fs/nfs/flexfilelayout/flexfilelayout.c

   1 /*
   2  * Module for pnfs flexfile layout driver.
   3  *
   4  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   5  *
   6  * Tao Peng <bergwolf@primarydata.com>
   7  */
   8
   9 #include <linux/nfs_fs.h>
  10 #include <linux/nfs_page.h>
  11 #include <linux/module.h>
  12
  13 #include <linux/sunrpc/metrics.h>
  14
  15 #include "flexfilelayout.h"
  16 #include "../nfs4session.h"
  17 #include "../nfs4idmap.h"
  18 #include "../internal.h"
  19 #include "../delegation.h"
  20 #include "../nfs4trace.h"
  21 #include "../iostat.h"
  22 #include "../nfs.h"
  23 #include "../nfs42.h"
  24
  25 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  26
  27 #define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
  28
  29 static struct pnfs_layout_hdr *
  30 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
  31 {
  32         struct nfs4_flexfile_layout *ffl;
  33
  34         ffl = kzalloc(sizeof(*ffl), gfp_flags);
  35         if (ffl) {
  36                 INIT_LIST_HEAD(&ffl->error_list);
  37                 INIT_LIST_HEAD(&ffl->mirrors);
  38                 return &ffl->generic_hdr;
  39         } else
  40                 return NULL;
  41 }
  42
  43 static void
  44 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
  45 {
  46         struct nfs4_ff_layout_ds_err *err, *n;
  47
  48         list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
  49                                  list) {
  50                 list_del(&err->list);
  51                 kfree(err);
  52         }
  53         kfree(FF_LAYOUT_FROM_HDR(lo));
  54 }
  55
  56 static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
  57 {
  58         __be32 *p;
  59
  60         p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
  61         if (unlikely(p == NULL))
  62                 return -ENOBUFS;
  63         memcpy(stateid, p, NFS4_STATEID_SIZE);
  64         dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
  65                 p[0], p[1], p[2], p[3]);
  66         return 0;
  67 }
  68
  69 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
  70 {
  71         __be32 *p;
  72
  73         p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
  74         if (unlikely(!p))
  75                 return -ENOBUFS;
  76         memcpy(devid, p, NFS4_DEVICEID4_SIZE);
  77         nfs4_print_deviceid(devid);
  78         return 0;
  79 }
  80
  81 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
  82 {
  83         __be32 *p;
  84
  85         p = xdr_inline_decode(xdr, 4);
  86         if (unlikely(!p))
  87                 return -ENOBUFS;
  88         fh->size = be32_to_cpup(p++);
  89         if (fh->size > sizeof(struct nfs_fh)) {
  90                 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
  91                        fh->size);
  92                 return -EOVERFLOW;
  93         }
  94         /* fh.data */
  95         p = xdr_inline_decode(xdr, fh->size);
  96         if (unlikely(!p))
  97                 return -ENOBUFS;
  98         memcpy(&fh->data, p, fh->size);
  99         dprintk("%s: fh len %d\n", __func__, fh->size);
 100
 101         return 0;
 102 }
 103
 104 /*
 105  * Currently only stringified uids and gids are accepted.
 106  * I.e., kerberos is not supported to the DSes, so no pricipals.
 107  *
 108  * That means that one common function will suffice, but when
 109  * principals are added, this should be split to accomodate
 110  * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
 111  */
 112 static int
 113 decode_name(struct xdr_stream *xdr, u32 *id)
 114 {
 115         __be32 *p;
 116         int len;
 117
 118         /* opaque_length(4)*/
 119         p = xdr_inline_decode(xdr, 4);
 120         if (unlikely(!p))
 121                 return -ENOBUFS;
 122         len = be32_to_cpup(p++);
 123         if (len < 0)
 124                 return -EINVAL;
 125
 126         dprintk("%s: len %u\n", __func__, len);
 127
 128         /* opaque body */
 129         p = xdr_inline_decode(xdr, len);
 130         if (unlikely(!p))
 131                 return -ENOBUFS;
 132
 133         if (!nfs_map_string_to_numeric((char *)p, len, id))
 134                 return -EINVAL;
 135
 136         return 0;
 137 }
 138
 139 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1,
 140                 const struct nfs4_ff_layout_mirror *m2)
 141 {
 142         int i, j;
 143
 144         if (m1->fh_versions_cnt != m2->fh_versions_cnt)
 145                 return false;
 146         for (i = 0; i < m1->fh_versions_cnt; i++) {
 147                 bool found_fh = false;
 148                 for (j = 0; j < m2->fh_versions_cnt; i++) {
 149                         if (nfs_compare_fh(&m1->fh_versions[i],
 150                                         &m2->fh_versions[j]) == 0) {
 151                                 found_fh = true;
 152                                 break;
 153                         }
 154                 }
 155                 if (!found_fh)
 156                         return false;
 157         }
 158         return true;
 159 }
 160
 161 static struct nfs4_ff_layout_mirror *
 162 ff_layout_add_mirror(struct pnfs_layout_hdr *lo,
 163                 struct nfs4_ff_layout_mirror *mirror)
 164 {
 165         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
 166         struct nfs4_ff_layout_mirror *pos;
 167         struct inode *inode = lo->plh_inode;
 168
 169         spin_lock(&inode->i_lock);
 170         list_for_each_entry(pos, &ff_layout->mirrors, mirrors) {
 171                 if (mirror->mirror_ds != pos->mirror_ds)
 172                         continue;
 173                 if (!ff_mirror_match_fh(mirror, pos))
 174                         continue;
 175                 if (atomic_inc_not_zero(&pos->ref)) {
 176                         spin_unlock(&inode->i_lock);
 177                         return pos;
 178                 }
 179         }
 180         list_add(&mirror->mirrors, &ff_layout->mirrors);
 181         mirror->layout = lo;
 182         spin_unlock(&inode->i_lock);
 183         return mirror;
 184 }
 185
 186 static void
 187 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror)
 188 {
 189         struct inode *inode;
 190         if (mirror->layout == NULL)
 191                 return;
 192         inode = mirror->layout->plh_inode;
 193         spin_lock(&inode->i_lock);
 194         list_del(&mirror->mirrors);
 195         spin_unlock(&inode->i_lock);
 196         mirror->layout = NULL;
 197 }
 198
 199 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(gfp_t gfp_flags)
 200 {
 201         struct nfs4_ff_layout_mirror *mirror;
 202
 203         mirror = kzalloc(sizeof(*mirror), gfp_flags);
 204         if (mirror != NULL) {
 205                 spin_lock_init(&mirror->lock);
 206                 atomic_set(&mirror->ref, 1);
 207                 INIT_LIST_HEAD(&mirror->mirrors);
 208         }
 209         return mirror;
 210 }
 211
 212 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror)
 213 {
 214         ff_layout_remove_mirror(mirror);
 215         kfree(mirror->fh_versions);
 216         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 217         kfree(mirror);
 218 }
 219
 220 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 221 {
 222         if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
 223                 ff_layout_free_mirror(mirror);
 224 }
 225
 226 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 227 {
 228         int i;
 229
 230         if (fls->mirror_array) {
 231                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 232                         /* normally mirror_ds is freed in
 233                          * .free_deviceid_node but we still do it here
 234                          * for .alloc_lseg error path */
 235                         ff_layout_put_mirror(fls->mirror_array[i]);
 236                 }
 237                 kfree(fls->mirror_array);
 238                 fls->mirror_array = NULL;
 239         }
 240 }
 241
 242 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 243 {
 244         int ret = 0;
 245
 246         dprintk("--> %s\n", __func__);
 247
 248         /* FIXME: remove this check when layout segment support is added */
 249         if (lgr->range.offset != 0 ||
 250             lgr->range.length != NFS4_MAX_UINT64) {
 251                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 252                         __func__);
 253                 ret = -EINVAL;
 254         }
 255
 256         dprintk("--> %s returns %d\n", __func__, ret);
 257         return ret;
 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_range_is_after(const struct pnfs_layout_range *l1,
 270                 const struct pnfs_layout_range *l2)
 271 {
 272         u64 end1, end2;
 273
 274         if (l1->iomode != l2->iomode)
 275                 return l1->iomode != IOMODE_READ;
 276         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 277         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 278         if (end1 < l2->offset)
 279                 return false;
 280         if (end2 < l1->offset)
 281                 return true;
 282         return l2->offset <= l1->offset;
 283 }
 284
 285 static bool
 286 ff_lseg_merge(struct pnfs_layout_segment *new,
 287                 struct pnfs_layout_segment *old)
 288 {
 289         u64 new_end, old_end;
 290
 291         if (new->pls_range.iomode != old->pls_range.iomode)
 292                 return false;
 293         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 294                         old->pls_range.length);
 295         if (old_end < new->pls_range.offset)
 296                 return false;
 297         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 298                         new->pls_range.length);
