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 > NFS_MAXFHSIZE) {
  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; j++) {
 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         if (mirror->cred)
 217                 put_rpccred(mirror->cred);
 218         nfs4_ff_layout_put_deviceid(mirror->mirror_ds);
 219         kfree(mirror);
 220 }
 221
 222 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror)
 223 {
 224         if (mirror != NULL && atomic_dec_and_test(&mirror->ref))
 225                 ff_layout_free_mirror(mirror);
 226 }
 227
 228 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
 229 {
 230         int i;
 231
 232         if (fls->mirror_array) {
 233                 for (i = 0; i < fls->mirror_array_cnt; i++) {
 234                         /* normally mirror_ds is freed in
 235                          * .free_deviceid_node but we still do it here
 236                          * for .alloc_lseg error path */
 237                         ff_layout_put_mirror(fls->mirror_array[i]);
 238                 }
 239                 kfree(fls->mirror_array);
 240                 fls->mirror_array = NULL;
 241         }
 242 }
 243
 244 static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
 245 {
 246         int ret = 0;
 247
 248         dprintk("--> %s\n", __func__);
 249
 250         /* FIXME: remove this check when layout segment support is added */
 251         if (lgr->range.offset != 0 ||
 252             lgr->range.length != NFS4_MAX_UINT64) {
 253                 dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
 254                         __func__);
 255                 ret = -EINVAL;
 256         }
 257
 258         dprintk("--> %s returns %d\n", __func__, ret);
 259         return ret;
 260 }
 261
 262 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 263 {
 264         if (fls) {
 265                 ff_layout_free_mirror_array(fls);
 266                 kfree(fls);
 267         }
 268 }
 269
 270 static bool
 271 ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
 272                 const struct pnfs_layout_range *l2)
 273 {
 274         u64 end1, end2;
 275
 276         if (l1->iomode != l2->iomode)
 277                 return l1->iomode != IOMODE_READ;
 278         end1 = pnfs_calc_offset_end(l1->offset, l1->length);
 279         end2 = pnfs_calc_offset_end(l2->offset, l2->length);
 280         if (end1 < l2->offset)
 281                 return false;
 282         if (end2 < l1->offset)
 283                 return true;
 284         return l2->offset <= l1->offset;
 285 }
 286
 287 static bool
 288 ff_lseg_merge(struct pnfs_layout_segment *new,
 289                 struct pnfs_layout_segment *old)
 290 {
 291         u64 new_end, old_end;
 292
 293         if (new->pls_range.iomode != old->pls_range.iomode)
 294                 return false;
 295         old_end = pnfs_calc_offset_end(old->pls_range.offset,
 296                         old->pls_range.length);
 297         if (old_end < new->pls_range.offset)
 298                 return false;
 299         new_end = pnfs_calc_offset_end(new->pls_range.offset,
 300                         new->pls_range.length);
 301         if (new_end < old->pls_range.offset)
 302                 return false;
 303
 304         /* Mergeable: copy info from 'old' to 'new' */
 305         if (new_end < old_end)
 306                 new_end = old_end;
 307         if (new->pls_range.offset < old->pls_range.offset)
 308                 new->pls_range.offset = old->pls_range.offset;
 309         new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
 310                         new_end);
 311         if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
 312                 set_bit(NFS_LSEG_ROC, &new->pls_flags);
 313         if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
 314                 set_bit(NFS_LSEG_LAYOUTRETURN, &new->pls_flags);
 315         return true;
 316 }
 317
 318 static void
 319 ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
 320                 struct pnfs_layout_segment *lseg,
 321                 struct list_head *free_me)
 322 {
 323         pnfs_generic_layout_insert_lseg(lo, lseg,
 324                         ff_lseg_range_is_after,
 325                         ff_lseg_merge,
 326                         free_me);
 327 }
 328
 329 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 330 {
 331         int i, j;
 332
 333         for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
 334                 for (j = i + 1; j < fls->mirror_array_cnt; j++)
 335                         if (fls->mirror_array[i]->efficiency <
 336                             fls->mirror_array[j]->efficiency)
 337                                 swap(fls->mirror_array[i],
 338                                      fls->mirror_array[j]);
 339         }
 340 }
 341
 342 static void ff_layout_mark_devices_valid(struct nfs4_ff_layout_segment *fls)
 343 {
 344         struct nfs4_deviceid_node *node;
 345         int i;
 346
 347         if (!(fls->flags & FF_FLAGS_NO_IO_THRU_MDS))
 348                 return;
 349         for (i = 0; i < fls->mirror_array_cnt; i++) {
 350                 node = &fls->mirror_array[i]->mirror_ds->id_node;
 351                 clear_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags);
 352         }
 353 }
 354
 355 static struct pnfs_layout_segment *
 356 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
 357                      struct nfs4_layoutget_res *lgr,
 358                      gfp_t gfp_flags)
 359 {
 360         struct pnfs_layout_segment *ret;
 361         struct nfs4_ff_layout_segment *fls = NULL;
 362         struct xdr_stream stream;
 363         struct xdr_buf buf;
 364         struct page *scratch;
 365         u64 stripe_unit;
 366         u32 mirror_array_cnt;
 367         __be32 *p;
 368         int i, rc;
 369
 370         dprintk("--> %s\n", __func__);
 371         scratch = alloc_page(gfp_flags);
 372         if (!scratch)
 373                 return ERR_PTR(-ENOMEM);
 374
 375         xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
 376                               lgr->layoutp->len);
 377         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 378
 379         /* stripe unit and mirror_array_cnt */
 380         rc = -EIO;
 381         p = xdr_inline_decode(&stream, 8 + 4);
 382         if (!p)
 383                 goto out_err_free;
 384
 385         p = xdr_decode_hyper(p, &stripe_unit);
 386         mirror_array_cnt = be32_to_cpup(p++);
 387         dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
 388                 stripe_unit, mirror_array_cnt);
 389
 390         if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
 391             mirror_array_cnt == 0)
 392                 goto out_err_free;
 393
 394         rc = -ENOMEM;
 395         fls = kzalloc(sizeof(*fls), gfp_flags);
 396         if (!fls)
 397                 goto out_err_free;
 398
 399         fls->mirror_array_cnt = mirror_array_cnt;
 400         fls->stripe_unit = stripe_unit;
 401         fls->mirror_array = kcalloc(fls->mirror_array_cnt,
 402                                     sizeof(fls->mirror_array[0]), gfp_flags);
 403         if (fls->mirror_array == NULL)
 404                 goto out_err_free;
 405
 406         for (i = 0; i < fls->mirror_array_cnt; i++) {
 407                 struct nfs4_ff_layout_mirror *mirror;
 408                 struct nfs4_deviceid devid;
 409                 struct nfs4_deviceid_node *idnode;
 410                 u32 ds_count;
 411                 u32 fh_count;
 412                 int j;
 413
 414                 rc = -EIO;
 415                 p = xdr_inline_decode(&stream, 4);
 416                 if (!p)
 417                         goto out_err_free;
 418                 ds_count = be32_to_cpup(p);
 419
 420                 /* FIXME: allow for striping? */
 421                 if (ds_count != 1)
 422                         goto out_err_free;
 423
 424                 fls->mirror_array[i] = ff_layout_alloc_mirror(gfp_flags);
 425                 if (fls->mirror_array[i] == NULL) {
 426                         rc = -ENOMEM;
 427                         goto out_err_free;
 428                 }
 429
 430                 fls->mirror_array[i]->ds_count = ds_count;
 431
 432                 /* deviceid */
