_time, timer);
 647         iostat->aggregate_completion_time =
 648                         ktime_add(iostat->aggregate_completion_time,
 649                                         completion_time);
 650 }
 651
 652 static void
 653 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 654                 struct nfs4_ff_layout_mirror *mirror,
 655                 __u64 requested, ktime_t now)
 656 {
 657         bool report;
 658
 659         spin_lock(&mirror->lock);
 660         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 661         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 662         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 663         spin_unlock(&mirror->lock);
 664
 665         if (report)
 666                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 667 }
 668
 669 static void
 670 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 671                 struct nfs4_ff_layout_mirror *mirror,
 672                 __u64 requested,
 673                 __u64 completed)
 674 {
 675         spin_lock(&mirror->lock);
 676         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 677                         requested, completed,
 678                         ktime_get(), task->tk_start);
 679         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 680         spin_unlock(&mirror->lock);
 681 }
 682
 683 static void
 684 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 685                 struct nfs4_ff_layout_mirror *mirror,
 686                 __u64 requested, ktime_t now)
 687 {
 688         bool report;
 689
 690         spin_lock(&mirror->lock);
 691         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 692         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 693         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 694         spin_unlock(&mirror->lock);
 695
 696         if (report)
 697                 pnfs_report_layoutstat(inode, GFP_NOIO);
 698 }
 699
 700 static void
 701 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 702                 struct nfs4_ff_layout_mirror *mirror,
 703                 __u64 requested,
 704                 __u64 completed,
 705                 enum nfs3_stable_how committed)
 706 {
 707         if (committed == NFS_UNSTABLE)
 708                 requested = completed = 0;
 709
 710         spin_lock(&mirror->lock);
 711         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 712                         requested, completed, ktime_get(), task->tk_start);
 713         set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags);
 714         spin_unlock(&mirror->lock);
 715 }
 716
 717 static void
 718 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx)
 719 {
 720         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 721
 722         if (devid)
 723                 nfs4_mark_deviceid_unavailable(devid);
 724 }
 725
 726 static void
 727 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx)
 728 {
 729         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 730
 731         if (devid)
 732                 nfs4_mark_deviceid_available(devid);
 733 }
 734
 735 static struct nfs4_pnfs_ds *
 736 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg,
 737                              u32 start_idx, u32 *best_idx,
 738                              bool check_device)
 739 {
 740         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 741         struct nfs4_ff_layout_mirror *mirror;
 742         struct nfs4_pnfs_ds *ds;
 743         bool fail_return = false;
 744         u32 idx;
 745
 746         /* mirrors are initially sorted by efficiency */
 747         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 748                 if (idx+1 == fls->mirror_array_cnt)
 749                         fail_return = !check_device;
 750
 751                 mirror = FF_LAYOUT_COMP(lseg, idx);
 752                 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, fail_return);
 753                 if (!ds)
 754                         continue;
 755
 756                 if (check_device &&
 757                     nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node))
 758                         continue;
 759
 760                 *best_idx = idx;
 761                 return ds;
 762         }
 763
 764         return NULL;
 765 }
 766
 767 static struct nfs4_pnfs_ds *
 768 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg,
 769                                  u32 start_idx, u32 *best_idx)
 770 {
 771         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false);
 772 }
 773
 774 static struct nfs4_pnfs_ds *
 775 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg,
 776                                    u32 start_idx, u32 *best_idx)
 777 {
 778         return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true);
 779 }
 780
 781 static struct nfs4_pnfs_ds *
 782 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 783                                   u32 start_idx, u32 *best_idx)
 784 {
 785         struct nfs4_pnfs_ds *ds;
 786
 787         ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx);
 788         if (ds)
 789                 return ds;
 790         return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx);
 791 }
 792
 793 static struct nfs4_pnfs_ds *
 794 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio,
 795                           u32 *best_idx)
 796 {
 797         struct pnfs_layout_segment *lseg = pgio->pg_lseg;
 798         struct nfs4_pnfs_ds *ds;
 799
 800         ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx,
 801                                                best_idx);
 802         if (ds || !pgio->pg_mirror_idx)
 803                 return ds;
 804         return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx);
 805 }
 806
 807 static void
 808 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio,
 809                       struct nfs_page *req,
 810                       bool strict_iomode)
 811 {
 812         pnfs_put_lseg(pgio->pg_lseg);
 813         pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 814                                            nfs_req_openctx(req),
 815                                            req_offset(req),
 816                                            req->wb_bytes,
 817                                            IOMODE_READ,
 818                                            strict_iomode,
 819                                            GFP_KERNEL);
 820         if (IS_ERR(pgio->pg_lseg)) {
 821                 pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 822                 pgio->pg_lseg = NULL;
 823         }
 824 }
 825
 826 static void
 827 ff_layout_pg_check_layout(struct nfs_pageio_descriptor *pgio,
 828                           struct nfs_page *req)
 829 {
 830         pnfs_generic_pg_check_layout(pgio);
 831         pnfs_generic_pg_check_range(pgio, req);
 832 }
 833
 834 static void
 835 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 836                         struct nfs_page *req)
 837 {
 838         struct nfs_pgio_mirror *pgm;
 839         struct nfs4_ff_layout_mirror *mirror;
 840         struct nfs4_pnfs_ds *ds;
 841         u32 ds_idx, i;
 842
 843 retry:
 844         ff_layout_pg_check_layout(pgio, req);
 845         /* Use full layout for now */
 846         if (!pgio->pg_lseg) {
 847                 ff_layout_pg_get_read(pgio, req, false);
 848                 if (!pgio->pg_lseg)
 849                         goto out_nolseg;
 850         }
 851         if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) {
 852                 ff_layout_pg_get_read(pgio, req, true);
 853                 if (!pgio->pg_lseg)
 854                         goto out_nolseg;
 855         }
 856
 857         ds = ff_layout_get_ds_for_read(pgio, &ds_idx);
 858         if (!ds) {
 859                 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 860                         goto out_mds;
 861                 pnfs_generic_pg_cleanup(pgio);
 862                 /* Sleep for 1 second before retrying */
 863                 ssleep(1);
 864                 goto retry;
 865         }
 866
 867         for (i = 0; i < pgio->pg_mirror_count; i++) {
 868                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 869                 pgm = &pgio->pg_mirrors[i];
 870                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 871         }
 872
 873         pgio->pg_mirror_idx = ds_idx;
 874
 875         if (NFS_SERVER(pgio->pg_inode)->flags &
 876                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 877                 pgio->pg_maxretrans = io_maxretrans;
 878         return;
 879 out_nolseg:
 880         if (pgio->pg_error < 0)
 881                 return;
 882 out_mds:
 883         trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode,
 884                         0, NFS4_MAX_UINT64, IOMODE_READ,
 885                         NFS_I(pgio->pg_inode)->layout,
 886                         pgio->pg_lseg);
 887         pgio->pg_maxretrans = 0;
 888         nfs_pageio_reset_read_mds(pgio);
 889 }
 890
 891 static void
 892 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 893                         struct nfs_page *req)
 894 {
 895         struct nfs4_ff_layout_mirror *mirror;
 896         struct nfs_pgio_mirror *pgm;
 897         struct nfs4_pnfs_ds *ds;
 898         u32 i;
 899
 900 retry:
 901         ff_layout_pg_check_layout(pgio, req);
 902         if (!pgio->pg_lseg) {
 903                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 904                                                    nfs_req_openctx(req),