 299         if (new_end < old->pls_range.offset)
 300                 return false;
 301
 302         /* Mergeable: copy info from 'old' to 'new' */
 303         if (new_end < old_end)
 304                 new_end = old_end;
 305         if (new->pls_range.offset < old->pls_range.offset)
 306                 new->pls_range.offset = old->pls_range.offset;
 307         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 308                         new_end);
 309         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 310                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 311         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 312                 set_bit(NFS_LSEG_LAYOUTRETURN, &new->pls_flags);
 313         return true;
 314 }
 315
 316 static void
 317 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 318                 struct pnfs_layout_segment *lseg,
 319                 struct list_head *free_me)
 320 {
 321         pnfs_generic_layout_insert_lseg(lo, lseg,
 322                         ff_lseg_range_is_after,
 323                         ff_lseg_merge,
 324                         free_me);
 325 }
 326
 327 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 328 {
 329         int i, j;
 330
 331         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 332                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 333                         if (fls->mirror_array[i]->efficiency <
 334                             fls->mirror_array[j]->efficiency)
 335                                 swap(fls->mirror_array[i],
 336                                      fls->mirror_array[j]);
 337         }
 338 }
 339
 340 static struct pnfs_layout_segment *
 341 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 342                      struct nfs4_layoutget_res *lgr,
 343                      gfp_t gfp_flags)
 344 {
 345         struct pnfs_layout_segment *ret;
 346         struct nfs4_ff_layout_segment *fls = NULL;
 347         struct xdr_stream stream;
 348         struct xdr_buf buf;
 349         struct page *scratch;
 350         u64 stripe_unit;
 351         u32 mirror_array_cnt;
 352         __be32 *p;
 353         int i, rc;
 354
 355         dprintk("--> %s\n", __func__);
 356         scratch = alloc_page(gfp_flags);
 357         if (!scratch)
 358                 return ERR_PTR(-ENOMEM);
 359
 360         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 361                               lgr->layoutp->len);
 362         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 363
 364         /* stripe unit and mirror_array_cnt */
 365         rc = -EIO;
 366         p = xdr_inline_decode(&stream, 8 + 4);
 367         if (!p)
 368                 goto out_err_free;
 369
 370         p = xdr_decode_hyper(p, &stripe_unit);
 371         mirror_array_cnt = be32_to_cpup(p++);
 372         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 373                 stripe_unit, mirror_array_cnt);
 374
 375         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 376             mirror_array_cnt == 0)
 377                 goto out_err_free;
 378
 379         rc = -ENOMEM;
 380         fls = kzalloc(sizeof(*fls), gfp_flags);
 381         if (!fls)
 382                 goto out_err_free;
 383
 384         fls->mirror_array_cnt = mirror_array_cnt;
 385         fls->stripe_unit = stripe_unit;
 386         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 387                                     sizeof(fls->mirror_array[0]), gfp_flags);
 388         if (fls->mirror_array == NULL)
 389                 goto out_err_free;
 390
 391         for (i = 0; i < fls->mirror_array_cnt; i++) {
 392                 struct nfs4_ff_layout_mirror *mirror;
 393                 struct nfs4_deviceid devid;
 394                 struct nfs4_deviceid_node *idnode;
 395                 u32 ds_count;
 396                 u32 fh_count;
 397                 int j;
 398
 399                 rc = -EIO;
 400                 p = xdr_inline_decode(&stream, 4);
 401                 if (!p)
 402                         goto out_err_free;
 403                 ds_count = be32_to_cpup(p);
 404
 405                 /* FIXME: allow for striping? */
 406                 if (ds_count != 1)
 407                         goto out_err_free;
 408
 409                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 410                 if (fls->mirror_array[i] == NULL) {
 411                         rc = -ENOMEM;
 412                         goto out_err_free;
 413                 }
 414
 415                 fls->mirror_array[i]->ds_count = ds_count;
 416
 417                 /* deviceid */
 418                 rc = decode_deviceid(&stream, &devid);
 419                 if (rc)
 420                         goto out_err_free;
 421
 422                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 423                                                 &devid, lh->plh_lc_cred,
 424                                                 gfp_flags);
 425                 /*
 426                  * upon success, mirror_ds is allocated by previous
 427                  * getdeviceinfo, or newly by .alloc_deviceid_node
 428                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 429                  */
 430                 if (idnode)
 431                         fls->mirror_array[i]->mirror_ds =
 432                                 FF_LAYOUT_MIRROR_DS(idnode);
 433                 else
 434                         goto out_err_free;
 435
 436                 /* efficiency */
 437                 rc = -EIO;
 438                 p = xdr_inline_decode(&stream, 4);
 439                 if (!p)
 440                         goto out_err_free;
 441                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 442
 443                 /* stateid */
 444                 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
 445                 if (rc)
 446                         goto out_err_free;
 447
 448                 /* fh */
 449                 p = xdr_inline_decode(&stream, 4);
 450                 if (!p)
 451                         goto out_err_free;
 452                 fh_count = be32_to_cpup(p);
 453
 454                 fls->mirror_array[i]->fh_versions =
 455                         kzalloc(fh_count * sizeof(struct nfs_fh),
 456                                 gfp_flags);
 457                 if (fls->mirror_array[i]->fh_versions == NULL) {
 458                         rc = -ENOMEM;
 459                         goto out_err_free;
 460                 }
 461
 462                 for (j = 0; j < fh_count; j++) {
 463                         rc = decode_nfs_fh(&stream,
 464                                            &fls->mirror_array[i]->fh_versions[j]);
 465                         if (rc)
 466                                 goto out_err_free;
 467                 }
 468
 469                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 470
 471                 /* user */
 472                 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
 473                 if (rc)
 474                         goto out_err_free;
 475
 476                 /* group */
 477                 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
 478                 if (rc)
 479                         goto out_err_free;
 480
 481                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 482                 if (mirror != fls->mirror_array[i]) {
 483                         ff_layout_free_mirror(fls->mirror_array[i]);
 484                         fls->mirror_array[i] = mirror;
 485                 }
 486
 487                 dprintk("%s: uid %d gid %d\n", __func__,
 488                         fls->mirror_array[i]->uid,
 489                         fls->mirror_array[i]->gid);
 490         }
 491
 492         p = xdr_inline_decode(&stream, 4);
 493         if (p)
 494                 fls->flags = be32_to_cpup(p);
 495
 496         ff_layout_sort_mirrors(fls);
 497         rc = ff_layout_check_layout(lgr);
 498         if (rc)
 499                 goto out_err_free;
 500
 501         ret = &fls->generic_hdr;
 502         dprintk("<-- %s (success)\n", __func__);
 503 out_free_page:
 504         __free_page(scratch);
 505         return ret;
 506 out_err_free:
 507         _ff_layout_free_lseg(fls);
 508         ret = ERR_PTR(rc);
 509         dprintk("<-- %s (%d)\n", __func__, rc);
 510         goto out_free_page;
 511 }
 512
 513 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 514 {
 515         struct pnfs_layout_segment *lseg;
 516
 517         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 518                 if (lseg->pls_range.iomode == IOMODE_RW)
 519                         return true;
 520
 521         return false;
 522 }
 523
 524 static void
 525 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 526 {
 527         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 528         int i;
 529
 530         dprintk("--> %s\n", __func__);
 531
 532         for (i = 0; i < fls->mirror_array_cnt; i++) {
 533                 if (fls->mirror_array[i]) {
 534                         nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
 535                         fls->mirror_array[i]->mirror_ds = NULL;
 536                         if (fls->mirror_array[i]->cred) {
 537                                 put_rpccred(fls->mirror_array[i]->cred);
 538                                 fls->mirror_array[i]->cred = NULL;
 539                         }
 540                 }
 541         }
 542
 543         if (lseg->pls_range.iomode == IOMODE_RW) {
 544                 struct nfs4_flexfile_layout *ffl;
 545                 struct inode *inode;
 546