 433                 rc = decode_deviceid(&stream, &devid);
 434                 if (rc)
 435                         goto out_err_free;
 436
 437                 idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
 438                                                 &devid, lh->plh_lc_cred,
 439                                                 gfp_flags);
 440                 /*
 441                  * upon success, mirror_ds is allocated by previous
 442                  * getdeviceinfo, or newly by .alloc_deviceid_node
 443                  * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
 444                  */
 445                 if (idnode)
 446                         fls->mirror_array[i]->mirror_ds =
 447                                 FF_LAYOUT_MIRROR_DS(idnode);
 448                 else
 449                         goto out_err_free;
 450
 451                 /* efficiency */
 452                 rc = -EIO;
 453                 p = xdr_inline_decode(&stream, 4);
 454                 if (!p)
 455                         goto out_err_free;
 456                 fls->mirror_array[i]->efficiency = be32_to_cpup(p);
 457
 458                 /* stateid */
 459                 rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
 460                 if (rc)
 461                         goto out_err_free;
 462
 463                 /* fh */
 464                 rc = -EIO;
 465                 p = xdr_inline_decode(&stream, 4);
 466                 if (!p)
 467                         goto out_err_free;
 468                 fh_count = be32_to_cpup(p);
 469
 470                 fls->mirror_array[i]->fh_versions =
 471                         kzalloc(fh_count * sizeof(struct nfs_fh),
 472                                 gfp_flags);
 473                 if (fls->mirror_array[i]->fh_versions == NULL) {
 474                         rc = -ENOMEM;
 475                         goto out_err_free;
 476                 }
 477
 478                 for (j = 0; j < fh_count; j++) {
 479                         rc = decode_nfs_fh(&stream,
 480                                            &fls->mirror_array[i]->fh_versions[j]);
 481                         if (rc)
 482                                 goto out_err_free;
 483                 }
 484
 485                 fls->mirror_array[i]->fh_versions_cnt = fh_count;
 486
 487                 /* user */
 488                 rc = decode_name(&stream, &fls->mirror_array[i]->uid);
 489                 if (rc)
 490                         goto out_err_free;
 491
 492                 /* group */
 493                 rc = decode_name(&stream, &fls->mirror_array[i]->gid);
 494                 if (rc)
 495                         goto out_err_free;
 496
 497                 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]);
 498                 if (mirror != fls->mirror_array[i]) {
 499                         ff_layout_free_mirror(fls->mirror_array[i]);
 500                         fls->mirror_array[i] = mirror;
 501                 }
 502
 503                 dprintk("%s: uid %d gid %d\n", __func__,
 504                         fls->mirror_array[i]->uid,
 505                         fls->mirror_array[i]->gid);
 506         }
 507
 508         p = xdr_inline_decode(&stream, 4);
 509         if (p)
 510                 fls->flags = be32_to_cpup(p);
 511
 512         ff_layout_sort_mirrors(fls);
 513         rc = ff_layout_check_layout(lgr);
 514         if (rc)
 515                 goto out_err_free;
 516         ff_layout_mark_devices_valid(fls);
 517
 518         ret = &fls->generic_hdr;
 519         dprintk("<-- %s (success)\n", __func__);
 520 out_free_page:
 521         __free_page(scratch);
 522         return ret;
 523 out_err_free:
 524         _ff_layout_free_lseg(fls);
 525         ret = ERR_PTR(rc);
 526         dprintk("<-- %s (%d)\n", __func__, rc);
 527         goto out_free_page;
 528 }
 529
 530 static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
 531 {
 532         struct pnfs_layout_segment *lseg;
 533
 534         list_for_each_entry(lseg, &layout->plh_segs, pls_list)
 535                 if (lseg->pls_range.iomode == IOMODE_RW)
 536                         return true;
 537
 538         return false;
 539 }
 540
 541 static void
 542 ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
 543 {
 544         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 545
 546         dprintk("--> %s\n", __func__);
 547
 548         if (lseg->pls_range.iomode == IOMODE_RW) {
 549                 struct nfs4_flexfile_layout *ffl;
 550                 struct inode *inode;
 551
 552                 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
 553                 inode = ffl->generic_hdr.plh_inode;
 554                 spin_lock(&inode->i_lock);
 555                 if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
 556                         ffl->commit_info.nbuckets = 0;
 557                         kfree(ffl->commit_info.buckets);
 558                         ffl->commit_info.buckets = NULL;
 559                 }
 560                 spin_unlock(&inode->i_lock);
 561         }
 562         _ff_layout_free_lseg(fls);
 563 }
 564
 565 /* Return 1 until we have multiple lsegs support */
 566 static int
 567 ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
 568 {
 569         return 1;
 570 }
 571
 572 static void
 573 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 574 {
 575         /* first IO request? */
 576         if (atomic_inc_return(&timer->n_ops) == 1) {
 577                 timer->start_time = now;
 578         }
 579 }
 580
 581 static ktime_t
 582 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now)
 583 {
 584         ktime_t start;
 585
 586         if (atomic_dec_return(&timer->n_ops) < 0)
 587                 WARN_ON_ONCE(1);
 588
 589         start = timer->start_time;
 590         timer->start_time = now;
 591         return ktime_sub(now, start);
 592 }
 593
 594 static bool
 595 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror,
 596                             struct nfs4_ff_layoutstat *layoutstat,
 597                             ktime_t now)
 598 {
 599         static const ktime_t notime = {0};
 600         s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL;
 601
 602         nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now);
 603         if (ktime_equal(mirror->start_time, notime))
 604                 mirror->start_time = now;
 605         if (ktime_equal(mirror->last_report_time, notime))
 606                 mirror->last_report_time = now;
 607         if (layoutstats_timer != 0)
 608                 report_interval = (s64)layoutstats_timer * 1000LL;
 609         if (ktime_to_ms(ktime_sub(now, mirror->last_report_time)) >=
 610                         report_interval) {
 611                 mirror->last_report_time = now;
 612                 return true;
 613         }
 614
 615         return false;
 616 }
 617
 618 static void
 619 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat,
 620                 __u64 requested)
 621 {
 622         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 623
 624         iostat->ops_requested++;
 625         iostat->bytes_requested += requested;
 626 }
 627
 628 static void
 629 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat,
 630                 __u64 requested,
 631                 __u64 completed,
 632                 ktime_t time_completed,
 633                 ktime_t time_started)
 634 {
 635         struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat;
 636         ktime_t completion_time = ktime_sub(time_completed, time_started);
 637         ktime_t timer;
 638
 639         iostat->ops_completed++;
 640         iostat->bytes_completed += completed;
 641         iostat->bytes_not_delivered += requested - completed;
 642
 643         timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed);
 644         iostat->total_busy_time =
 645                         ktime_add(iostat->total_busy_time, timer);
 646         iostat->aggregate_completion_time =
 647                         ktime_add(iostat->aggregate_completion_time,
 648                                         completion_time);
 649 }
 650
 651 static void
 652 nfs4_ff_layout_stat_io_start_read(struct inode *inode,
 653                 struct nfs4_ff_layout_mirror *mirror,