 905                                                    req_offset(req),
 906                                                    req->wb_bytes,
 907                                                    IOMODE_RW,
 908                                                    false,
 909                                                    GFP_NOFS);
 910                 if (IS_ERR(pgio->pg_lseg)) {
 911                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 912                         pgio->pg_lseg = NULL;
 913                         return;
 914                 }
 915         }
 916         /* If no lseg, fall back to write through mds */
 917         if (pgio->pg_lseg == NULL)
 918                 goto out_mds;
 919
 920         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 921         if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
 922                 goto out_eagain;
 923
 924         for (i = 0; i < pgio->pg_mirror_count; i++) {
 925                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 926                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true);
 927                 if (!ds) {
 928                         if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg))
 929                                 goto out_mds;
 930                         pnfs_generic_pg_cleanup(pgio);
 931                         /* Sleep for 1 second before retrying */
 932                         ssleep(1);
 933                         goto retry;
 934                 }
 935                 pgm = &pgio->pg_mirrors[i];
 936                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 937         }
 938
 939         if (NFS_SERVER(pgio->pg_inode)->flags &
 940                         (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR))
 941                 pgio->pg_maxretrans = io_maxretrans;
 942         return;
 943 out_eagain:
 944         pnfs_generic_pg_cleanup(pgio);
 945         pgio->pg_error = -EAGAIN;
 946         return;
 947 out_mds:
 948         trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode,
 949                         0, NFS4_MAX_UINT64, IOMODE_RW,
 950                         NFS_I(pgio->pg_inode)->layout,
 951                         pgio->pg_lseg);
 952         pgio->pg_maxretrans = 0;
 953         nfs_pageio_reset_write_mds(pgio);
 954         pgio->pg_error = -EAGAIN;
 955 }
 956
 957 static unsigned int
 958 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 959                                     struct nfs_page *req)
 960 {
 961         if (!pgio->pg_lseg) {
 962                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 963                                                    nfs_req_openctx(req),
 964                                                    req_offset(req),
 965                                                    req->wb_bytes,
 966                                                    IOMODE_RW,
 967                                                    false,
 968                                                    GFP_NOFS);
 969                 if (IS_ERR(pgio->pg_lseg)) {
 970                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 971                         pgio->pg_lseg = NULL;
 972                         goto out;
 973                 }
 974         }
 975         if (pgio->pg_lseg)
 976                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 977
 978         trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode,
 979                         0, NFS4_MAX_UINT64, IOMODE_RW,
 980                         NFS_I(pgio->pg_inode)->layout,
 981                         pgio->pg_lseg);
 982         /* no lseg means that pnfs is not in use, so no mirroring here */
 983         nfs_pageio_reset_write_mds(pgio);
 984 out:
 985         return 1;
 986 }
 987
 988 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 989         .pg_init = ff_layout_pg_init_read,
 990         .pg_test = pnfs_generic_pg_test,
 991         .pg_doio = pnfs_generic_pg_readpages,
 992         .pg_cleanup = pnfs_generic_pg_cleanup,
 993 };
 994
 995 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 996         .pg_init = ff_layout_pg_init_write,
 997         .pg_test = pnfs_generic_pg_test,
 998         .pg_doio = pnfs_generic_pg_writepages,
 999         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
1000         .pg_cleanup = pnfs_generic_pg_cleanup,
1001 };
1002
1003 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
1004 {
1005         struct rpc_task *task = &hdr->task;
1006
1007         pnfs_layoutcommit_inode(hdr->inode, false);
1008
1009         if (retry_pnfs) {
1010                 dprintk("%s Reset task %5u for i/o through pNFS "
1011                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1012                         hdr->task.tk_pid,
1013                         hdr->inode->i_sb->s_id,
1014                         (unsigned long long)NFS_FILEID(hdr->inode),
1015                         hdr->args.count,
1016                         (unsigned long long)hdr->args.offset);
1017
1018                 hdr->completion_ops->reschedule_io(hdr);
1019                 return;
1020         }
1021
1022         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1023                 dprintk("%s Reset task %5u for i/o through MDS "
1024                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1025                         hdr->task.tk_pid,
1026                         hdr->inode->i_sb->s_id,
1027                         (unsigned long long)NFS_FILEID(hdr->inode),
1028                         hdr->args.count,
1029                         (unsigned long long)hdr->args.offset);
1030
1031                 trace_pnfs_mds_fallback_write_done(hdr->inode,
1032                                 hdr->args.offset, hdr->args.count,
1033                                 IOMODE_RW, NFS_I(hdr->inode)->layout,
1034                                 hdr->lseg);
1035                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
1036         }
1037 }
1038
1039 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr)
1040 {
1041         u32 idx = hdr->pgio_mirror_idx + 1;
1042         u32 new_idx = 0;
1043
1044         if (ff_layout_choose_any_ds_for_read(hdr->lseg, idx + 1, &new_idx))
1045                 ff_layout_send_layouterror(hdr->lseg);
1046         else
1047                 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1048         pnfs_read_resend_pnfs(hdr, new_idx);
1049 }
1050
1051 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
1052 {
1053         struct rpc_task *task = &hdr->task;
1054
1055         pnfs_layoutcommit_inode(hdr->inode, false);
1056         pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg);
1057
1058         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
1059                 dprintk("%s Reset task %5u for i/o through MDS "
1060                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
1061                         hdr->task.tk_pid,
1062                         hdr->inode->i_sb->s_id,
1063                         (unsigned long long)NFS_FILEID(hdr->inode),
1064                         hdr->args.count,
1065                         (unsigned long long)hdr->args.offset);
1066
1067                 trace_pnfs_mds_fallback_read_done(hdr->inode,
1068                                 hdr->args.offset, hdr->args.count,
1069                                 IOMODE_READ, NFS_I(hdr->inode)->layout,
1070                                 hdr->lseg);
1071                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
1072         }
1073 }
1074
1075 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
1076                                            struct nfs4_state *state,
1077                                            struct nfs_client *clp,
1078                                            struct pnfs_layout_segment *lseg,
1079                                            u32 idx)
1080 {
1081         struct pnfs_layout_hdr *lo = lseg->pls_layout;
1082         struct inode *inode = lo->plh_inode;
1083         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1084         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
1085
1086         switch (task->tk_status) {
1087         case -NFS4ERR_BADSESSION:
1088         case -NFS4ERR_BADSLOT:
1089         case -NFS4ERR_BAD_HIGH_SLOT:
1090         case -NFS4ERR_DEADSESSION:
1091         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1092         case -NFS4ERR_SEQ_FALSE_RETRY:
1093         case -NFS4ERR_SEQ_MISORDERED:
1094                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1095                         "flags 0x%x\n", __func__, task->tk_status,
1096                         clp->cl_exchange_flags);
1097                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1098                 break;
1099         case -NFS4ERR_DELAY:
1100         case -NFS4ERR_GRACE:
1101                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1102                 break;
1103         case -NFS4ERR_RETRY_UNCACHED_REP:
1104                 break;
1105         /* Invalidate Layout errors */
1106         case -NFS4ERR_PNFS_NO_LAYOUT:
1107         case -ESTALE:           /* mapped NFS4ERR_STALE */
1108         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1109         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1110         case -NFS4ERR_FHEXPIRED:
1111         case -NFS4ERR_WRONG_TYPE:
1112                 dprintk("%s Invalid layout error %d\n", __func__,
1113                         task->tk_status);
1114                 /*
1115                  * Destroy layout so new i/o will get a new layout.