 547                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 548                 inode = ffl->generic_hdr.plh_inode;
 549                 spin_lock(&inode->i_lock);
 550                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 551                         ffl->commit_info.nbuckets = 0;
 552                         kfree(ffl->commit_info.buckets);
 553                         ffl->commit_info.buckets = NULL;
 554                 }
 555                 spin_unlock(&inode->i_lock);
 556         }
 557         _ff_layout_free_lseg(fls);
 558 }
 559
 560 /* Return 1 until we have multiple lsegs support */
 561 static int
 562 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 563 {
 564         return 1;
 565 }
 566
 567 static void
 568 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 569 {
 570         /* first IO request? */
 571         if (atomic_inc_return(&timer->n_ops) == 1) {
 572                 timer->start_time = now;
 573         }
 574 }
 575
 576 static ktime_t
 577 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 578 {
 579         ktime_t start;
 580
 581         if (atomic_dec_return(&timer->n_ops) < 0)
 582                 WARN_ON_ONCE(1);
 583
 584         start = timer->start_time;
 585         timer->start_time = now;
 586         return ktime_sub(now, start);
 587 }
 588
 589 static bool
 590 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 591                             struct nfs4_ff_layoutstat *layoutstat,
 592                             ktime_t now)
 593 {
 594         static const ktime_t notime = {0};
 595         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 596
 597         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 598         if (ktime_equal(mirror->start_time, notime))
 599                 mirror->start_time = now;
 600         if (ktime_equal(mirror->last_report_time, notime))
 601                 mirror->last_report_time = now;
 602         if (layoutstats_timer != 0)
 603                 report_interval = (s64)layoutstats_timer * 1000LL;
 604         if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
 605                         report_interval) {
 606                 mirror->last_report_time = now;
 607                 return true;
 608         }
 609
 610         return false;
 611 }
 612
 613 static void
 614 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 615                 __u64 requested)
 616 {
 617         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 618
 619         iostat->ops_requested++;
 620         iostat->bytes_requested += requested;
 621 }
 622
 623 static void
 624 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 625                 __u64 requested,
 626                 __u64 completed,
 627                 ktime_t time_completed,
 628                 ktime_t time_started)
 629 {
 630         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 631         ktime_t completion_time = ktime_sub(time_completed, time_started);
 632         ktime_t timer;
 633
 634         iostat->ops_completed++;
 635         iostat->bytes_completed += completed;
 636         iostat->bytes_not_delivered += requested - completed;
 637
 638         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 639         iostat->total_busy_time =
 640                         ktime_add(iostat->total_busy_time, timer);
 641         iostat->aggregate_completion_time =
 642                         ktime_add(iostat->aggregate_completion_time,
 643                                         completion_time);
 644 }
 645
 646 static void
 647 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 648                 struct nfs4_ff_layout_mirror *mirror,
 649                 __u64 requested, ktime_t now)
 650 {
 651         bool report;
 652
 653         spin_lock(&mirror->lock);
 654         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 655         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 656         spin_unlock(&mirror->lock);
 657
 658         if (report)
 659                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 660 }
 661
 662 static void
 663 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 664                 struct nfs4_ff_layout_mirror *mirror,
 665                 __u64 requested,
 666                 __u64 completed)
 667 {
 668         spin_lock(&mirror->lock);
 669         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 670                         requested, completed,
 671                         ktime_get(), task->tk_start);
 672         spin_unlock(&mirror->lock);
 673 }
 674
 675 static void
 676 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 677                 struct nfs4_ff_layout_mirror *mirror,
 678                 __u64 requested, ktime_t now)
 679 {
 680         bool report;
 681
 682         spin_lock(&mirror->lock);
 683         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 684         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 685         spin_unlock(&mirror->lock);
 686
 687         if (report)
 688                 pnfs_report_layoutstat(inode, GFP_NOIO);
 689 }
 690
 691 static void
 692 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 693                 struct nfs4_ff_layout_mirror *mirror,
 694                 __u64 requested,
 695                 __u64 completed,
 696                 enum nfs3_stable_how committed)
 697 {
 698         if (committed == NFS_UNSTABLE)
 699                 requested = completed = 0;
 700
 701         spin_lock(&mirror->lock);
 702         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 703                         requested, completed, ktime_get(), task->tk_start);
 704         spin_unlock(&mirror->lock);
 705 }
 706
 707 static int
 708 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 709                             struct nfs_commit_info *cinfo,
 710                             gfp_t gfp_flags)
 711 {
 712         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 713         struct pnfs_commit_bucket *buckets;
 714         int size;
 715
 716         if (cinfo->ds->nbuckets != 0) {
 717                 /* This assumes there is only one RW lseg per file.
 718                  * To support multiple lseg per file, we need to
 719                  * change struct pnfs_commit_bucket to allow dynamic
 720                  * increasing nbuckets.
 721                  */
 722                 return 0;
 723         }
 724
 725         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 726
 727         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 728                           gfp_flags);
 729         if (!buckets)
 730                 return -ENOMEM;
 731         else {
 732                 int i;
 733
 734                 spin_lock(cinfo->lock);
 735                 if (cinfo->ds->nbuckets != 0)
 736                         kfree(buckets);
 737                 else {
 738                         cinfo->ds->buckets = buckets;
 739                         cinfo->ds->nbuckets = size;
 740                         for (i = 0; i < size; i++) {
 741                                 INIT_LIST_HEAD(&buckets[i].written);
 742                                 INIT_LIST_HEAD(&buckets[i].committing);
 743                                 /* mark direct verifier as unset */
 744                                 buckets[i].direct_verf.committed =
 745                                         NFS_INVALID_STABLE_HOW;
 746                         }
 747                 }
 748                 spin_unlock(cinfo->lock);
 749                 return 0;
 750         }
 751 }
 752
 753 static struct nfs4_pnfs_ds *
 754 ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
 755                                   int *best_idx)
 756 {
 757         struct nfs4_ff_layout_segment *fls;
 758         struct nfs4_pnfs_ds *ds;
 759         int idx;
 760
 761         fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
 762         /* mirrors are sorted by efficiency */
 763         for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
 764                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
 765                 if (ds) {
 766                         *best_idx = idx;
 767                         return ds;
 768                 }
 769         }
 770
 771         return NULL;
 772 }
 773
 774 static void
 775 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 776                         struct nfs_page *req)
 777 {
 778         struct nfs_pgio_mirror *pgm;
 779         struct nfs4_ff_layout_mirror *mirror;
 780         struct nfs4_pnfs_ds *ds;
 781         int ds_idx;
 782
 783         /* Use full layout for now */
 784         if (!pgio->pg_lseg)
 785                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 786                                                    req->wb_context,
 787                                                    0,
 788                                                    NFS4_MAX_UINT64,
 789                                                    IOMODE_READ,
 790                                                    GFP_KERNEL);
 791         /* If no lseg, fall back to read through mds */
 792         if (pgio->pg_lseg == NULL)
 793                 goto out_mds;
 794
 795         ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
 796         if (!ds)
 797                 goto out_mds;
 798         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 799
 800         pgio->pg_mirror_idx = ds_idx;
 801
 802         /* read always uses only one mirror - idx 0 for pgio layer */
 803         pgm = &pgio->pg_mirrors[0];
 804         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 805
 806         return;
 807 out_mds:
 808         pnfs_put_lseg(pgio->pg_lseg);
 809         pgio->pg_lseg = NULL;