 654                 __u64 requested, ktime_t now)
 655 {
 656         bool report;
 657
 658         spin_lock(&mirror->lock);
 659         report = nfs4_ff_layoutstat_start_io(mirror, &mirror->read_stat, now);
 660         nfs4_ff_layout_stat_io_update_requested(&mirror->read_stat, requested);
 661         spin_unlock(&mirror->lock);
 662
 663         if (report)
 664                 pnfs_report_layoutstat(inode, GFP_KERNEL);
 665 }
 666
 667 static void
 668 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task,
 669                 struct nfs4_ff_layout_mirror *mirror,
 670                 __u64 requested,
 671                 __u64 completed)
 672 {
 673         spin_lock(&mirror->lock);
 674         nfs4_ff_layout_stat_io_update_completed(&mirror->read_stat,
 675                         requested, completed,
 676                         ktime_get(), task->tk_start);
 677         spin_unlock(&mirror->lock);
 678 }
 679
 680 static void
 681 nfs4_ff_layout_stat_io_start_write(struct inode *inode,
 682                 struct nfs4_ff_layout_mirror *mirror,
 683                 __u64 requested, ktime_t now)
 684 {
 685         bool report;
 686
 687         spin_lock(&mirror->lock);
 688         report = nfs4_ff_layoutstat_start_io(mirror , &mirror->write_stat, now);
 689         nfs4_ff_layout_stat_io_update_requested(&mirror->write_stat, requested);
 690         spin_unlock(&mirror->lock);
 691
 692         if (report)
 693                 pnfs_report_layoutstat(inode, GFP_NOIO);
 694 }
 695
 696 static void
 697 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task,
 698                 struct nfs4_ff_layout_mirror *mirror,
 699                 __u64 requested,
 700                 __u64 completed,
 701                 enum nfs3_stable_how committed)
 702 {
 703         if (committed == NFS_UNSTABLE)
 704                 requested = completed = 0;
 705
 706         spin_lock(&mirror->lock);
 707         nfs4_ff_layout_stat_io_update_completed(&mirror->write_stat,
 708                         requested, completed, ktime_get(), task->tk_start);
 709         spin_unlock(&mirror->lock);
 710 }
 711
 712 static int
 713 ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
 714                             struct nfs_commit_info *cinfo,
 715                             gfp_t gfp_flags)
 716 {
 717         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 718         struct pnfs_commit_bucket *buckets;
 719         int size;
 720
 721         if (cinfo->ds->nbuckets != 0) {
 722                 /* This assumes there is only one RW lseg per file.
 723                  * To support multiple lseg per file, we need to
 724                  * change struct pnfs_commit_bucket to allow dynamic
 725                  * increasing nbuckets.
 726                  */
 727                 return 0;
 728         }
 729
 730         size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
 731
 732         buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
 733                           gfp_flags);
 734         if (!buckets)
 735                 return -ENOMEM;
 736         else {
 737                 int i;
 738
 739                 spin_lock(cinfo->lock);
 740                 if (cinfo->ds->nbuckets != 0)
 741                         kfree(buckets);
 742                 else {
 743                         cinfo->ds->buckets = buckets;
 744                         cinfo->ds->nbuckets = size;
 745                         for (i = 0; i < size; i++) {
 746                                 INIT_LIST_HEAD(&buckets[i].written);
 747                                 INIT_LIST_HEAD(&buckets[i].committing);
 748                                 /* mark direct verifier as unset */
 749                                 buckets[i].direct_verf.committed =
 750                                         NFS_INVALID_STABLE_HOW;
 751                         }
 752                 }
 753                 spin_unlock(cinfo->lock);
 754                 return 0;
 755         }
 756 }
 757
 758 static struct nfs4_pnfs_ds *
 759 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 760                                   int start_idx,
 761                                   int *best_idx)
 762 {
 763         struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
 764         struct nfs4_pnfs_ds *ds;
 765         int idx;
 766
 767         /* mirrors are sorted by efficiency */
 768         for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) {
 769                 ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
 770                 if (ds) {
 771                         *best_idx = idx;
 772                         return ds;
 773                 }
 774         }
 775
 776         return NULL;
 777 }
 778
 779 static void
 780 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 781                         struct nfs_page *req)
 782 {
 783         struct nfs_pgio_mirror *pgm;
 784         struct nfs4_ff_layout_mirror *mirror;
 785         struct nfs4_pnfs_ds *ds;
 786         int ds_idx;
 787
 788         /* Use full layout for now */
 789         if (!pgio->pg_lseg) {
 790                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 791                                                    req->wb_context,
 792                                                    0,
 793                                                    NFS4_MAX_UINT64,
 794                                                    IOMODE_READ,
 795                                                    GFP_KERNEL);
 796                 if (IS_ERR(pgio->pg_lseg)) {
 797                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 798                         pgio->pg_lseg = NULL;
 799                         return;
 800                 }
 801         }
 802         /* If no lseg, fall back to read through mds */
 803         if (pgio->pg_lseg == NULL)
 804                 goto out_mds;
 805
 806         ds = ff_layout_choose_best_ds_for_read(pgio->pg_lseg, 0, &ds_idx);
 807         if (!ds)
 808                 goto out_mds;
 809         mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
 810
 811         pgio->pg_mirror_idx = ds_idx;
 812
 813         /* read always uses only one mirror - idx 0 for pgio layer */
 814         pgm = &pgio->pg_mirrors[0];
 815         pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
 816
 817         return;
 818 out_mds:
 819         pnfs_put_lseg(pgio->pg_lseg);
 820         pgio->pg_lseg = NULL;
 821         nfs_pageio_reset_read_mds(pgio);
 822 }
 823
 824 static void
 825 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
 826                         struct nfs_page *req)
 827 {
 828         struct nfs4_ff_layout_mirror *mirror;
 829         struct nfs_pgio_mirror *pgm;
 830         struct nfs_commit_info cinfo;
 831         struct nfs4_pnfs_ds *ds;
 832         int i;
 833         int status;
 834
 835         if (!pgio->pg_lseg) {
 836                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 837                                                    req->wb_context,
 838                                                    0,
 839                                                    NFS4_MAX_UINT64,
 840                                                    IOMODE_RW,
 841                                                    GFP_NOFS);
 842                 if (IS_ERR(pgio->pg_lseg)) {
 843                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 844                         pgio->pg_lseg = NULL;
 845                         return;
 846                 }
 847         }
 848         /* If no lseg, fall back to write through mds */
 849         if (pgio->pg_lseg == NULL)
 850                 goto out_mds;
 851
 852         nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
 853         status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
 854         if (status < 0)
 855                 goto out_mds;
 856
 857         /* Use a direct mapping of ds_idx to pgio mirror_idx */
 858         if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg))
 859                 goto out_eagain;
 860
 861         for (i = 0; i < pgio->pg_mirror_count; i++) {
 862                 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
 863                 if (!ds)
 864                         goto out_mds;