1116                  * Layout will not be destroyed until all current lseg
1117                  * references are put. Mark layout as invalid to resend failed
1118                  * i/o and all i/o waiting on the slot table to the MDS until
1119                  * layout is destroyed and a new valid layout is obtained.
1120                  */
1121                 pnfs_destroy_layout(NFS_I(inode));
1122                 rpc_wake_up(&tbl->slot_tbl_waitq);
1123                 goto reset;
1124         /* RPC connection errors */
1125         case -ECONNREFUSED:
1126         case -EHOSTDOWN:
1127         case -EHOSTUNREACH:
1128         case -ENETUNREACH:
1129         case -EIO:
1130         case -ETIMEDOUT:
1131         case -EPIPE:
1132                 dprintk("%s DS connection error %d\n", __func__,
1133                         task->tk_status);
1134                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1135                                 &devid->deviceid);
1136                 rpc_wake_up(&tbl->slot_tbl_waitq);
1137                 fallthrough;
1138         default:
1139                 if (ff_layout_avoid_mds_available_ds(lseg))
1140                         return -NFS4ERR_RESET_TO_PNFS;
1141 reset:
1142                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1143                         task->tk_status);
1144                 return -NFS4ERR_RESET_TO_MDS;
1145         }
1146         task->tk_status = 0;
1147         return -EAGAIN;
1148 }
1149
1150 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1151 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1152                                            struct pnfs_layout_segment *lseg,
1153                                            u32 idx)
1154 {
1155         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1156
1157         switch (task->tk_status) {
1158         /* File access problems. Don't mark the device as unavailable */
1159         case -EACCES:
1160         case -ESTALE:
1161         case -EISDIR:
1162         case -EBADHANDLE:
1163         case -ELOOP:
1164         case -ENOSPC:
1165                 break;
1166         case -EJUKEBOX:
1167                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1168                 goto out_retry;
1169         default:
1170                 dprintk("%s DS connection error %d\n", __func__,
1171                         task->tk_status);
1172                 nfs4_delete_deviceid(devid->ld, devid->nfs_client,
1173                                 &devid->deviceid);
1174         }
1175         /* FIXME: Need to prevent infinite looping here. */
1176         return -NFS4ERR_RESET_TO_PNFS;
1177 out_retry:
1178         task->tk_status = 0;
1179         rpc_restart_call_prepare(task);
1180         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1181         return -EAGAIN;
1182 }
1183
1184 static int ff_layout_async_handle_error(struct rpc_task *task,
1185                                         struct nfs4_state *state,
1186                                         struct nfs_client *clp,
1187                                         struct pnfs_layout_segment *lseg,
1188                                         u32 idx)
1189 {
1190         int vers = clp->cl_nfs_mod->rpc_vers->number;
1191
1192         if (task->tk_status >= 0) {
1193                 ff_layout_mark_ds_reachable(lseg, idx);
1194                 return 0;
1195         }
1196
1197         /* Handle the case of an invalid layout segment */
1198         if (!pnfs_is_valid_lseg(lseg))
1199                 return -NFS4ERR_RESET_TO_PNFS;
1200
1201         switch (vers) {
1202         case 3:
1203                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1204         case 4:
1205                 return ff_layout_async_handle_error_v4(task, state, clp,
1206                                                        lseg, idx);
1207         default:
1208                 /* should never happen */
1209                 WARN_ON_ONCE(1);
1210                 return 0;
1211         }
1212 }
1213
1214 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1215                                         u32 idx, u64 offset, u64 length,
1216                                         u32 *op_status, int opnum, int error)
1217 {
1218         struct nfs4_ff_layout_mirror *mirror;
1219         u32 status = *op_status;
1220         int err;
1221
1222         if (status == 0) {
1223                 switch (error) {
1224                 case -ETIMEDOUT:
1225                 case -EPFNOSUPPORT:
1226                 case -EPROTONOSUPPORT:
1227                 case -EOPNOTSUPP:
1228                 case -ECONNREFUSED:
1229                 case -ECONNRESET:
1230                 case -EHOSTDOWN:
1231                 case -EHOSTUNREACH:
1232                 case -ENETUNREACH:
1233                 case -EADDRINUSE:
1234                 case -ENOBUFS:
1235                 case -EPIPE:
1236                 case -EPERM:
1237                         *op_status = status = NFS4ERR_NXIO;
1238                         break;
1239                 case -EACCES:
1240                         *op_status = status = NFS4ERR_ACCESS;
1241                         break;
1242                 default:
1243                         return;
1244                 }
1245         }
1246
1247         mirror = FF_LAYOUT_COMP(lseg, idx);
1248         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1249                                        mirror, offset, length, status, opnum,
1250                                        GFP_NOIO);
1251
1252         switch (status) {
1253         case NFS4ERR_DELAY:
1254         case NFS4ERR_GRACE:
1255                 break;
1256         case NFS4ERR_NXIO:
1257                 ff_layout_mark_ds_unreachable(lseg, idx);
1258                 /*
1259                  * Don't return the layout if this is a read and we still
1260                  * have layouts to try
1261                  */
1262                 if (opnum == OP_READ)
1263                         break;
1264                 fallthrough;
1265         default:
1266                 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode,
1267                                                   lseg);
1268         }
1269
1270         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1271 }
1272
1273 /* NFS_PROTO call done callback routines */
1274 static int ff_layout_read_done_cb(struct rpc_task *task,
1275                                 struct nfs_pgio_header *hdr)
1276 {
1277         int err;
1278
1279         if (task->tk_status < 0) {
1280                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1281                                             hdr->args.offset, hdr->args.count,
1282                                             &hdr->res.op_status, OP_READ,
1283                                             task->tk_status);
1284                 trace_ff_layout_read_error(hdr);
1285         }
1286
1287         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1288                                            hdr->ds_clp, hdr->lseg,
1289                                            hdr->pgio_mirror_idx);
1290
1291         trace_nfs4_pnfs_read(hdr, err);
1292         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1293         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1294         switch (err) {
1295         case -NFS4ERR_RESET_TO_PNFS:
1296                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1297                 return task->tk_status;
1298         case -NFS4ERR_RESET_TO_MDS:
1299                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1300                 return task->tk_status;
1301         case -EAGAIN:
1302                 goto out_eagain;
1303         }
1304
1305         return 0;
1306 out_eagain:
1307         rpc_restart_call_prepare(task);
1308         return -EAGAIN;
1309 }
1310
1311 static bool
1312 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1313 {
1314         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1315 }
1316
1317 /*
1318  * We reference the rpc_cred of the first WRITE that triggers the need for
1319  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1320  * rfc5661 is not clear about which credential should be used.
1321  *
1322  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1323  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1324  * we always send layoutcommit after DS writes.
1325  */
1326 static void
1327 ff_layout_set_layoutcommit(struct inode *inode,
1328                 struct pnfs_layout_segment *lseg,
1329                 loff_t end_offset)
1330 {
1331         if (!ff_layout_need_layoutcommit(lseg))
1332                 return;
1333
1334         pnfs_set_layoutcommit(inode, lseg, end_offset);
1335         dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino,
1336                 (unsigned long long) NFS_I(inode)->layout->plh_lwb);
1337 }
1338
1339 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task,
1340                 struct nfs_pgio_header *hdr)
1341 {
1342         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1343                 return;
1344         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1345                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1346                         hdr->args.count,
1347                         task->tk_start);
1348 }
1349
1350 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
1351                 struct nfs_pgio_header *hdr)
1352 {
1353         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1354                 return;
1355         nfs4_ff_layout_stat_io_end_read(task,
1356                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1357                         hdr->args.count,
1358                         hdr->res.count);
1359         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1360 }
1361
1362 static int ff_layout_read_prepare_common(struct rpc_task *task,
1363                                          struct nfs_pgio_header *hdr)
1364 {
1365         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1366                 rpc_exit(task, -EIO);
1367                 return -EIO;
1368         }
1369
1370         ff_layout_read_record_layoutstats_start(task, hdr);
1371         return 0;
1372 }
1373
1374 /*
1375  * Call ops for the async read/write cases
1376  * In the case of dense layouts, the offset needs to be reset to its
1377  * original value.