 810         nfs_pageio_reset_read_mds(pgio);
 811 }
 812
 813 static void
 814 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 815                         struct nfs_page *req)
 816 {
 817         struct nfs4_ff_layout_mirror *mirror;
 818         struct nfs_pgio_mirror *pgm;
 819         struct nfs_commit_info cinfo;
 820         struct nfs4_pnfs_ds *ds;
 821         int i;
 822         int status;
 823
 824         if (!pgio->pg_lseg)
 825                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 826                                                    req->wb_context,
 827                                                    0,
 828                                                    NFS4_MAX_UINT64,
 829                                                    IOMODE_RW,
 830                                                    GFP_NOFS);
 831         /* If no lseg, fall back to write through mds */
 832         if (pgio->pg_lseg == NULL)
 833                 goto out_mds;
 834
 835         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 836         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 837         if (status < 0)
 838                 goto out_mds;
 839
 840         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 841         if (WARN_ON_ONCE(pgio->pg_mirror_count !=
 842             FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
 843                 goto out_mds;
 844
 845         for (i = 0; i < pgio->pg_mirror_count; i++) {
 846                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 847                 if (!ds)
 848                         goto out_mds;
 849                 pgm = &pgio->pg_mirrors[i];
 850                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 851                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 852         }
 853
 854         return;
 855
 856 out_mds:
 857         pnfs_put_lseg(pgio->pg_lseg);
 858         pgio->pg_lseg = NULL;
 859         nfs_pageio_reset_write_mds(pgio);
 860 }
 861
 862 static unsigned int
 863 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 864                                     struct nfs_page *req)
 865 {
 866         if (!pgio->pg_lseg)
 867                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 868                                                    req->wb_context,
 869                                                    0,
 870                                                    NFS4_MAX_UINT64,
 871                                                    IOMODE_RW,
 872                                                    GFP_NOFS);
 873         if (pgio->pg_lseg)
 874                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 875
 876         /* no lseg means that pnfs is not in use, so no mirroring here */
 877         nfs_pageio_reset_write_mds(pgio);
 878         return 1;
 879 }
 880
 881 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 882         .pg_init = ff_layout_pg_init_read,
 883         .pg_test = pnfs_generic_pg_test,
 884         .pg_doio = pnfs_generic_pg_readpages,
 885         .pg_cleanup = pnfs_generic_pg_cleanup,
 886 };
 887
 888 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 889         .pg_init = ff_layout_pg_init_write,
 890         .pg_test = pnfs_generic_pg_test,
 891         .pg_doio = pnfs_generic_pg_writepages,
 892         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 893         .pg_cleanup = pnfs_generic_pg_cleanup,
 894 };
 895
 896 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 897 {
 898         struct rpc_task *task = &hdr->task;
 899
 900         pnfs_layoutcommit_inode(hdr->inode, false);
 901
 902         if (retry_pnfs) {
 903                 dprintk("%s Reset task %5u for i/o through pNFS "
 904                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 905                         hdr->task.tk_pid,
 906                         hdr->inode->i_sb->s_id,
 907                         (unsigned long long)NFS_FILEID(hdr->inode),
 908                         hdr->args.count,
 909                         (unsigned long long)hdr->args.offset);
 910
 911                 if (!hdr->dreq) {
 912                         struct nfs_open_context *ctx;
 913
 914                         ctx = nfs_list_entry(hdr->pages.next)->wb_context;
 915                         set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
 916                         hdr->completion_ops->error_cleanup(&hdr->pages);
 917                 } else {
 918                         nfs_direct_set_resched_writes(hdr->dreq);
 919                         /* fake unstable write to let common nfs resend pages */
 920                         hdr->verf.committed = NFS_UNSTABLE;
 921                         hdr->good_bytes = hdr->args.count;
 922                 }
 923                 return;
 924         }
 925
 926         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 927                 dprintk("%s Reset task %5u for i/o through MDS "
 928                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 929                         hdr->task.tk_pid,
 930                         hdr->inode->i_sb->s_id,
 931                         (unsigned long long)NFS_FILEID(hdr->inode),
 932                         hdr->args.count,
 933                         (unsigned long long)hdr->args.offset);
 934
 935                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
 936         }
 937 }
 938
 939 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
 940 {
 941         struct rpc_task *task = &hdr->task;
 942
 943         pnfs_layoutcommit_inode(hdr->inode, false);
 944
 945         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 946                 dprintk("%s Reset task %5u for i/o through MDS "
 947                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 948                         hdr->task.tk_pid,
 949                         hdr->inode->i_sb->s_id,
 950                         (unsigned long long)NFS_FILEID(hdr->inode),
 951                         hdr->args.count,
 952                         (unsigned long long)hdr->args.offset);
 953
 954                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
 955         }
 956 }
 957
 958 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
 959                                            struct nfs4_state *state,
 960                                            struct nfs_client *clp,
 961                                            struct pnfs_layout_segment *lseg,
 962                                            int idx)
 963 {
 964         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 965         struct inode *inode = lo->plh_inode;
 966         struct nfs_server *mds_server = NFS_SERVER(inode);
 967
 968         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 969         struct nfs_client *mds_client = mds_server->nfs_client;
 970         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 971
 972         if (task->tk_status >= 0)
 973                 return 0;
 974
 975         switch (task->tk_status) {
 976         /* MDS state errors */
 977         case -NFS4ERR_DELEG_REVOKED:
 978         case -NFS4ERR_ADMIN_REVOKED:
 979         case -NFS4ERR_BAD_STATEID:
 980                 if (state == NULL)
 981                         break;
 982                 nfs_remove_bad_delegation(state->inode);
 983         case -NFS4ERR_OPENMODE:
 984                 if (state == NULL)
 985                         break;
 986                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 987                         goto out_bad_stateid;
 988                 goto wait_on_recovery;
 989         case -NFS4ERR_EXPIRED:
 990                 if (state != NULL) {
 991                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
 992                                 goto out_bad_stateid;
 993                 }
 994                 nfs4_schedule_lease_recovery(mds_client);
 995                 goto wait_on_recovery;
 996         /* DS session errors */
 997         case -NFS4ERR_BADSESSION:
 998         case -NFS4ERR_BADSLOT:
 999         case -NFS4ERR_BAD_HIGH_SLOT:
1000         case -NFS4ERR_DEADSESSION:
1001         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1002         case -NFS4ERR_SEQ_FALSE_RETRY:
1003         case -NFS4ERR_SEQ_MISORDERED:
1004                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1005                         "flags 0x%x\n", __func__, task->tk_status,
1006                         clp->cl_exchange_flags);
1007                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1008                 break;
1009         case -NFS4ERR_DELAY:
1010         case -NFS4ERR_GRACE:
1011                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1012                 break;
1013         case -NFS4ERR_RETRY_UNCACHED_REP:
1014                 break;
1015         /* Invalidate Layout errors */
1016         case -NFS4ERR_PNFS_NO_LAYOUT:
1017         case -ESTALE:           /* mapped NFS4ERR_STALE */
1018         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1019         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1020         case -NFS4ERR_FHEXPIRED:
1021         case -NFS4ERR_WRONG_TYPE:
1022                 dprintk("%s Invalid layout error %d\n", __func__,
1023                         task->tk_status);
1024                 /*
1025                  * Destroy layout so new i/o will get a new layout.
1026                  * Layout will not be destroyed until all current lseg
1027                  * references are put. Mark layout as invalid to resend failed
1028                  * i/o and all i/o waiting on the slot table to the MDS until
1029                  * layout is destroyed and a new valid layout is obtained.