 865                 pgm = &pgio->pg_mirrors[i];
 866                 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
 867                 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
 868         }
 869
 870         return;
 871 out_eagain:
 872         pnfs_generic_pg_cleanup(pgio);
 873         pgio->pg_error = -EAGAIN;
 874         return;
 875 out_mds:
 876         pnfs_put_lseg(pgio->pg_lseg);
 877         pgio->pg_lseg = NULL;
 878         nfs_pageio_reset_write_mds(pgio);
 879         pgio->pg_error = -EAGAIN;
 880 }
 881
 882 static unsigned int
 883 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
 884                                     struct nfs_page *req)
 885 {
 886         if (!pgio->pg_lseg) {
 887                 pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
 888                                                    req->wb_context,
 889                                                    0,
 890                                                    NFS4_MAX_UINT64,
 891                                                    IOMODE_RW,
 892                                                    GFP_NOFS);
 893                 if (IS_ERR(pgio->pg_lseg)) {
 894                         pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 895                         pgio->pg_lseg = NULL;
 896                         goto out;
 897                 }
 898         }
 899         if (pgio->pg_lseg)
 900                 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
 901
 902         /* no lseg means that pnfs is not in use, so no mirroring here */
 903         nfs_pageio_reset_write_mds(pgio);
 904 out:
 905         return 1;
 906 }
 907
 908 static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
 909         .pg_init = ff_layout_pg_init_read,
 910         .pg_test = pnfs_generic_pg_test,
 911         .pg_doio = pnfs_generic_pg_readpages,
 912         .pg_cleanup = pnfs_generic_pg_cleanup,
 913 };
 914
 915 static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
 916         .pg_init = ff_layout_pg_init_write,
 917         .pg_test = pnfs_generic_pg_test,
 918         .pg_doio = pnfs_generic_pg_writepages,
 919         .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
 920         .pg_cleanup = pnfs_generic_pg_cleanup,
 921 };
 922
 923 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
 924 {
 925         struct rpc_task *task = &hdr->task;
 926
 927         pnfs_layoutcommit_inode(hdr->inode, false);
 928
 929         if (retry_pnfs) {
 930                 dprintk("%s Reset task %5u for i/o through pNFS "
 931                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 932                         hdr->task.tk_pid,
 933                         hdr->inode->i_sb->s_id,
 934                         (unsigned long long)NFS_FILEID(hdr->inode),
 935                         hdr->args.count,
 936                         (unsigned long long)hdr->args.offset);
 937
 938                 if (!hdr->dreq) {
 939                         struct nfs_open_context *ctx;
 940
 941                         ctx = nfs_list_entry(hdr->pages.next)->wb_context;
 942                         set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
 943                         hdr->completion_ops->error_cleanup(&hdr->pages);
 944                 } else {
 945                         nfs_direct_set_resched_writes(hdr->dreq);
 946                         /* fake unstable write to let common nfs resend pages */
 947                         hdr->verf.committed = NFS_UNSTABLE;
 948                         hdr->good_bytes = hdr->args.count;
 949                 }
 950                 return;
 951         }
 952
 953         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 954                 dprintk("%s Reset task %5u for i/o through MDS "
 955                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 956                         hdr->task.tk_pid,
 957                         hdr->inode->i_sb->s_id,
 958                         (unsigned long long)NFS_FILEID(hdr->inode),
 959                         hdr->args.count,
 960                         (unsigned long long)hdr->args.offset);
 961
 962                 task->tk_status = pnfs_write_done_resend_to_mds(hdr);
 963         }
 964 }
 965
 966 static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
 967 {
 968         struct rpc_task *task = &hdr->task;
 969
 970         pnfs_layoutcommit_inode(hdr->inode, false);
 971
 972         if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
 973                 dprintk("%s Reset task %5u for i/o through MDS "
 974                         "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
 975                         hdr->task.tk_pid,
 976                         hdr->inode->i_sb->s_id,
 977                         (unsigned long long)NFS_FILEID(hdr->inode),
 978                         hdr->args.count,
 979                         (unsigned long long)hdr->args.offset);
 980
 981                 task->tk_status = pnfs_read_done_resend_to_mds(hdr);
 982         }
 983 }
 984
 985 static int ff_layout_async_handle_error_v4(struct rpc_task *task,
 986                                            struct nfs4_state *state,
 987                                            struct nfs_client *clp,
 988                                            struct pnfs_layout_segment *lseg,
 989                                            int idx)
 990 {
 991         struct pnfs_layout_hdr *lo = lseg->pls_layout;
 992         struct inode *inode = lo->plh_inode;
 993         struct nfs_server *mds_server = NFS_SERVER(inode);
 994
 995         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
 996         struct nfs_client *mds_client = mds_server->nfs_client;
 997         struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
 998
 999         if (task->tk_status >= 0)
1000                 return 0;
1001
1002         switch (task->tk_status) {
1003         /* MDS state errors */
1004         case -NFS4ERR_DELEG_REVOKED:
1005         case -NFS4ERR_ADMIN_REVOKED:
1006         case -NFS4ERR_BAD_STATEID:
1007                 if (state == NULL)
1008                         break;
1009                 nfs_remove_bad_delegation(state->inode);
1010         case -NFS4ERR_OPENMODE:
1011                 if (state == NULL)
1012                         break;
1013                 if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1014                         goto out_bad_stateid;
1015                 goto wait_on_recovery;
1016         case -NFS4ERR_EXPIRED:
1017                 if (state != NULL) {
1018                         if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
1019                                 goto out_bad_stateid;
1020                 }
1021                 nfs4_schedule_lease_recovery(mds_client);
1022                 goto wait_on_recovery;
1023         /* DS session errors */
1024         case -NFS4ERR_BADSESSION:
1025         case -NFS4ERR_BADSLOT:
1026         case -NFS4ERR_BAD_HIGH_SLOT:
1027         case -NFS4ERR_DEADSESSION:
1028         case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1029         case -NFS4ERR_SEQ_FALSE_RETRY:
1030         case -NFS4ERR_SEQ_MISORDERED:
1031                 dprintk("%s ERROR %d, Reset session. Exchangeid "
1032                         "flags 0x%x\n", __func__, task->tk_status,
1033                         clp->cl_exchange_flags);
1034                 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
1035                 break;
1036         case -NFS4ERR_DELAY:
1037         case -NFS4ERR_GRACE:
1038                 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
1039                 break;
1040         case -NFS4ERR_RETRY_UNCACHED_REP:
1041                 break;
1042         /* Invalidate Layout errors */
1043         case -NFS4ERR_PNFS_NO_LAYOUT:
1044         case -ESTALE:           /* mapped NFS4ERR_STALE */
1045         case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
1046         case -EISDIR:           /* mapped NFS4ERR_ISDIR */
1047         case -NFS4ERR_FHEXPIRED:
1048         case -NFS4ERR_WRONG_TYPE:
1049                 dprintk("%s Invalid layout error %d\n", __func__,
1050                         task->tk_status);
1051                 /*
1052                  * Destroy layout so new i/o will get a new layout.