1378  */
1379 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1380 {
1381         struct nfs_pgio_header *hdr = data;
1382
1383         if (ff_layout_read_prepare_common(task, hdr))
1384                 return;
1385
1386         rpc_call_start(task);
1387 }
1388
1389 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1390 {
1391         struct nfs_pgio_header *hdr = data;
1392
1393         if (nfs4_setup_sequence(hdr->ds_clp,
1394                                 &hdr->args.seq_args,
1395                                 &hdr->res.seq_res,
1396                                 task))
1397                 return;
1398
1399         ff_layout_read_prepare_common(task, hdr);
1400 }
1401
1402 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1403 {
1404         struct nfs_pgio_header *hdr = data;
1405
1406         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1407
1408         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1409             task->tk_status == 0) {
1410                 nfs4_sequence_done(task, &hdr->res.seq_res);
1411                 return;
1412         }
1413
1414         /* Note this may cause RPC to be resent */
1415         hdr->mds_ops->rpc_call_done(task, hdr);
1416 }
1417
1418 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1419 {
1420         struct nfs_pgio_header *hdr = data;
1421
1422         ff_layout_read_record_layoutstats_done(task, hdr);
1423         rpc_count_iostats_metrics(task,
1424             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1425 }
1426
1427 static void ff_layout_read_release(void *data)
1428 {
1429         struct nfs_pgio_header *hdr = data;
1430
1431         ff_layout_read_record_layoutstats_done(&hdr->task, hdr);
1432         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags))
1433                 ff_layout_resend_pnfs_read(hdr);
1434         else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1435                 ff_layout_reset_read(hdr);
1436         pnfs_generic_rw_release(data);
1437 }
1438
1439
1440 static int ff_layout_write_done_cb(struct rpc_task *task,
1441                                 struct nfs_pgio_header *hdr)
1442 {
1443         loff_t end_offs = 0;
1444         int err;
1445
1446         if (task->tk_status < 0) {
1447                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1448                                             hdr->args.offset, hdr->args.count,
1449                                             &hdr->res.op_status, OP_WRITE,
1450                                             task->tk_status);
1451                 trace_ff_layout_write_error(hdr);
1452         }
1453
1454         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1455                                            hdr->ds_clp, hdr->lseg,
1456                                            hdr->pgio_mirror_idx);
1457
1458         trace_nfs4_pnfs_write(hdr, err);
1459         clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1460         clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1461         switch (err) {
1462         case -NFS4ERR_RESET_TO_PNFS:
1463                 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags);
1464                 return task->tk_status;
1465         case -NFS4ERR_RESET_TO_MDS:
1466                 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags);
1467                 return task->tk_status;
1468         case -EAGAIN:
1469                 return -EAGAIN;
1470         }
1471
1472         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1473             hdr->res.verf->committed == NFS_DATA_SYNC)
1474                 end_offs = hdr->mds_offset + (loff_t)hdr->res.count;
1475
1476         /* Note: if the write is unstable, don't set end_offs until commit */
1477         ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs);
1478
1479         /* zero out fattr since we don't care DS attr at all */
1480         hdr->fattr.valid = 0;
1481         if (task->tk_status >= 0)
1482                 nfs_writeback_update_inode(hdr);
1483
1484         return 0;
1485 }
1486
1487 static int ff_layout_commit_done_cb(struct rpc_task *task,
1488                                      struct nfs_commit_data *data)
1489 {
1490         int err;
1491
1492         if (task->tk_status < 0) {
1493                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1494                                             data->args.offset, data->args.count,
1495                                             &data->res.op_status, OP_COMMIT,
1496                                             task->tk_status);
1497                 trace_ff_layout_commit_error(data);
1498         }
1499
1500         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1501                                            data->lseg, data->ds_commit_index);
1502
1503         trace_nfs4_pnfs_commit_ds(data, err);
1504         switch (err) {
1505         case -NFS4ERR_RESET_TO_PNFS:
1506                 pnfs_generic_prepare_to_resend_writes(data);
1507                 return -EAGAIN;
1508         case -NFS4ERR_RESET_TO_MDS:
1509                 pnfs_generic_prepare_to_resend_writes(data);
1510                 return -EAGAIN;
1511         case -EAGAIN:
1512                 rpc_restart_call_prepare(task);
1513                 return -EAGAIN;
1514         }
1515
1516         ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb);
1517
1518         return 0;
1519 }
1520
1521 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task,
1522                 struct nfs_pgio_header *hdr)
1523 {
1524         if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags))
1525                 return;
1526         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1527                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1528                         hdr->args.count,
1529                         task->tk_start);
1530 }
1531
1532 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task,
1533                 struct nfs_pgio_header *hdr)
1534 {
1535         if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags))
1536                 return;
1537         nfs4_ff_layout_stat_io_end_write(task,
1538                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1539                         hdr->args.count, hdr->res.count,
1540                         hdr->res.verf->committed);
1541         set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags);
1542 }
1543
1544 static int ff_layout_write_prepare_common(struct rpc_task *task,
1545                                           struct nfs_pgio_header *hdr)
1546 {
1547         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1548                 rpc_exit(task, -EIO);
1549                 return -EIO;
1550         }
1551
1552         ff_layout_write_record_layoutstats_start(task, hdr);
1553         return 0;
1554 }
1555
1556 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1557 {
1558         struct nfs_pgio_header *hdr = data;
1559
1560         if (ff_layout_write_prepare_common(task, hdr))
1561                 return;
1562
1563         rpc_call_start(task);
1564 }
1565
1566 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1567 {
1568         struct nfs_pgio_header *hdr = data;
1569
1570         if (nfs4_setup_sequence(hdr->ds_clp,
1571                                 &hdr->args.seq_args,
1572                                 &hdr->res.seq_res,
1573                                 task))
1574                 return;
1575
1576         ff_layout_write_prepare_common(task, hdr);
1577 }
1578
1579 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1580 {
1581         struct nfs_pgio_header *hdr = data;
1582
1583         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1584             task->tk_status == 0) {
1585                 nfs4_sequence_done(task, &hdr->res.seq_res);
1586                 return;
1587         }
1588
1589         /* Note this may cause RPC to be resent */
1590         hdr->mds_ops->rpc_call_done(task, hdr);
1591 }
1592
1593 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1594 {
1595         struct nfs_pgio_header *hdr = data;
1596
1597         ff_layout_write_record_layoutstats_done(task, hdr);
1598         rpc_count_iostats_metrics(task,
1599             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1600 }
1601
1602 static void ff_layout_write_release(void *data)
1603 {
1604         struct nfs_pgio_header *hdr = data;
1605
1606         ff_layout_write_record_layoutstats_done(&hdr->task, hdr);
1607         if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) {
1608                 ff_layout_send_layouterror(hdr->lseg);
1609                 ff_layout_reset_write(hdr, true);
1610         } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags))
1611                 ff_layout_reset_write(hdr, false);
1612         pnfs_generic_rw_release(data);
1613 }
1614
1615 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task,
1616                 struct nfs_commit_data *cdata)
1617 {