1030                  */
1031                 pnfs_destroy_layout(NFS_I(inode));
1032                 rpc_wake_up(&tbl->slot_tbl_waitq);
1033                 goto reset;
1034         /* RPC connection errors */
1035         case -ECONNREFUSED:
1036         case -EHOSTDOWN:
1037         case -EHOSTUNREACH:
1038         case -ENETUNREACH:
1039         case -EIO:
1040         case -ETIMEDOUT:
1041         case -EPIPE:
1042                 dprintk("%s DS connection error %d\n", __func__,
1043                         task->tk_status);
1044                 nfs4_mark_deviceid_unavailable(devid);
1045                 rpc_wake_up(&tbl->slot_tbl_waitq);
1046                 /* fall through */
1047         default:
1048                 if (ff_layout_has_available_ds(lseg))
1049                         return -NFS4ERR_RESET_TO_PNFS;
1050 reset:
1051                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1052                         task->tk_status);
1053                 return -NFS4ERR_RESET_TO_MDS;
1054         }
1055 out:
1056         task->tk_status = 0;
1057         return -EAGAIN;
1058 out_bad_stateid:
1059         task->tk_status = -EIO;
1060         return 0;
1061 wait_on_recovery:
1062         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
1063         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
1064                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
1065         goto out;
1066 }
1067
1068 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1069 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1070                                            struct pnfs_layout_segment *lseg,
1071                                            int idx)
1072 {
1073         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1074
1075         if (task->tk_status >= 0)
1076                 return 0;
1077
1078         switch (task->tk_status) {
1079         /* File access problems. Don't mark the device as unavailable */
1080         case -EACCES:
1081         case -ESTALE:
1082         case -EISDIR:
1083         case -EBADHANDLE:
1084         case -ELOOP:
1085         case -ENOSPC:
1086                 break;
1087         case -EJUKEBOX:
1088                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1089                 goto out_retry;
1090         default:
1091                 dprintk("%s DS connection error %d\n", __func__,
1092                         task->tk_status);
1093                 nfs4_mark_deviceid_unavailable(devid);
1094         }
1095         /* FIXME: Need to prevent infinite looping here. */
1096         return -NFS4ERR_RESET_TO_PNFS;
1097 out_retry:
1098         task->tk_status = 0;
1099         rpc_restart_call(task);
1100         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1101         return -EAGAIN;
1102 }
1103
1104 static int ff_layout_async_handle_error(struct rpc_task *task,
1105                                         struct nfs4_state *state,
1106                                         struct nfs_client *clp,
1107                                         struct pnfs_layout_segment *lseg,
1108                                         int idx)
1109 {
1110         int vers = clp->cl_nfs_mod->rpc_vers->number;
1111
1112         switch (vers) {
1113         case 3:
1114                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1115         case 4:
1116                 return ff_layout_async_handle_error_v4(task, state, clp,
1117                                                        lseg, idx);
1118         default:
1119                 /* should never happen */
1120                 WARN_ON_ONCE(1);
1121                 return 0;
1122         }
1123 }
1124
1125 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1126                                         int idx, u64 offset, u64 length,
1127                                         u32 status, int opnum, int error)
1128 {
1129         struct nfs4_ff_layout_mirror *mirror;
1130         int err;
1131
1132         if (status == 0) {
1133                 switch (error) {
1134                 case -ETIMEDOUT:
1135                 case -EPFNOSUPPORT:
1136                 case -EPROTONOSUPPORT:
1137                 case -EOPNOTSUPP:
1138                 case -ECONNREFUSED:
1139                 case -ECONNRESET:
1140                 case -EHOSTDOWN:
1141                 case -EHOSTUNREACH:
1142                 case -ENETUNREACH:
1143                 case -EADDRINUSE:
1144                 case -ENOBUFS:
1145                 case -EPIPE:
1146                 case -EPERM:
1147                         status = NFS4ERR_NXIO;
1148                         break;
1149                 case -EACCES:
1150                         status = NFS4ERR_ACCESS;
1151                         break;
1152                 default:
1153                         return;
1154                 }
1155         }
1156
1157         mirror = FF_LAYOUT_COMP(lseg, idx);
1158         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1159                                        mirror, offset, length, status, opnum,
1160                                        GFP_NOIO);
1161         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1162 }
1163
1164 /* NFS_PROTO call done callback routines */
1165
1166 static int ff_layout_read_done_cb(struct rpc_task *task,
1167                                 struct nfs_pgio_header *hdr)
1168 {
1169         struct inode *inode;
1170         int err;
1171
1172         trace_nfs4_pnfs_read(hdr, task->tk_status);
1173         if (task->tk_status < 0)
1174                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1175                                             hdr->args.offset, hdr->args.count,
1176                                             hdr->res.op_status, OP_READ,
1177                                             task->tk_status);
1178         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1179                                            hdr->ds_clp, hdr->lseg,
1180                                            hdr->pgio_mirror_idx);
1181
1182         switch (err) {
1183         case -NFS4ERR_RESET_TO_PNFS:
1184                 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1185                         &hdr->lseg->pls_layout->plh_flags);
1186                 pnfs_read_resend_pnfs(hdr);
1187                 return task->tk_status;
1188         case -NFS4ERR_RESET_TO_MDS:
1189                 inode = hdr->lseg->pls_layout->plh_inode;
1190                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1191                 ff_layout_reset_read(hdr);
1192                 return task->tk_status;
1193         case -EAGAIN:
1194                 rpc_restart_call_prepare(task);
1195                 return -EAGAIN;
1196         }
1197
1198         return 0;
1199 }
1200
1201 static bool
1202 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1203 {
1204         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1205 }
1206
1207 /*
1208  * We reference the rpc_cred of the first WRITE that triggers the need for
1209  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1210  * rfc5661 is not clear about which credential should be used.
1211  *
1212  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1213  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1214  * we always send layoutcommit after DS writes.
1215  */
1216 static void
1217 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
1218 {
1219         if (!ff_layout_need_layoutcommit(hdr->lseg))
1220                 return;
1221
1222         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1223                         hdr->mds_offset + hdr->res.count);
1224         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
1225                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
1226 }
1227
1228 static bool
1229 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
1230 {
1231         /* No mirroring for now */
1232         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1233
1234         return ff_layout_test_devid_unavailable(node);
1235 }
1236
1237 static int ff_layout_read_prepare_common(struct rpc_task *task,
1238                                          struct nfs_pgio_header *hdr)
1239 {
1240         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1241                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1242                         hdr->args.count,
1243                         task->tk_start);
1244
1245         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1246                 rpc_exit(task, -EIO);
1247                 return -EIO;
1248         }
1249         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1250                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
1251                 if (ff_layout_has_available_ds(hdr->lseg))
1252                         pnfs_read_resend_pnfs(hdr);
1253                 else
1254                         ff_layout_reset_read(hdr);
1255                 rpc_exit(task, 0);
1256                 return -EAGAIN;
1257         }
1258         hdr->pgio_done_cb = ff_layout_read_done_cb;
1259
1260         return 0;
1261 }
1262
1263 /*
1264  * Call ops for the async read/write cases
1265  * In the case of dense layouts, the offset needs to be reset to its
1266  * original value.