1053                  * Layout will not be destroyed until all current lseg
1054                  * references are put. Mark layout as invalid to resend failed
1055                  * i/o and all i/o waiting on the slot table to the MDS until
1056                  * layout is destroyed and a new valid layout is obtained.
1057                  */
1058                 pnfs_destroy_layout(NFS_I(inode));
1059                 rpc_wake_up(&tbl->slot_tbl_waitq);
1060                 goto reset;
1061         /* RPC connection errors */
1062         case -ECONNREFUSED:
1063         case -EHOSTDOWN:
1064         case -EHOSTUNREACH:
1065         case -ENETUNREACH:
1066         case -EIO:
1067         case -ETIMEDOUT:
1068         case -EPIPE:
1069                 dprintk("%s DS connection error %d\n", __func__,
1070                         task->tk_status);
1071                 nfs4_mark_deviceid_unavailable(devid);
1072                 rpc_wake_up(&tbl->slot_tbl_waitq);
1073                 /* fall through */
1074         default:
1075                 if (ff_layout_no_fallback_to_mds(lseg) ||
1076                     ff_layout_has_available_ds(lseg))
1077                         return -NFS4ERR_RESET_TO_PNFS;
1078 reset:
1079                 dprintk("%s Retry through MDS. Error %d\n", __func__,
1080                         task->tk_status);
1081                 return -NFS4ERR_RESET_TO_MDS;
1082         }
1083 out:
1084         task->tk_status = 0;
1085         return -EAGAIN;
1086 out_bad_stateid:
1087         task->tk_status = -EIO;
1088         return 0;
1089 wait_on_recovery:
1090         rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
1091         if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
1092                 rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
1093         goto out;
1094 }
1095
1096 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
1097 static int ff_layout_async_handle_error_v3(struct rpc_task *task,
1098                                            struct pnfs_layout_segment *lseg,
1099                                            int idx)
1100 {
1101         struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
1102
1103         if (task->tk_status >= 0)
1104                 return 0;
1105
1106         switch (task->tk_status) {
1107         /* File access problems. Don't mark the device as unavailable */
1108         case -EACCES:
1109         case -ESTALE:
1110         case -EISDIR:
1111         case -EBADHANDLE:
1112         case -ELOOP:
1113         case -ENOSPC:
1114                 break;
1115         case -EJUKEBOX:
1116                 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
1117                 goto out_retry;
1118         default:
1119                 dprintk("%s DS connection error %d\n", __func__,
1120                         task->tk_status);
1121                 nfs4_mark_deviceid_unavailable(devid);
1122         }
1123         /* FIXME: Need to prevent infinite looping here. */
1124         return -NFS4ERR_RESET_TO_PNFS;
1125 out_retry:
1126         task->tk_status = 0;
1127         rpc_restart_call(task);
1128         rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
1129         return -EAGAIN;
1130 }
1131
1132 static int ff_layout_async_handle_error(struct rpc_task *task,
1133                                         struct nfs4_state *state,
1134                                         struct nfs_client *clp,
1135                                         struct pnfs_layout_segment *lseg,
1136                                         int idx)
1137 {
1138         int vers = clp->cl_nfs_mod->rpc_vers->number;
1139
1140         switch (vers) {
1141         case 3:
1142                 return ff_layout_async_handle_error_v3(task, lseg, idx);
1143         case 4:
1144                 return ff_layout_async_handle_error_v4(task, state, clp,
1145                                                        lseg, idx);
1146         default:
1147                 /* should never happen */
1148                 WARN_ON_ONCE(1);
1149                 return 0;
1150         }
1151 }
1152
1153 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
1154                                         int idx, u64 offset, u64 length,
1155                                         u32 status, int opnum, int error)
1156 {
1157         struct nfs4_ff_layout_mirror *mirror;
1158         int err;
1159
1160         if (status == 0) {
1161                 switch (error) {
1162                 case -ETIMEDOUT:
1163                 case -EPFNOSUPPORT:
1164                 case -EPROTONOSUPPORT:
1165                 case -EOPNOTSUPP:
1166                 case -ECONNREFUSED:
1167                 case -ECONNRESET:
1168                 case -EHOSTDOWN:
1169                 case -EHOSTUNREACH:
1170                 case -ENETUNREACH:
1171                 case -EADDRINUSE:
1172                 case -ENOBUFS:
1173                 case -EPIPE:
1174                 case -EPERM:
1175                         status = NFS4ERR_NXIO;
1176                         break;
1177                 case -EACCES:
1178                         status = NFS4ERR_ACCESS;
1179                         break;
1180                 default:
1181                         return;
1182                 }
1183         }
1184
1185         mirror = FF_LAYOUT_COMP(lseg, idx);
1186         err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
1187                                        mirror, offset, length, status, opnum,
1188                                        GFP_NOIO);
1189         pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, lseg);
1190         dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
1191 }
1192
1193 /* NFS_PROTO call done callback routines */
1194 static int ff_layout_read_done_cb(struct rpc_task *task,
1195                                 struct nfs_pgio_header *hdr)
1196 {
1197         int err;
1198
1199         trace_nfs4_pnfs_read(hdr, task->tk_status);
1200         if (task->tk_status < 0)
1201                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1202                                             hdr->args.offset, hdr->args.count,
1203                                             hdr->res.op_status, OP_READ,
1204                                             task->tk_status);
1205         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1206                                            hdr->ds_clp, hdr->lseg,
1207                                            hdr->pgio_mirror_idx);
1208
1209         switch (err) {
1210         case -NFS4ERR_RESET_TO_PNFS:
1211                 if (ff_layout_choose_best_ds_for_read(hdr->lseg,
1212                                         hdr->pgio_mirror_idx + 1,
1213                                         &hdr->pgio_mirror_idx))
1214                         goto out_eagain;
1215                 set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
1216                         &hdr->lseg->pls_layout->plh_flags);
1217                 pnfs_read_resend_pnfs(hdr);
1218                 return task->tk_status;
1219         case -NFS4ERR_RESET_TO_MDS:
1220                 ff_layout_reset_read(hdr);
1221                 return task->tk_status;
1222         case -EAGAIN:
1223                 goto out_eagain;
1224         }
1225
1226         return 0;
1227 out_eagain:
1228         rpc_restart_call_prepare(task);
1229         return -EAGAIN;
1230 }
1231
1232 static bool
1233 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg)
1234 {
1235         return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT);
1236 }
1237
1238 /*
1239  * We reference the rpc_cred of the first WRITE that triggers the need for
1240  * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
1241  * rfc5661 is not clear about which credential should be used.
1242  *
1243  * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
1244  * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
1245  * we always send layoutcommit after DS writes.