1618         if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags))
1619                 return;
1620         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1621                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1622                         0, task->tk_start);
1623 }
1624
1625 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task,
1626                 struct nfs_commit_data *cdata)
1627 {
1628         struct nfs_page *req;
1629         __u64 count = 0;
1630
1631         if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags))
1632                 return;
1633
1634         if (task->tk_status == 0) {
1635                 list_for_each_entry(req, &cdata->pages, wb_list)
1636                         count += req->wb_bytes;
1637         }
1638         nfs4_ff_layout_stat_io_end_write(task,
1639                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1640                         count, count, NFS_FILE_SYNC);
1641         set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags);
1642 }
1643
1644 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1645                 struct nfs_commit_data *cdata)
1646 {
1647         ff_layout_commit_record_layoutstats_start(task, cdata);
1648 }
1649
1650 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1651 {
1652         ff_layout_commit_prepare_common(task, data);
1653         rpc_call_start(task);
1654 }
1655
1656 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1657 {
1658         struct nfs_commit_data *wdata = data;
1659
1660         if (nfs4_setup_sequence(wdata->ds_clp,
1661                                 &wdata->args.seq_args,
1662                                 &wdata->res.seq_res,
1663                                 task))
1664                 return;
1665         ff_layout_commit_prepare_common(task, data);
1666 }
1667
1668 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1669 {
1670         pnfs_generic_write_commit_done(task, data);
1671 }
1672
1673 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1674 {
1675         struct nfs_commit_data *cdata = data;
1676
1677         ff_layout_commit_record_layoutstats_done(task, cdata);
1678         rpc_count_iostats_metrics(task,
1679             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1680 }
1681
1682 static void ff_layout_commit_release(void *data)
1683 {
1684         struct nfs_commit_data *cdata = data;
1685
1686         ff_layout_commit_record_layoutstats_done(&cdata->task, cdata);
1687         pnfs_generic_commit_release(data);
1688 }
1689
1690 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1691         .rpc_call_prepare = ff_layout_read_prepare_v3,
1692         .rpc_call_done = ff_layout_read_call_done,
1693         .rpc_count_stats = ff_layout_read_count_stats,
1694         .rpc_release = ff_layout_read_release,
1695 };
1696
1697 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1698         .rpc_call_prepare = ff_layout_read_prepare_v4,
1699         .rpc_call_done = ff_layout_read_call_done,
1700         .rpc_count_stats = ff_layout_read_count_stats,
1701         .rpc_release = ff_layout_read_release,
1702 };
1703
1704 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1705         .rpc_call_prepare = ff_layout_write_prepare_v3,
1706         .rpc_call_done = ff_layout_write_call_done,
1707         .rpc_count_stats = ff_layout_write_count_stats,
1708         .rpc_release = ff_layout_write_release,
1709 };
1710
1711 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1712         .rpc_call_prepare = ff_layout_write_prepare_v4,
1713         .rpc_call_done = ff_layout_write_call_done,
1714         .rpc_count_stats = ff_layout_write_count_stats,
1715         .rpc_release = ff_layout_write_release,
1716 };
1717
1718 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1719         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1720         .rpc_call_done = ff_layout_commit_done,
1721         .rpc_count_stats = ff_layout_commit_count_stats,
1722         .rpc_release = ff_layout_commit_release,
1723 };
1724
1725 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1726         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1727         .rpc_call_done = ff_layout_commit_done,
1728         .rpc_count_stats = ff_layout_commit_count_stats,
1729         .rpc_release = ff_layout_commit_release,
1730 };
1731
1732 static enum pnfs_try_status
1733 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1734 {
1735         struct pnfs_layout_segment *lseg = hdr->lseg;
1736         struct nfs4_pnfs_ds *ds;
1737         struct rpc_clnt *ds_clnt;
1738         struct nfs4_ff_layout_mirror *mirror;
1739         const struct cred *ds_cred;
1740         loff_t offset = hdr->args.offset;
1741         u32 idx = hdr->pgio_mirror_idx;
1742         int vers;
1743         struct nfs_fh *fh;
1744
1745         dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n",
1746                 __func__, hdr->inode->i_ino,
1747                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1748
1749         mirror = FF_LAYOUT_COMP(lseg, idx);
1750         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false);
1751         if (!ds)
1752                 goto out_failed;
1753
1754         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1755                                                    hdr->inode);
1756         if (IS_ERR(ds_clnt))
1757                 goto out_failed;
1758
1759         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1760         if (!ds_cred)
1761                 goto out_failed;
1762
1763         vers = nfs4_ff_layout_ds_version(mirror);
1764
1765         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1766                 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers);
1767
1768         hdr->pgio_done_cb = ff_layout_read_done_cb;
1769         refcount_inc(&ds->ds_clp->cl_count);
1770         hdr->ds_clp = ds->ds_clp;
1771         fh = nfs4_ff_layout_select_ds_fh(mirror);
1772         if (fh)
1773                 hdr->args.fh = fh;
1774
1775         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1776
1777         /*
1778          * Note that if we ever decide to split across DSes,
1779          * then we may need to handle dense-like offsets.
1780          */
1781         hdr->args.offset = offset;
1782         hdr->mds_offset = offset;
1783
1784         /* Perform an asynchronous read to ds */
1785         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1786                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1787                                       &ff_layout_read_call_ops_v4,
1788                           0, RPC_TASK_SOFTCONN);
1789         put_cred(ds_cred);
1790         return PNFS_ATTEMPTED;
1791
1792 out_failed:
1793         if (ff_layout_avoid_mds_available_ds(lseg))
1794                 return PNFS_TRY_AGAIN;
1795         trace_pnfs_mds_fallback_read_pagelist(hdr->inode,
1796                         hdr->args.offset, hdr->args.count,
1797                         IOMODE_READ, NFS_I(hdr->inode)->layout, lseg);
1798         return PNFS_NOT_ATTEMPTED;
1799 }
1800
1801 /* Perform async writes. */
1802 static enum pnfs_try_status
1803 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1804 {
1805         struct pnfs_layout_segment *lseg = hdr->lseg;
1806         struct nfs4_pnfs_ds *ds;
1807         struct rpc_clnt *ds_clnt;
1808         struct nfs4_ff_layout_mirror *mirror;
1809         const struct cred *ds_cred;
1810         loff_t offset = hdr->args.offset;
1811         int vers;
1812         struct nfs_fh *fh;
1813         u32 idx = hdr->pgio_mirror_idx;
1814
1815         mirror = FF_LAYOUT_COMP(lseg, idx);
1816         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1817         if (!ds)
1818                 goto out_failed;
1819
1820         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1821                                                    hdr->inode);
1822         if (IS_ERR(ds_clnt))
1823                 goto out_failed;
1824
1825         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred);
1826         if (!ds_cred)
1827                 goto out_failed;
1828
1829         vers = nfs4_ff_layout_ds_version(mirror);
1830
1831         dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n",
1832                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1833                 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count),
1834                 vers);
1835
1836         hdr->pgio_done_cb = ff_layout_write_done_cb;
1837         refcount_inc(&ds->ds_clp->cl_count);
1838         hdr->ds_clp = ds->ds_clp;
1839         hdr->ds_commit_idx = idx;
1840         fh = nfs4_ff_layout_select_ds_fh(mirror);
1841         if (fh)
1842                 hdr->args.fh = fh;
1843
1844         nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid);
1845
1846         /*
1847          * Note that if we ever decide to split across DSes,
1848          * then we may need to handle dense-like offsets.