1267  */
1268 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1269 {
1270         struct nfs_pgio_header *hdr = data;
1271
1272         if (ff_layout_read_prepare_common(task, hdr))
1273                 return;
1274
1275         rpc_call_start(task);
1276 }
1277
1278 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1279                                     struct nfs4_sequence_args *args,
1280                                     struct nfs4_sequence_res *res,
1281                                     struct rpc_task *task)
1282 {
1283         if (ds_clp->cl_session)
1284                 return nfs41_setup_sequence(ds_clp->cl_session,
1285                                            args,
1286                                            res,
1287                                            task);
1288         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1289                                    args,
1290                                    res,
1291                                    task);
1292 }
1293
1294 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1295 {
1296         struct nfs_pgio_header *hdr = data;
1297
1298         if (ff_layout_setup_sequence(hdr->ds_clp,
1299                                      &hdr->args.seq_args,
1300                                      &hdr->res.seq_res,
1301                                      task))
1302                 return;
1303
1304         if (ff_layout_read_prepare_common(task, hdr))
1305                 return;
1306
1307         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1308                         hdr->args.lock_context, FMODE_READ) == -EIO)
1309                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1310 }
1311
1312 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1313 {
1314         struct nfs_pgio_header *hdr = data;
1315
1316         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1317
1318         nfs4_ff_layout_stat_io_end_read(task,
1319                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1320                         hdr->args.count, hdr->res.count);
1321
1322         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1323             task->tk_status == 0) {
1324                 nfs4_sequence_done(task, &hdr->res.seq_res);
1325                 return;
1326         }
1327
1328         /* Note this may cause RPC to be resent */
1329         hdr->mds_ops->rpc_call_done(task, hdr);
1330 }
1331
1332 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1333 {
1334         struct nfs_pgio_header *hdr = data;
1335
1336         rpc_count_iostats_metrics(task,
1337             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1338 }
1339
1340 static int ff_layout_write_done_cb(struct rpc_task *task,
1341                                 struct nfs_pgio_header *hdr)
1342 {
1343         struct inode *inode;
1344         int err;
1345
1346         trace_nfs4_pnfs_write(hdr, task->tk_status);
1347         if (task->tk_status < 0)
1348                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1349                                             hdr->args.offset, hdr->args.count,
1350                                             hdr->res.op_status, OP_WRITE,
1351                                             task->tk_status);
1352         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1353                                            hdr->ds_clp, hdr->lseg,
1354                                            hdr->pgio_mirror_idx);
1355
1356         switch (err) {
1357         case -NFS4ERR_RESET_TO_PNFS:
1358         case -NFS4ERR_RESET_TO_MDS:
1359                 inode = hdr->lseg->pls_layout->plh_inode;
1360                 pnfs_error_mark_layout_for_return(inode, hdr->lseg);
1361                 if (err == -NFS4ERR_RESET_TO_PNFS) {
1362                         pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1363                         ff_layout_reset_write(hdr, true);
1364                 } else {
1365                         pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1366                         ff_layout_reset_write(hdr, false);
1367                 }
1368                 return task->tk_status;
1369         case -EAGAIN:
1370                 rpc_restart_call_prepare(task);
1371                 return -EAGAIN;
1372         }
1373
1374         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1375             hdr->res.verf->committed == NFS_DATA_SYNC)
1376                 ff_layout_set_layoutcommit(hdr);
1377
1378         /* zero out fattr since we don't care DS attr at all */
1379         hdr->fattr.valid = 0;
1380         if (task->tk_status >= 0)
1381                 nfs_writeback_update_inode(hdr);
1382
1383         return 0;
1384 }
1385
1386 static int ff_layout_commit_done_cb(struct rpc_task *task,
1387                                      struct nfs_commit_data *data)
1388 {
1389         struct inode *inode;
1390         int err;
1391
1392         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1393         if (task->tk_status < 0)
1394                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1395                                             data->args.offset, data->args.count,
1396                                             data->res.op_status, OP_COMMIT,
1397                                             task->tk_status);
1398         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1399                                            data->lseg, data->ds_commit_index);
1400
1401         switch (err) {
1402         case -NFS4ERR_RESET_TO_PNFS:
1403         case -NFS4ERR_RESET_TO_MDS:
1404                 inode = data->lseg->pls_layout->plh_inode;
1405                 pnfs_error_mark_layout_for_return(inode, data->lseg);
1406                 if (err == -NFS4ERR_RESET_TO_PNFS)
1407                         pnfs_set_retry_layoutget(data->lseg->pls_layout);
1408                 else
1409                         pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1410                 pnfs_generic_prepare_to_resend_writes(data);
1411                 return -EAGAIN;
1412         case -EAGAIN:
1413                 rpc_restart_call_prepare(task);
1414                 return -EAGAIN;
1415         }
1416
1417         if (data->verf.committed == NFS_UNSTABLE
1418             && ff_layout_need_layoutcommit(data->lseg))
1419                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1420
1421         return 0;
1422 }
1423
1424 static int ff_layout_write_prepare_common(struct rpc_task *task,
1425                                           struct nfs_pgio_header *hdr)
1426 {
1427         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1428                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1429                         hdr->args.count,
1430                         task->tk_start);
1431
1432         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1433                 rpc_exit(task, -EIO);
1434                 return -EIO;
1435         }
1436
1437         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1438                 bool retry_pnfs;
1439
1440                 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1441                 dprintk("%s task %u reset io to %s\n", __func__,
1442                         task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1443                 ff_layout_reset_write(hdr, retry_pnfs);
1444                 rpc_exit(task, 0);
1445                 return -EAGAIN;
1446         }
1447
1448         return 0;
1449 }
1450
1451 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1452 {
1453         struct nfs_pgio_header *hdr = data;
1454
1455         if (ff_layout_write_prepare_common(task, hdr))
1456                 return;
1457
1458         rpc_call_start(task);
1459 }
1460
1461 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1462 {
1463         struct nfs_pgio_header *hdr = data;
1464
1465         if (ff_layout_setup_sequence(hdr->ds_clp,
1466                                      &hdr->args.seq_args,
1467                                      &hdr->res.seq_res,
1468                                      task))
1469                 return;
1470
1471         if (ff_layout_write_prepare_common(task, hdr))
1472                 return;
1473
1474         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1475                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1476                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1477 }
1478
1479 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1480 {
1481         struct nfs_pgio_header *hdr = data;
1482
1483         nfs4_ff_layout_stat_io_end_write(task,
1484                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1485                         hdr->args.count, hdr->res.count,
1486                         hdr->res.verf->committed);
1487
1488         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1489             task->tk_status == 0) {
1490                 nfs4_sequence_done(task, &hdr->res.seq_res);
1491                 return;
1492         }
1493
1494         /* Note this may cause RPC to be resent */
1495         hdr->mds_ops->rpc_call_done(task, hdr);
1496 }
1497
1498 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1499 {
1500         struct nfs_pgio_header *hdr = data;
1501
1502         rpc_count_iostats_metrics(task,
1503             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1504 }
1505
1506 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1507                 struct nfs_commit_data *cdata)
1508 {
1509         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1510                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1511                         0, task->tk_start);
1512 }
1513
1514 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1515 {
1516         ff_layout_commit_prepare_common(task, data);
1517         rpc_call_start(task);
1518 }
1519
1520 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1521 {
1522         struct nfs_commit_data *wdata = data;
1523
1524         if (ff_layout_setup_sequence(wdata->ds_clp,
1525                                  &wdata->args.seq_args,
1526                                  &wdata->res.seq_res,
1527                                  task))
1528                 return;
1529         ff_layout_commit_prepare_common(task, data);
1530 }
1531
1532 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1533 {
1534         struct nfs_commit_data *cdata = data;
1535         struct nfs_page *req;
1536         __u64 count = 0;
1537
1538         if (task->tk_status == 0) {
1539                 list_for_each_entry(req, &cdata->pages, wb_list)
1540                         count += req->wb_bytes;
1541         }
1542
1543         nfs4_ff_layout_stat_io_end_write(task,
1544                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1545                         count, count, NFS_FILE_SYNC);
1546
1547         pnfs_generic_write_commit_done(task, data);
1548 }
1549
1550 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1551 {
1552         struct nfs_commit_data *cdata = data;
1553
1554         rpc_count_iostats_metrics(task,
1555             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1556 }
1557
1558 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1559         .rpc_call_prepare = ff_layout_read_prepare_v3,
1560         .rpc_call_done = ff_layout_read_call_done,
1561         .rpc_count_stats = ff_layout_read_count_stats,
1562         .rpc_release = pnfs_generic_rw_release,
1563 };
1564
1565 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1566         .rpc_call_prepare = ff_layout_read_prepare_v4,
1567         .rpc_call_done = ff_layout_read_call_done,
1568         .rpc_count_stats = ff_layout_read_count_stats,
1569         .rpc_release = pnfs_generic_rw_release,
1570 };
1571
1572 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1573         .rpc_call_prepare = ff_layout_write_prepare_v3,
1574         .rpc_call_done = ff_layout_write_call_done,
1575         .rpc_count_stats = ff_layout_write_count_stats,
1576         .rpc_release = pnfs_generic_rw_release,
1577 };
1578
1579 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1580         .rpc_call_prepare = ff_layout_write_prepare_v4,
1581         .rpc_call_done = ff_layout_write_call_done,
1582         .rpc_count_stats = ff_layout_write_count_stats,
1583         .rpc_release = pnfs_generic_rw_release,
1584 };
1585
1586 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1587         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1588         .rpc_call_done = ff_layout_commit_done,
1589         .rpc_count_stats = ff_layout_commit_count_stats,
1590         .rpc_release = pnfs_generic_commit_release,
1591 };
1592
1593 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1594         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1595         .rpc_call_done = ff_layout_commit_done,
1596         .rpc_count_stats = ff_layout_commit_count_stats,
1597         .rpc_release = pnfs_generic_commit_release,
1598 };
1599
1600 static enum pnfs_try_status
1601 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1602 {
1603         struct pnfs_layout_segment *lseg = hdr->lseg;
1604         struct nfs4_pnfs_ds *ds;
1605         struct rpc_clnt *ds_clnt;
1606         struct rpc_cred *ds_cred;
1607         loff_t offset = hdr->args.offset;
1608         u32 idx = hdr->pgio_mirror_idx;
1609         int vers;
1610         struct nfs_fh *fh;
1611
1612         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1613                 __func__, hdr->inode->i_ino,
1614                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1615
1616         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1617         if (!ds)
1618                 goto out_failed;
1619
1620         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1621                                                    hdr->inode);
1622         if (IS_ERR(ds_clnt))
1623                 goto out_failed;
1624
1625         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1626         if (IS_ERR(ds_cred))
1627                 goto out_failed;
1628
1629         vers = nfs4_ff_layout_ds_version(lseg, idx);
1630
1631         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1632                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1633
1634         atomic_inc(&ds->ds_clp->cl_count);
1635         hdr->ds_clp = ds->ds_clp;
1636         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1637         if (fh)
1638                 hdr->args.fh = fh;
1639         /*
1640          * Note that if we ever decide to split across DSes,
1641          * then we may need to handle dense-like offsets.