1246  */
1247 static void
1248 ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
1249 {
1250         if (!ff_layout_need_layoutcommit(hdr->lseg))
1251                 return;
1252
1253         pnfs_set_layoutcommit(hdr->inode, hdr->lseg,
1254                         hdr->mds_offset + hdr->res.count);
1255         dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
1256                 (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
1257 }
1258
1259 static bool
1260 ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
1261 {
1262         /* No mirroring for now */
1263         struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
1264
1265         return ff_layout_test_devid_unavailable(node);
1266 }
1267
1268 static int ff_layout_read_prepare_common(struct rpc_task *task,
1269                                          struct nfs_pgio_header *hdr)
1270 {
1271         nfs4_ff_layout_stat_io_start_read(hdr->inode,
1272                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1273                         hdr->args.count,
1274                         task->tk_start);
1275
1276         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1277                 rpc_exit(task, -EIO);
1278                 return -EIO;
1279         }
1280         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1281                 dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
1282                 if (ff_layout_has_available_ds(hdr->lseg))
1283                         pnfs_read_resend_pnfs(hdr);
1284                 else
1285                         ff_layout_reset_read(hdr);
1286                 rpc_exit(task, 0);
1287                 return -EAGAIN;
1288         }
1289         hdr->pgio_done_cb = ff_layout_read_done_cb;
1290
1291         return 0;
1292 }
1293
1294 /*
1295  * Call ops for the async read/write cases
1296  * In the case of dense layouts, the offset needs to be reset to its
1297  * original value.
1298  */
1299 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
1300 {
1301         struct nfs_pgio_header *hdr = data;
1302
1303         if (ff_layout_read_prepare_common(task, hdr))
1304                 return;
1305
1306         rpc_call_start(task);
1307 }
1308
1309 static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
1310                                     struct nfs4_sequence_args *args,
1311                                     struct nfs4_sequence_res *res,
1312                                     struct rpc_task *task)
1313 {
1314         if (ds_clp->cl_session)
1315                 return nfs41_setup_sequence(ds_clp->cl_session,
1316                                            args,
1317                                            res,
1318                                            task);
1319         return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
1320                                    args,
1321                                    res,
1322                                    task);
1323 }
1324
1325 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
1326 {
1327         struct nfs_pgio_header *hdr = data;
1328
1329         if (ff_layout_setup_sequence(hdr->ds_clp,
1330                                      &hdr->args.seq_args,
1331                                      &hdr->res.seq_res,
1332                                      task))
1333                 return;
1334
1335         if (ff_layout_read_prepare_common(task, hdr))
1336                 return;
1337
1338         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1339                         hdr->args.lock_context, FMODE_READ) == -EIO)
1340                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1341 }
1342
1343 static void ff_layout_read_call_done(struct rpc_task *task, void *data)
1344 {
1345         struct nfs_pgio_header *hdr = data;
1346
1347         dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
1348
1349         nfs4_ff_layout_stat_io_end_read(task,
1350                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1351                         hdr->args.count, hdr->res.count);
1352
1353         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1354             task->tk_status == 0) {
1355                 nfs4_sequence_done(task, &hdr->res.seq_res);
1356                 return;
1357         }
1358
1359         /* Note this may cause RPC to be resent */
1360         hdr->mds_ops->rpc_call_done(task, hdr);
1361 }
1362
1363 static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
1364 {
1365         struct nfs_pgio_header *hdr = data;
1366
1367         rpc_count_iostats_metrics(task,
1368             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
1369 }
1370
1371 static int ff_layout_write_done_cb(struct rpc_task *task,
1372                                 struct nfs_pgio_header *hdr)
1373 {
1374         int err;
1375
1376         trace_nfs4_pnfs_write(hdr, task->tk_status);
1377         if (task->tk_status < 0)
1378                 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
1379                                             hdr->args.offset, hdr->args.count,
1380                                             hdr->res.op_status, OP_WRITE,
1381                                             task->tk_status);
1382         err = ff_layout_async_handle_error(task, hdr->args.context->state,
1383                                            hdr->ds_clp, hdr->lseg,
1384                                            hdr->pgio_mirror_idx);
1385
1386         switch (err) {
1387         case -NFS4ERR_RESET_TO_PNFS:
1388                 pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
1389                 ff_layout_reset_write(hdr, true);
1390                 return task->tk_status;
1391         case -NFS4ERR_RESET_TO_MDS:
1392                 pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
1393                 ff_layout_reset_write(hdr, false);
1394                 return task->tk_status;
1395         case -EAGAIN:
1396                 rpc_restart_call_prepare(task);
1397                 return -EAGAIN;
1398         }
1399
1400         if (hdr->res.verf->committed == NFS_FILE_SYNC ||
1401             hdr->res.verf->committed == NFS_DATA_SYNC)
1402                 ff_layout_set_layoutcommit(hdr);
1403
1404         /* zero out fattr since we don't care DS attr at all */
1405         hdr->fattr.valid = 0;
1406         if (task->tk_status >= 0)
1407                 nfs_writeback_update_inode(hdr);
1408
1409         return 0;
1410 }
1411
1412 static int ff_layout_commit_done_cb(struct rpc_task *task,
1413                                      struct nfs_commit_data *data)
1414 {
1415         int err;
1416
1417         trace_nfs4_pnfs_commit_ds(data, task->tk_status);
1418         if (task->tk_status < 0)
1419                 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
1420                                             data->args.offset, data->args.count,
1421                                             data->res.op_status, OP_COMMIT,
1422                                             task->tk_status);
1423         err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
1424                                            data->lseg, data->ds_commit_index);
1425
1426         switch (err) {
1427         case -NFS4ERR_RESET_TO_PNFS:
1428                 pnfs_set_retry_layoutget(data->lseg->pls_layout);
1429                 pnfs_generic_prepare_to_resend_writes(data);
1430                 return -EAGAIN;
1431         case -NFS4ERR_RESET_TO_MDS:
1432                 pnfs_clear_retry_layoutget(data->lseg->pls_layout);
1433                 pnfs_generic_prepare_to_resend_writes(data);
1434                 return -EAGAIN;
1435         case -EAGAIN:
1436                 rpc_restart_call_prepare(task);
1437                 return -EAGAIN;
1438         }
1439
1440         if (ff_layout_need_layoutcommit(data->lseg))
1441                 pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb);
1442
1443         return 0;
1444 }
1445
1446 static int ff_layout_write_prepare_common(struct rpc_task *task,
1447                                           struct nfs_pgio_header *hdr)
1448 {
1449         nfs4_ff_layout_stat_io_start_write(hdr->inode,