1849          */
1850         hdr->args.offset = offset;
1851
1852         /* Perform an asynchronous write */
1853         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1854                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1855                                       &ff_layout_write_call_ops_v4,
1856                           sync, RPC_TASK_SOFTCONN);
1857         put_cred(ds_cred);
1858         return PNFS_ATTEMPTED;
1859
1860 out_failed:
1861         if (ff_layout_avoid_mds_available_ds(lseg))
1862                 return PNFS_TRY_AGAIN;
1863         trace_pnfs_mds_fallback_write_pagelist(hdr->inode,
1864                         hdr->args.offset, hdr->args.count,
1865                         IOMODE_RW, NFS_I(hdr->inode)->layout, lseg);
1866         return PNFS_NOT_ATTEMPTED;
1867 }
1868
1869 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1870 {
1871         return i;
1872 }
1873
1874 static struct nfs_fh *
1875 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1876 {
1877         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1878
1879         /* FIXME: Assume that there is only one NFS version available
1880          * for the DS.
1881          */
1882         return &flseg->mirror_array[i]->fh_versions[0];
1883 }
1884
1885 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1886 {
1887         struct pnfs_layout_segment *lseg = data->lseg;
1888         struct nfs4_pnfs_ds *ds;
1889         struct rpc_clnt *ds_clnt;
1890         struct nfs4_ff_layout_mirror *mirror;
1891         const struct cred *ds_cred;
1892         u32 idx;
1893         int vers, ret;
1894         struct nfs_fh *fh;
1895
1896         if (!lseg || !(pnfs_is_valid_lseg(lseg) ||
1897             test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags)))
1898                 goto out_err;
1899
1900         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1901         mirror = FF_LAYOUT_COMP(lseg, idx);
1902         ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true);
1903         if (!ds)
1904                 goto out_err;
1905
1906         ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp,
1907                                                    data->inode);
1908         if (IS_ERR(ds_clnt))
1909                 goto out_err;
1910
1911         ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred);
1912         if (!ds_cred)
1913                 goto out_err;
1914
1915         vers = nfs4_ff_layout_ds_version(mirror);
1916
1917         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1918                 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count),
1919                 vers);
1920         data->commit_done_cb = ff_layout_commit_done_cb;
1921         data->cred = ds_cred;
1922         refcount_inc(&ds->ds_clp->cl_count);
1923         data->ds_clp = ds->ds_clp;
1924         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1925         if (fh)
1926                 data->args.fh = fh;
1927
1928         ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1929                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1930                                                &ff_layout_commit_call_ops_v4,
1931                                    how, RPC_TASK_SOFTCONN);
1932         put_cred(ds_cred);
1933         return ret;
1934 out_err:
1935         pnfs_generic_prepare_to_resend_writes(data);
1936         pnfs_generic_commit_release(data);
1937         return -EAGAIN;
1938 }
1939
1940 static int
1941 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1942                            int how, struct nfs_commit_info *cinfo)
1943 {
1944         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1945                                             ff_layout_initiate_commit);
1946 }
1947
1948 static struct pnfs_ds_commit_info *
1949 ff_layout_get_ds_info(struct inode *inode)
1950 {
1951         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1952
1953         if (layout == NULL)
1954                 return NULL;
1955
1956         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1957 }
1958
1959 static void
1960 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1961                 struct pnfs_layout_segment *lseg)
1962 {
1963         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1964         struct inode *inode = lseg->pls_layout->plh_inode;
1965         struct pnfs_commit_array *array, *new;
1966
1967         new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, GFP_NOIO);
1968         if (new) {
1969                 spin_lock(&inode->i_lock);
1970                 array = pnfs_add_commit_array(fl_cinfo, new, lseg);
1971                 spin_unlock(&inode->i_lock);
1972                 if (array != new)
1973                         pnfs_free_commit_array(new);
1974         }
1975 }
1976
1977 static void
1978 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo,
1979                 struct inode *inode)
1980 {
1981         spin_lock(&inode->i_lock);
1982         pnfs_generic_ds_cinfo_destroy(fl_cinfo);
1983         spin_unlock(&inode->i_lock);
1984 }
1985
1986 static void
1987 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1988 {
1989         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1990                                                   id_node));
1991 }
1992
1993 static int ff_layout_encode_ioerr(struct xdr_stream *xdr,
1994                                   const struct nfs4_layoutreturn_args *args,
1995                                   const struct nfs4_flexfile_layoutreturn_args *ff_args)
1996 {
1997         __be32 *start;
1998
1999         start = xdr_reserve_space(xdr, 4);
2000         if (unlikely(!start))
2001                 return -E2BIG;
2002
2003         *start = cpu_to_be32(ff_args->num_errors);
2004         /* This assume we always return _ALL_ layouts */
2005         return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors);
2006 }
2007
2008 static void
2009 encode_opaque_fixed(struct xdr_stream *xdr, const void *buf, size_t len)
2010 {
2011         WARN_ON_ONCE(xdr_stream_encode_opaque_fixed(xdr, buf, len) < 0);
2012 }
2013
2014 static void
2015 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr,
2016                             const nfs4_stateid *stateid,
2017                             const struct nfs42_layoutstat_devinfo *devinfo)
2018 {
2019         __be32 *p;
2020
2021         p = xdr_reserve_space(xdr, 8 + 8);
2022         p = xdr_encode_hyper(p, devinfo->offset);
2023         p = xdr_encode_hyper(p, devinfo->length);
2024         encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE);
2025         p = xdr_reserve_space(xdr, 4*8);
2026         p = xdr_encode_hyper(p, devinfo->read_count);
2027         p = xdr_encode_hyper(p, devinfo->read_bytes);
2028         p = xdr_encode_hyper(p, devinfo->write_count);
2029         p = xdr_encode_hyper(p, devinfo->write_bytes);
2030         encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE);
2031 }
2032
2033 static void
2034 ff_layout_encode_ff_iostat(struct xdr_stream *xdr,
2035                             const nfs4_stateid *stateid,
2036                             const struct nfs42_layoutstat_devinfo *devinfo)
2037 {
2038         ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo);
2039         ff_layout_encode_ff_layoutupdate(xdr, devinfo,
2040                         devinfo->ld_private.data);
2041 }
2042
2043 /* report nothing for now */
2044 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr,