1642          */
1643         hdr->args.offset = offset;
1644         hdr->mds_offset = offset;
1645
1646         /* Perform an asynchronous read to ds */
1647         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1648                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1649                                       &ff_layout_read_call_ops_v4,
1650                           0, RPC_TASK_SOFTCONN);
1651
1652         return PNFS_ATTEMPTED;
1653
1654 out_failed:
1655         if (ff_layout_has_available_ds(lseg))
1656                 return PNFS_TRY_AGAIN;
1657         return PNFS_NOT_ATTEMPTED;
1658 }
1659
1660 /* Perform async writes. */
1661 static enum pnfs_try_status
1662 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1663 {
1664         struct pnfs_layout_segment *lseg = hdr->lseg;
1665         struct nfs4_pnfs_ds *ds;
1666         struct rpc_clnt *ds_clnt;
1667         struct rpc_cred *ds_cred;
1668         loff_t offset = hdr->args.offset;
1669         int vers;
1670         struct nfs_fh *fh;
1671         int idx = hdr->pgio_mirror_idx;
1672
1673         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1674         if (!ds)
1675                 return PNFS_NOT_ATTEMPTED;
1676
1677         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1678                                                    hdr->inode);
1679         if (IS_ERR(ds_clnt))
1680                 return PNFS_NOT_ATTEMPTED;
1681
1682         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1683         if (IS_ERR(ds_cred))
1684                 return PNFS_NOT_ATTEMPTED;
1685
1686         vers = nfs4_ff_layout_ds_version(lseg, idx);
1687
1688         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1689                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1690                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1691                 vers);
1692
1693         hdr->pgio_done_cb = ff_layout_write_done_cb;
1694         atomic_inc(&ds->ds_clp->cl_count);
1695         hdr->ds_clp = ds->ds_clp;
1696         hdr->ds_commit_idx = idx;
1697         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1698         if (fh)
1699                 hdr->args.fh = fh;
1700
1701         /*
1702          * Note that if we ever decide to split across DSes,
1703          * then we may need to handle dense-like offsets.
1704          */
1705         hdr->args.offset = offset;
1706
1707         /* Perform an asynchronous write */
1708         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1709                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1710                                       &ff_layout_write_call_ops_v4,
1711                           sync, RPC_TASK_SOFTCONN);
1712         return PNFS_ATTEMPTED;
1713 }
1714
1715 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1716 {
1717         return i;
1718 }
1719
1720 static struct nfs_fh *
1721 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1722 {
1723         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1724
1725         /* FIXME: Assume that there is only one NFS version available
1726          * for the DS.
1727          */
1728         return &flseg->mirror_array[i]->fh_versions[0];
1729 }
1730
1731 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1732 {
1733         struct pnfs_layout_segment *lseg = data->lseg;
1734         struct nfs4_pnfs_ds *ds;
1735         struct rpc_clnt *ds_clnt;
1736         struct rpc_cred *ds_cred;
1737         u32 idx;
1738         int vers;
1739         struct nfs_fh *fh;
1740
1741         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1742         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1743         if (!ds)
1744                 goto out_err;
1745
1746         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1747                                                    data->inode);
1748         if (IS_ERR(ds_clnt))
1749                 goto out_err;
1750
1751         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1752         if (IS_ERR(ds_cred))
1753                 goto out_err;
1754
1755         vers = nfs4_ff_layout_ds_version(lseg, idx);
1756
1757         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1758                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1759                 vers);
1760         data->commit_done_cb = ff_layout_commit_done_cb;
1761         data->cred = ds_cred;
1762         atomic_inc(&ds->ds_clp->cl_count);
1763         data->ds_clp = ds->ds_clp;
1764         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1765         if (fh)
1766                 data->args.fh = fh;
1767
1768         return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1769                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1770                                                &ff_layout_commit_call_ops_v4,
1771                                    how, RPC_TASK_SOFTCONN);
1772 out_err:
1773         pnfs_generic_prepare_to_resend_writes(data);
1774         pnfs_generic_commit_release(data);
1775         return -EAGAIN;
1776 }
1777
1778 static int
1779 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1780                            int how, struct nfs_commit_info *cinfo)
1781 {
1782         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1783                                             ff_layout_initiate_commit);
1784 }
1785
1786 static struct pnfs_ds_commit_info *
1787 ff_layout_get_ds_info(struct inode *inode)
1788 {
1789         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1790
1791         if (layout == NULL)
1792                 return NULL;
1793
1794         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1795 }
1796
1797 static void
1798 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1799 {
1800         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1801                                                   id_node));
1802 }
1803
1804 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1805                                   struct xdr_stream *xdr,
1806                                   const struct nfs4_layoutreturn_args *args)
1807 {
1808         struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1809         __be32 *start;
1810         int count = 0, ret = 0;
1811
1812         start = xdr_reserve_space(xdr, 4);
1813         if (unlikely(!start))
1814                 return -E2BIG;
1815
1816         /* This assume we always return _ALL_ layouts */
1817         spin_lock(&hdr->plh_inode->i_lock);
1818         ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1819         spin_unlock(&hdr->plh_inode->i_lock);
1820
1821         *start = cpu_to_be32(count);
1822
1823         return ret;
1824 }
1825
1826 /* report nothing for now */
1827 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1828                                      struct xdr_stream *xdr,
1829                                      const struct nfs4_layoutreturn_args *args)
1830 {
1831         __be32 *p;
1832
1833         p = xdr_reserve_space(xdr, 4);
1834         if (likely(p))
1835                 *p = cpu_to_be32(0);
1836 }
1837
1838 static struct nfs4_deviceid_node *
1839 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1840                               struct pnfs_device *pdev, gfp_t gfp_flags)
1841 {
1842         struct nfs4_ff_layout_ds *dsaddr;
1843
1844         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1845         if (!dsaddr)
1846                 return NULL;
1847         return &dsaddr->id_node;
1848 }
1849
1850 static void
1851 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1852                               struct xdr_stream *xdr,
1853                               const struct nfs4_layoutreturn_args *args)
1854 {
1855         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1856         __be32 *start;
1857
1858         dprintk("%s: Begin\n", __func__);
1859         start = xdr_reserve_space(xdr, 4);
1860         BUG_ON(!start);
1861
1862         if (ff_layout_encode_ioerr(flo, xdr, args))
1863                 goto out;
1864
1865         ff_layout_encode_iostats(flo, xdr, args);
1866 out:
1867         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1868         dprintk("%s: Return\n", __func__);
1869 }
1870
1871 static int
1872 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
1873 {
1874         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
1875
1876         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
1877 }
1878
1879 static size_t
1880 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
1881                           const int buflen)
1882 {
1883         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
1884         const struct in6_addr *addr = &sin6->sin6_addr;
1885
1886         /*
1887          * RFC 4291, Section 2.2.2
1888          *
1889          * Shorthanded ANY address
1890          */
1891         if (ipv6_addr_any(addr))
1892                 return snprintf(buf, buflen, "::");
1893
1894         /*
1895          * RFC 4291, Section 2.2.2
1896          *
1897          * Shorthanded loopback address
1898          */
1899         if (ipv6_addr_loopback(addr))
1900                 return snprintf(buf, buflen, "::1");
1901
1902         /*
1903          * RFC 4291, Section 2.2.3
1904          *
1905          * Special presentation address format for mapped v4
1906          * addresses.