1450                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1451                         hdr->args.count,
1452                         task->tk_start);
1453
1454         if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
1455                 rpc_exit(task, -EIO);
1456                 return -EIO;
1457         }
1458
1459         if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
1460                 bool retry_pnfs;
1461
1462                 retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
1463                 dprintk("%s task %u reset io to %s\n", __func__,
1464                         task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
1465                 ff_layout_reset_write(hdr, retry_pnfs);
1466                 rpc_exit(task, 0);
1467                 return -EAGAIN;
1468         }
1469
1470         return 0;
1471 }
1472
1473 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
1474 {
1475         struct nfs_pgio_header *hdr = data;
1476
1477         if (ff_layout_write_prepare_common(task, hdr))
1478                 return;
1479
1480         rpc_call_start(task);
1481 }
1482
1483 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
1484 {
1485         struct nfs_pgio_header *hdr = data;
1486
1487         if (ff_layout_setup_sequence(hdr->ds_clp,
1488                                      &hdr->args.seq_args,
1489                                      &hdr->res.seq_res,
1490                                      task))
1491                 return;
1492
1493         if (ff_layout_write_prepare_common(task, hdr))
1494                 return;
1495
1496         if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
1497                         hdr->args.lock_context, FMODE_WRITE) == -EIO)
1498                 rpc_exit(task, -EIO); /* lost lock, terminate I/O */
1499 }
1500
1501 static void ff_layout_write_call_done(struct rpc_task *task, void *data)
1502 {
1503         struct nfs_pgio_header *hdr = data;
1504
1505         nfs4_ff_layout_stat_io_end_write(task,
1506                         FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx),
1507                         hdr->args.count, hdr->res.count,
1508                         hdr->res.verf->committed);
1509
1510         if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
1511             task->tk_status == 0) {
1512                 nfs4_sequence_done(task, &hdr->res.seq_res);
1513                 return;
1514         }
1515
1516         /* Note this may cause RPC to be resent */
1517         hdr->mds_ops->rpc_call_done(task, hdr);
1518 }
1519
1520 static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
1521 {
1522         struct nfs_pgio_header *hdr = data;
1523
1524         rpc_count_iostats_metrics(task,
1525             &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
1526 }
1527
1528 static void ff_layout_commit_prepare_common(struct rpc_task *task,
1529                 struct nfs_commit_data *cdata)
1530 {
1531         nfs4_ff_layout_stat_io_start_write(cdata->inode,
1532                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1533                         0, task->tk_start);
1534 }
1535
1536 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
1537 {
1538         ff_layout_commit_prepare_common(task, data);
1539         rpc_call_start(task);
1540 }
1541
1542 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
1543 {
1544         struct nfs_commit_data *wdata = data;
1545
1546         if (ff_layout_setup_sequence(wdata->ds_clp,
1547                                  &wdata->args.seq_args,
1548                                  &wdata->res.seq_res,
1549                                  task))
1550                 return;
1551         ff_layout_commit_prepare_common(task, data);
1552 }
1553
1554 static void ff_layout_commit_done(struct rpc_task *task, void *data)
1555 {
1556         struct nfs_commit_data *cdata = data;
1557         struct nfs_page *req;
1558         __u64 count = 0;
1559
1560         if (task->tk_status == 0) {
1561                 list_for_each_entry(req, &cdata->pages, wb_list)
1562                         count += req->wb_bytes;
1563         }
1564
1565         nfs4_ff_layout_stat_io_end_write(task,
1566                         FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index),
1567                         count, count, NFS_FILE_SYNC);
1568
1569         pnfs_generic_write_commit_done(task, data);
1570 }
1571
1572 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
1573 {
1574         struct nfs_commit_data *cdata = data;
1575
1576         rpc_count_iostats_metrics(task,
1577             &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
1578 }
1579
1580 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
1581         .rpc_call_prepare = ff_layout_read_prepare_v3,
1582         .rpc_call_done = ff_layout_read_call_done,
1583         .rpc_count_stats = ff_layout_read_count_stats,
1584         .rpc_release = pnfs_generic_rw_release,
1585 };
1586
1587 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
1588         .rpc_call_prepare = ff_layout_read_prepare_v4,
1589         .rpc_call_done = ff_layout_read_call_done,
1590         .rpc_count_stats = ff_layout_read_count_stats,
1591         .rpc_release = pnfs_generic_rw_release,
1592 };
1593
1594 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
1595         .rpc_call_prepare = ff_layout_write_prepare_v3,
1596         .rpc_call_done = ff_layout_write_call_done,
1597         .rpc_count_stats = ff_layout_write_count_stats,
1598         .rpc_release = pnfs_generic_rw_release,
1599 };
1600
1601 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
1602         .rpc_call_prepare = ff_layout_write_prepare_v4,
1603         .rpc_call_done = ff_layout_write_call_done,
1604         .rpc_count_stats = ff_layout_write_count_stats,
1605         .rpc_release = pnfs_generic_rw_release,
1606 };
1607
1608 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
1609         .rpc_call_prepare = ff_layout_commit_prepare_v3,
1610         .rpc_call_done = ff_layout_commit_done,
1611         .rpc_count_stats = ff_layout_commit_count_stats,
1612         .rpc_release = pnfs_generic_commit_release,
1613 };
1614
1615 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
1616         .rpc_call_prepare = ff_layout_commit_prepare_v4,
1617         .rpc_call_done = ff_layout_commit_done,
1618         .rpc_count_stats = ff_layout_commit_count_stats,
1619         .rpc_release = pnfs_generic_commit_release,
1620 };
1621
1622 static enum pnfs_try_status
1623 ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
1624 {
1625         struct pnfs_layout_segment *lseg = hdr->lseg;
1626         struct nfs4_pnfs_ds *ds;
1627         struct rpc_clnt *ds_clnt;
1628         struct rpc_cred *ds_cred;
1629         loff_t offset = hdr->args.offset;
1630         u32 idx = hdr->pgio_mirror_idx;
1631         int vers;
1632         struct nfs_fh *fh;
1633
1634         dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
1635                 __func__, hdr->inode->i_ino,
1636                 hdr->args.pgbase, (size_t)hdr->args.count, offset);
1637
1638         ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
1639         if (!ds)
1640                 goto out_failed;
1641
1642         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1643                                                    hdr->inode);
1644         if (IS_ERR(ds_clnt))
1645                 goto out_failed;
1646
1647         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1648         if (IS_ERR(ds_cred))
1649                 goto out_failed;
1650
1651         vers = nfs4_ff_layout_ds_version(lseg, idx);
1652
1653         dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
1654                 ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
1655
1656         atomic_inc(&ds->ds_clp->cl_count);
1657         hdr->ds_clp = ds->ds_clp;
1658         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1659         if (fh)
1660                 hdr->args.fh = fh;
1661         /*
1662          * Note that if we ever decide to split across DSes,
1663          * then we may need to handle dense-like offsets.