2045                 const struct nfs4_layoutreturn_args *args,
2046                 struct nfs4_flexfile_layoutreturn_args *ff_args)
2047 {
2048         __be32 *p;
2049         int i;
2050
2051         p = xdr_reserve_space(xdr, 4);
2052         *p = cpu_to_be32(ff_args->num_dev);
2053         for (i = 0; i < ff_args->num_dev; i++)
2054                 ff_layout_encode_ff_iostat(xdr,
2055                                 &args->layout->plh_stateid,
2056                                 &ff_args->devinfo[i]);
2057 }
2058
2059 static void
2060 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo,
2061                 unsigned int num_entries)
2062 {
2063         unsigned int i;
2064
2065         for (i = 0; i < num_entries; i++) {
2066                 if (!devinfo[i].ld_private.ops)
2067                         continue;
2068                 if (!devinfo[i].ld_private.ops->free)
2069                         continue;
2070                 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private);
2071         }
2072 }
2073
2074 static struct nfs4_deviceid_node *
2075 ff_layout_alloc_deviceid_node(struct nfs_server *server,
2076                               struct pnfs_device *pdev, gfp_t gfp_flags)
2077 {
2078         struct nfs4_ff_layout_ds *dsaddr;
2079
2080         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
2081         if (!dsaddr)
2082                 return NULL;
2083         return &dsaddr->id_node;
2084 }
2085
2086 static void
2087 ff_layout_encode_layoutreturn(struct xdr_stream *xdr,
2088                 const void *voidargs,
2089                 const struct nfs4_xdr_opaque_data *ff_opaque)
2090 {
2091         const struct nfs4_layoutreturn_args *args = voidargs;
2092         struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data;
2093         struct xdr_buf tmp_buf = {
2094                 .head = {
2095                         [0] = {
2096                                 .iov_base = page_address(ff_args->pages[0]),
2097                         },
2098                 },
2099                 .buflen = PAGE_SIZE,
2100         };
2101         struct xdr_stream tmp_xdr;
2102         __be32 *start;
2103
2104         dprintk("%s: Begin\n", __func__);
2105
2106         xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL);
2107
2108         ff_layout_encode_ioerr(&tmp_xdr, args, ff_args);
2109         ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args);
2110
2111         start = xdr_reserve_space(xdr, 4);
2112         *start = cpu_to_be32(tmp_buf.len);
2113         xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len);
2114
2115         dprintk("%s: Return\n", __func__);
2116 }
2117
2118 static void
2119 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args)
2120 {
2121         struct nfs4_flexfile_layoutreturn_args *ff_args;
2122
2123         if (!args->data)
2124                 return;
2125         ff_args = args->data;
2126         args->data = NULL;
2127
2128         ff_layout_free_ds_ioerr(&ff_args->errors);
2129         ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev);
2130
2131         put_page(ff_args->pages[0]);
2132         kfree(ff_args);
2133 }
2134
2135 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = {
2136         .encode = ff_layout_encode_layoutreturn,
2137         .free = ff_layout_free_layoutreturn,
2138 };
2139
2140 static int
2141 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args)
2142 {
2143         struct nfs4_flexfile_layoutreturn_args *ff_args;
2144         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout);
2145
2146         ff_args = kmalloc(sizeof(*ff_args), GFP_KERNEL);
2147         if (!ff_args)
2148                 goto out_nomem;
2149         ff_args->pages[0] = alloc_page(GFP_KERNEL);
2150         if (!ff_args->pages[0])
2151                 goto out_nomem_free;
2152
2153         INIT_LIST_HEAD(&ff_args->errors);
2154         ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout,
2155                         &args->range, &ff_args->errors,
2156                         FF_LAYOUTRETURN_MAXERR);
2157
2158         spin_lock(&args->inode->i_lock);
2159         ff_args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2160                         &ff_args->devinfo[0], ARRAY_SIZE(ff_args->devinfo));
2161         spin_unlock(&args->inode->i_lock);
2162
2163         args->ld_private->ops = &layoutreturn_ops;
2164         args->ld_private->data = ff_args;
2165         return 0;
2166 out_nomem_free:
2167         kfree(ff_args);
2168 out_nomem:
2169         return -ENOMEM;
2170 }
2171
2172 #ifdef CONFIG_NFS_V4_2
2173 void
2174 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2175 {
2176         struct pnfs_layout_hdr *lo = lseg->pls_layout;
2177         struct nfs42_layout_error *errors;
2178         LIST_HEAD(head);
2179
2180         if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR))
2181                 return;
2182         ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1);
2183         if (list_empty(&head))
2184                 return;
2185
2186         errors = kmalloc_array(NFS42_LAYOUTERROR_MAX,
2187                         sizeof(*errors), GFP_NOFS);
2188         if (errors != NULL) {
2189                 const struct nfs4_ff_layout_ds_err *pos;
2190                 size_t n = 0;
2191
2192                 list_for_each_entry(pos, &head, list) {
2193                         errors[n].offset = pos->offset;
2194                         errors[n].length = pos->length;
2195                         nfs4_stateid_copy(&errors[n].stateid, &pos->stateid);
2196                         errors[n].errors[0].dev_id = pos->deviceid;
2197                         errors[n].errors[0].status = pos->status;
2198                         errors[n].errors[0].opnum = pos->opnum;
2199                         n++;
2200                         if (!list_is_last(&pos->list, &head) &&
2201                             n < NFS42_LAYOUTERROR_MAX)
2202                                 continue;
2203                         if (nfs42_proc_layouterror(lseg, errors, n) < 0)
2204                                 break;
2205                         n = 0;
2206                 }
2207                 kfree(errors);
2208         }
2209         ff_layout_free_ds_ioerr(&head);
2210 }
2211 #else
2212 void
2213 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg)
2214 {
2215 }
2216 #endif
2217
2218 static int
2219 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
2220 {
2221         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
2222
2223         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
2224 }
2225
2226 static size_t
2227 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
2228                           const int buflen)
2229 {
2230         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
2231         const struct in6_addr *addr = &sin6->sin6_addr;
2232
2233         /*
2234          * RFC 4291, Section 2.2.2
2235          *
2236          * Shorthanded ANY address
2237          */
2238         if (ipv6_addr_any(addr))
2239                 return snprintf(buf, buflen, "::");
2240
2241         /*
2242          * RFC 4291, Section 2.2.2
2243          *
2244          * Shorthanded loopback address
2245          */
2246         if (ipv6_addr_loopback(addr))
2247                 return snprintf(buf, buflen, "::1");
2248
2249         /*
2250          * RFC 4291, Section 2.2.3
2251          *
2252          * Special presentation address format for mapped v4
2253          * addresses.