1907          */
1908         if (ipv6_addr_v4mapped(addr))
1909                 return snprintf(buf, buflen, "::ffff:%pI4",
1910                                         &addr->s6_addr32[3]);
1911
1912         /*
1913          * RFC 4291, Section 2.2.1
1914          */
1915         return snprintf(buf, buflen, "%pI6c", addr);
1916 }
1917
1918 /* Derived from rpc_sockaddr2uaddr */
1919 static void
1920 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
1921 {
1922         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
1923         char portbuf[RPCBIND_MAXUADDRPLEN];
1924         char addrbuf[RPCBIND_MAXUADDRLEN];
1925         char *netid;
1926         unsigned short port;
1927         int len, netid_len;
1928         __be32 *p;
1929
1930         switch (sap->sa_family) {
1931         case AF_INET:
1932                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
1933                         return;
1934                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
1935                 netid = "tcp";
1936                 netid_len = 3;
1937                 break;
1938         case AF_INET6:
1939                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
1940                         return;
1941                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
1942                 netid = "tcp6";
1943                 netid_len = 4;
1944                 break;
1945         default:
1946                 /* we only support tcp and tcp6 */
1947                 WARN_ON_ONCE(1);
1948                 return;
1949         }
1950
1951         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
1952         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
1953
1954         p = xdr_reserve_space(xdr, 4 + netid_len);
1955         xdr_encode_opaque(p, netid, netid_len);
1956
1957         p = xdr_reserve_space(xdr, 4 + len);
1958         xdr_encode_opaque(p, addrbuf, len);
1959 }
1960
1961 static void
1962 ff_layout_encode_nfstime(struct xdr_stream *xdr,
1963                          ktime_t t)
1964 {
1965         struct timespec64 ts;
1966         __be32 *p;
1967
1968         p = xdr_reserve_space(xdr, 12);
1969         ts = ktime_to_timespec64(t);
1970         p = xdr_encode_hyper(p, ts.tv_sec);
1971         *p++ = cpu_to_be32(ts.tv_nsec);
1972 }
1973
1974 static void
1975 ff_layout_encode_io_latency(struct xdr_stream *xdr,
1976                             struct nfs4_ff_io_stat *stat)
1977 {
1978         __be32 *p;
1979
1980         p = xdr_reserve_space(xdr, 5 * 8);
1981         p = xdr_encode_hyper(p, stat->ops_requested);
1982         p = xdr_encode_hyper(p, stat->bytes_requested);
1983         p = xdr_encode_hyper(p, stat->ops_completed);
1984         p = xdr_encode_hyper(p, stat->bytes_completed);
1985         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
1986         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
1987         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
1988 }
1989
1990 static void
1991 ff_layout_encode_layoutstats(struct xdr_stream *xdr,
1992                              struct nfs42_layoutstat_args *args,
1993                              struct nfs42_layoutstat_devinfo *devinfo)
1994 {
1995         struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
1996         struct nfs4_pnfs_ds_addr *da;
1997         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
1998         struct nfs_fh *fh = &mirror->fh_versions[0];
1999         __be32 *p, *start;
2000
2001         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2002         dprintk("%s: DS %s: encoding address %s\n",
2003                 __func__, ds->ds_remotestr, da->da_remotestr);
2004         /* layoutupdate length */
2005         start = xdr_reserve_space(xdr, 4);
2006         /* netaddr4 */
2007         ff_layout_encode_netaddr(xdr, da);
2008         /* nfs_fh4 */
2009         p = xdr_reserve_space(xdr, 4 + fh->size);
2010         xdr_encode_opaque(p, fh->data, fh->size);
2011         /* ff_io_latency4 read */
2012         spin_lock(&mirror->lock);
2013         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2014         /* ff_io_latency4 write */
2015         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2016         spin_unlock(&mirror->lock);
2017         /* nfstime4 */
2018         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2019         /* bool */
2020         p = xdr_reserve_space(xdr, 4);
2021         *p = cpu_to_be32(false);
2022
2023         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2024 }
2025
2026 static int
2027 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
2028                                struct pnfs_layout_hdr *lo,
2029                                int dev_limit)
2030 {
2031         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2032         struct nfs4_ff_layout_mirror *mirror;
2033         struct nfs4_deviceid_node *dev;
2034         struct nfs42_layoutstat_devinfo *devinfo;
2035         int i = 0;
2036
2037         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2038                 if (i >= dev_limit)
2039                         break;
2040                 if (!mirror->mirror_ds)
2041                         continue;
2042                 /* mirror refcount put in cleanup_layoutstats */
2043                 if (!atomic_inc_not_zero(&mirror->ref))
2044                         continue;
2045                 dev = &mirror->mirror_ds->id_node;
2046                 devinfo = &args->devinfo[i];
2047                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2048                 devinfo->offset = 0;
2049                 devinfo->length = NFS4_MAX_UINT64;
2050                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2051                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2052                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2053                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2054                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2055                 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
2056                 devinfo->layout_private = mirror;
2057
2058                 i++;
2059         }
2060         return i;
2061 }
2062
2063 static int
2064 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2065 {
2066         struct nfs4_flexfile_layout *ff_layout;
2067         struct nfs4_ff_layout_mirror *mirror;
2068         int dev_count = 0;
2069
2070         spin_lock(&args->inode->i_lock);
2071         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2072         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2073                 if (atomic_read(&mirror->ref) != 0)
2074                         dev_count ++;
2075         }
2076         spin_unlock(&args->inode->i_lock);
2077         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2078         if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
2079                 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
2080                         __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
2081                 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2082         }
2083         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2084         if (!args->devinfo)
2085                 return -ENOMEM;
2086
2087         spin_lock(&args->inode->i_lock);
2088         args->num_dev = ff_layout_mirror_prepare_stats(args,
2089                         &ff_layout->generic_hdr, dev_count);
2090         spin_unlock(&args->inode->i_lock);
2091
2092         return 0;
2093 }
2094
2095 static void
2096 ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
2097 {
2098         struct nfs4_ff_layout_mirror *mirror;
2099         int i;
2100
2101         for (i = 0; i < data->args.num_dev; i++) {
2102                 mirror = data->args.devinfo[i].layout_private;
2103                 data->args.devinfo[i].layout_private = NULL;
2104                 ff_layout_put_mirror(mirror);
2105         }
2106 }
2107
2108 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2109         .id                     = LAYOUT_FLEX_FILES,
2110         .name                   = "LAYOUT_FLEX_FILES",
2111         .owner                  = THIS_MODULE,
2112         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2113         .free_layout_hdr        = ff_layout_free_layout_hdr,
2114         .alloc_lseg             = ff_layout_alloc_lseg,
2115         .free_lseg              = ff_layout_free_lseg,
2116         .add_lseg               = ff_layout_add_lseg,
2117         .pg_read_ops            = &ff_layout_pg_read_ops,
2118         .pg_write_ops           = &ff_layout_pg_write_ops,
2119         .get_ds_info            = ff_layout_get_ds_info,
2120         .free_deviceid_node     = ff_layout_free_deviceid_node,
2121         .mark_request_commit    = pnfs_layout_mark_request_commit,
2122         .clear_request_commit   = pnfs_generic_clear_request_commit,
2123         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2124         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2125         .commit_pagelist        = ff_layout_commit_pagelist,
2126         .read_pagelist          = ff_layout_read_pagelist,
2127         .write_pagelist         = ff_layout_write_pagelist,
2128         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2129         .encode_layoutreturn    = ff_layout_encode_layoutreturn,
2130         .sync                   = pnfs_nfs_generic_sync,
2131         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2132         .cleanup_layoutstats    = ff_layout_cleanup_layoutstats,
2133 };
2134
2135 static int __init nfs4flexfilelayout_init(void)
2136 {
2137         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2138                __func__);
2139         return pnfs_register_layoutdriver(&flexfilelayout_type);
2140 }
2141
2142 static void __exit nfs4flexfilelayout_exit(void)
2143 {
2144         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2145                __func__);
2146         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2147 }
2148
2149 MODULE_ALIAS("nfs-layouttype4-4");
2150
2151 MODULE_LICENSE("GPL");
2152 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2153
2154 module_init(nfs4flexfilelayout_init);
2155 module_exit(nfs4flexfilelayout_exit);