2254          */
2255         if (ipv6_addr_v4mapped(addr))
2256                 return snprintf(buf, buflen, "::ffff:%pI4",
2257                                         &addr->s6_addr32[3]);
2258
2259         /*
2260          * RFC 4291, Section 2.2.1
2261          */
2262         return snprintf(buf, buflen, "%pI6c", addr);
2263 }
2264
2265 /* Derived from rpc_sockaddr2uaddr */
2266 static void
2267 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
2268 {
2269         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
2270         char portbuf[RPCBIND_MAXUADDRPLEN];
2271         char addrbuf[RPCBIND_MAXUADDRLEN];
2272         char *netid;
2273         unsigned short port;
2274         int len, netid_len;
2275         __be32 *p;
2276
2277         switch (sap->sa_family) {
2278         case AF_INET:
2279                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
2280                         return;
2281                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
2282                 netid = "tcp";
2283                 netid_len = 3;
2284                 break;
2285         case AF_INET6:
2286                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
2287                         return;
2288                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
2289                 netid = "tcp6";
2290                 netid_len = 4;
2291                 break;
2292         default:
2293                 /* we only support tcp and tcp6 */
2294                 WARN_ON_ONCE(1);
2295                 return;
2296         }
2297
2298         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
2299         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
2300
2301         p = xdr_reserve_space(xdr, 4 + netid_len);
2302         xdr_encode_opaque(p, netid, netid_len);
2303
2304         p = xdr_reserve_space(xdr, 4 + len);
2305         xdr_encode_opaque(p, addrbuf, len);
2306 }
2307
2308 static void
2309 ff_layout_encode_nfstime(struct xdr_stream *xdr,
2310                          ktime_t t)
2311 {
2312         struct timespec64 ts;
2313         __be32 *p;
2314
2315         p = xdr_reserve_space(xdr, 12);
2316         ts = ktime_to_timespec64(t);
2317         p = xdr_encode_hyper(p, ts.tv_sec);
2318         *p++ = cpu_to_be32(ts.tv_nsec);
2319 }
2320
2321 static void
2322 ff_layout_encode_io_latency(struct xdr_stream *xdr,
2323                             struct nfs4_ff_io_stat *stat)
2324 {
2325         __be32 *p;
2326
2327         p = xdr_reserve_space(xdr, 5 * 8);
2328         p = xdr_encode_hyper(p, stat->ops_requested);
2329         p = xdr_encode_hyper(p, stat->bytes_requested);
2330         p = xdr_encode_hyper(p, stat->ops_completed);
2331         p = xdr_encode_hyper(p, stat->bytes_completed);
2332         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2333         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2334         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2335 }
2336
2337 static void
2338 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr,
2339                               const struct nfs42_layoutstat_devinfo *devinfo,
2340                               struct nfs4_ff_layout_mirror *mirror)
2341 {
2342         struct nfs4_pnfs_ds_addr *da;
2343         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2344         struct nfs_fh *fh = &mirror->fh_versions[0];
2345         __be32 *p;
2346
2347         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2348         dprintk("%s: DS %s: encoding address %s\n",
2349                 __func__, ds->ds_remotestr, da->da_remotestr);
2350         /* netaddr4 */
2351         ff_layout_encode_netaddr(xdr, da);
2352         /* nfs_fh4 */
2353         p = xdr_reserve_space(xdr, 4 + fh->size);
2354         xdr_encode_opaque(p, fh->data, fh->size);
2355         /* ff_io_latency4 read */
2356         spin_lock(&mirror->lock);
2357         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2358         /* ff_io_latency4 write */
2359         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2360         spin_unlock(&mirror->lock);
2361         /* nfstime4 */
2362         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2363         /* bool */
2364         p = xdr_reserve_space(xdr, 4);
2365         *p = cpu_to_be32(false);
2366 }
2367
2368 static void
2369 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args,
2370                              const struct nfs4_xdr_opaque_data *opaque)
2371 {
2372         struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque,
2373                         struct nfs42_layoutstat_devinfo, ld_private);
2374         __be32 *start;
2375
2376         /* layoutupdate length */
2377         start = xdr_reserve_space(xdr, 4);
2378         ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data);
2379
2380         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2381 }
2382
2383 static void
2384 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque)
2385 {
2386         struct nfs4_ff_layout_mirror *mirror = opaque->data;
2387
2388         ff_layout_put_mirror(mirror);
2389 }
2390
2391 static const struct nfs4_xdr_opaque_ops layoutstat_ops = {
2392         .encode = ff_layout_encode_layoutstats,
2393         .free   = ff_layout_free_layoutstats,
2394 };
2395
2396 static int
2397 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
2398                                struct nfs42_layoutstat_devinfo *devinfo,
2399                                int dev_limit)
2400 {
2401         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2402         struct nfs4_ff_layout_mirror *mirror;
2403         struct nfs4_deviceid_node *dev;
2404         int i = 0;
2405
2406         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2407                 if (i >= dev_limit)
2408                         break;
2409                 if (IS_ERR_OR_NULL(mirror->mirror_ds))
2410                         continue;
2411                 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags))
2412                         continue;
2413                 /* mirror refcount put in cleanup_layoutstats */
2414                 if (!refcount_inc_not_zero(&mirror->ref))
2415                         continue;
2416                 dev = &mirror->mirror_ds->id_node;
2417                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2418                 devinfo->offset = 0;
2419                 devinfo->length = NFS4_MAX_UINT64;
2420                 spin_lock(&mirror->lock);
2421                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2422                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2423                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2424                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2425                 spin_unlock(&mirror->lock);
2426                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2427                 devinfo->ld_private.ops = &layoutstat_ops;
2428                 devinfo->ld_private.data = mirror;
2429
2430                 devinfo++;
2431                 i++;
2432         }
2433         return i;
2434 }
2435
2436 static int
2437 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2438 {
2439         struct nfs4_flexfile_layout *ff_layout;
2440         const int dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2441
2442         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2443         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2444         if (!args->devinfo)
2445                 return -ENOMEM;
2446
2447         spin_lock(&args->inode->i_lock);
2448         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2449         args->num_dev = ff_layout_mirror_prepare_stats(&ff_layout->generic_hdr,
2450                         &args->devinfo[0], dev_count);
2451         spin_unlock(&args->inode->i_lock);
2452         if (!args->num_dev) {
2453                 kfree(args->devinfo);
2454                 args->devinfo = NULL;
2455                 return -ENOENT;
2456         }
2457
2458         return 0;
2459 }
2460
2461 static int
2462 ff_layout_set_layoutdriver(struct nfs_server *server,
2463                 const struct nfs_fh *dummy)
2464 {
2465 #if IS_ENABLED(CONFIG_NFS_V4_2)
2466         server->caps |= NFS_CAP_LAYOUTSTATS;
2467 #endif
2468         return 0;
2469 }
2470
2471 static const struct pnfs_commit_ops ff_layout_commit_ops = {
2472         .setup_ds_info          = ff_layout_setup_ds_info,
2473         .release_ds_info        = ff_layout_release_ds_info,
2474         .mark_request_commit    = pnfs_layout_mark_request_commit,
2475         .clear_request_commit   = pnfs_generic_clear_request_commit,
2476         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2477         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2478         .commit_pagelist        = ff_layout_commit_pagelist,
2479 };
2480
2481 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2482         .id                     = LAYOUT_FLEX_FILES,
2483         .name                   = "LAYOUT_FLEX_FILES",
2484         .owner                  = THIS_MODULE,
2485         .flags                  = PNFS_LAYOUTGET_ON_OPEN,
2486         .max_layoutget_response = 4096, /* 1 page or so... */
2487         .set_layoutdriver       = ff_layout_set_layoutdriver,
2488         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2489         .free_layout_hdr        = ff_layout_free_layout_hdr,
2490         .alloc_lseg             = ff_layout_alloc_lseg,
2491         .free_lseg              = ff_layout_free_lseg,
2492         .add_lseg               = ff_layout_add_lseg,
2493         .pg_read_ops            = &ff_layout_pg_read_ops,
2494         .pg_write_ops           = &ff_layout_pg_write_ops,
2495         .get_ds_info            = ff_layout_get_ds_info,
2496         .free_deviceid_node     = ff_layout_free_deviceid_node,
2497         .read_pagelist          = ff_layout_read_pagelist,
2498         .write_pagelist         = ff_layout_write_pagelist,
2499         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2500         .prepare_layoutreturn   = ff_layout_prepare_layoutreturn,
2501         .sync                   = pnfs_nfs_generic_sync,
2502         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2503 };
2504
2505 static int __init nfs4flexfilelayout_init(void)
2506 {
2507         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2508                __func__);
2509         return pnfs_register_layoutdriver(&flexfilelayout_type);
2510 }
2511
2512 static void __exit nfs4flexfilelayout_exit(void)
2513 {
2514         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2515                __func__);
2516         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2517 }
2518
2519 MODULE_ALIAS("nfs-layouttype4-4");
2520
2521 MODULE_LICENSE("GPL");
2522 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2523
2524 module_init(nfs4flexfilelayout_init);
2525 module_exit(nfs4flexfilelayout_exit);
2526
2527 module_param(io_maxretrans, ushort, 0644);
2528 MODULE_PARM_DESC(io_maxretrans, "The  number of times the NFSv4.1 client "
2529                         "retries an I/O request before returning an error. ");