1664          */
1665         hdr->args.offset = offset;
1666         hdr->mds_offset = offset;
1667
1668         /* Perform an asynchronous read to ds */
1669         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1670                           vers == 3 ? &ff_layout_read_call_ops_v3 :
1671                                       &ff_layout_read_call_ops_v4,
1672                           0, RPC_TASK_SOFTCONN);
1673
1674         return PNFS_ATTEMPTED;
1675
1676 out_failed:
1677         if (ff_layout_has_available_ds(lseg))
1678                 return PNFS_TRY_AGAIN;
1679         return PNFS_NOT_ATTEMPTED;
1680 }
1681
1682 /* Perform async writes. */
1683 static enum pnfs_try_status
1684 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
1685 {
1686         struct pnfs_layout_segment *lseg = hdr->lseg;
1687         struct nfs4_pnfs_ds *ds;
1688         struct rpc_clnt *ds_clnt;
1689         struct rpc_cred *ds_cred;
1690         loff_t offset = hdr->args.offset;
1691         int vers;
1692         struct nfs_fh *fh;
1693         int idx = hdr->pgio_mirror_idx;
1694
1695         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1696         if (!ds)
1697                 return PNFS_NOT_ATTEMPTED;
1698
1699         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1700                                                    hdr->inode);
1701         if (IS_ERR(ds_clnt))
1702                 return PNFS_NOT_ATTEMPTED;
1703
1704         ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
1705         if (IS_ERR(ds_cred))
1706                 return PNFS_NOT_ATTEMPTED;
1707
1708         vers = nfs4_ff_layout_ds_version(lseg, idx);
1709
1710         dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
1711                 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
1712                 offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
1713                 vers);
1714
1715         hdr->pgio_done_cb = ff_layout_write_done_cb;
1716         atomic_inc(&ds->ds_clp->cl_count);
1717         hdr->ds_clp = ds->ds_clp;
1718         hdr->ds_commit_idx = idx;
1719         fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
1720         if (fh)
1721                 hdr->args.fh = fh;
1722
1723         /*
1724          * Note that if we ever decide to split across DSes,
1725          * then we may need to handle dense-like offsets.
1726          */
1727         hdr->args.offset = offset;
1728
1729         /* Perform an asynchronous write */
1730         nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
1731                           vers == 3 ? &ff_layout_write_call_ops_v3 :
1732                                       &ff_layout_write_call_ops_v4,
1733                           sync, RPC_TASK_SOFTCONN);
1734         return PNFS_ATTEMPTED;
1735 }
1736
1737 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1738 {
1739         return i;
1740 }
1741
1742 static struct nfs_fh *
1743 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
1744 {
1745         struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
1746
1747         /* FIXME: Assume that there is only one NFS version available
1748          * for the DS.
1749          */
1750         return &flseg->mirror_array[i]->fh_versions[0];
1751 }
1752
1753 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
1754 {
1755         struct pnfs_layout_segment *lseg = data->lseg;
1756         struct nfs4_pnfs_ds *ds;
1757         struct rpc_clnt *ds_clnt;
1758         struct rpc_cred *ds_cred;
1759         u32 idx;
1760         int vers;
1761         struct nfs_fh *fh;
1762
1763         idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
1764         ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
1765         if (!ds)
1766                 goto out_err;
1767
1768         ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
1769                                                    data->inode);
1770         if (IS_ERR(ds_clnt))
1771                 goto out_err;
1772
1773         ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
1774         if (IS_ERR(ds_cred))
1775                 goto out_err;
1776
1777         vers = nfs4_ff_layout_ds_version(lseg, idx);
1778
1779         dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
1780                 data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
1781                 vers);
1782         data->commit_done_cb = ff_layout_commit_done_cb;
1783         data->cred = ds_cred;
1784         atomic_inc(&ds->ds_clp->cl_count);
1785         data->ds_clp = ds->ds_clp;
1786         fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
1787         if (fh)
1788                 data->args.fh = fh;
1789
1790         return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
1791                                    vers == 3 ? &ff_layout_commit_call_ops_v3 :
1792                                                &ff_layout_commit_call_ops_v4,
1793                                    how, RPC_TASK_SOFTCONN);
1794 out_err:
1795         pnfs_generic_prepare_to_resend_writes(data);
1796         pnfs_generic_commit_release(data);
1797         return -EAGAIN;
1798 }
1799
1800 static int
1801 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
1802                            int how, struct nfs_commit_info *cinfo)
1803 {
1804         return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
1805                                             ff_layout_initiate_commit);
1806 }
1807
1808 static struct pnfs_ds_commit_info *
1809 ff_layout_get_ds_info(struct inode *inode)
1810 {
1811         struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
1812
1813         if (layout == NULL)
1814                 return NULL;
1815
1816         return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
1817 }
1818
1819 static void
1820 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d)
1821 {
1822         nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
1823                                                   id_node));
1824 }
1825
1826 static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
1827                                   struct xdr_stream *xdr,
1828                                   const struct nfs4_layoutreturn_args *args)
1829 {
1830         struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
1831         __be32 *start;
1832         int count = 0, ret = 0;
1833
1834         start = xdr_reserve_space(xdr, 4);
1835         if (unlikely(!start))
1836                 return -E2BIG;
1837
1838         /* This assume we always return _ALL_ layouts */
1839         spin_lock(&hdr->plh_inode->i_lock);
1840         ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
1841         spin_unlock(&hdr->plh_inode->i_lock);
1842
1843         *start = cpu_to_be32(count);
1844
1845         return ret;
1846 }
1847
1848 /* report nothing for now */
1849 static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
1850                                      struct xdr_stream *xdr,
1851                                      const struct nfs4_layoutreturn_args *args)
1852 {
1853         __be32 *p;
1854
1855         p = xdr_reserve_space(xdr, 4);
1856         if (likely(p))
1857                 *p = cpu_to_be32(0);
1858 }
1859
1860 static struct nfs4_deviceid_node *
1861 ff_layout_alloc_deviceid_node(struct nfs_server *server,
1862                               struct pnfs_device *pdev, gfp_t gfp_flags)
1863 {
1864         struct nfs4_ff_layout_ds *dsaddr;
1865
1866         dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
1867         if (!dsaddr)
1868                 return NULL;
1869         return &dsaddr->id_node;
1870 }
1871
1872 static void
1873 ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
1874                               struct xdr_stream *xdr,
1875                               const struct nfs4_layoutreturn_args *args)
1876 {
1877         struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
1878         __be32 *start;
1879
1880         dprintk("%s: Begin\n", __func__);
1881         start = xdr_reserve_space(xdr, 4);
1882         BUG_ON(!start);
1883
1884         ff_layout_encode_ioerr(flo, xdr, args);
1885         ff_layout_encode_iostats(flo, xdr, args);
1886
1887         *start = cpu_to_be32((xdr->p - start - 1) * 4);
1888         dprintk("%s: Return\n", __func__);
1889 }
1890
1891 static int
1892 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen)
1893 {
1894         const struct sockaddr_in *sin = (struct sockaddr_in *)sap;
1895
1896         return snprintf(buf, buflen, "%pI4", &sin->sin_addr);
1897 }
1898
1899 static size_t
1900 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf,
1901                           const int buflen)
1902 {
1903         const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap;
1904         const struct in6_addr *addr = &sin6->sin6_addr;
1905
1906         /*
1907          * RFC 4291, Section 2.2.2
1908          *
1909          * Shorthanded ANY address
1910          */
1911         if (ipv6_addr_any(addr))
1912                 return snprintf(buf, buflen, "::");
1913
1914         /*
1915          * RFC 4291, Section 2.2.2
1916          *
1917          * Shorthanded loopback address
1918          */
1919         if (ipv6_addr_loopback(addr))
1920                 return snprintf(buf, buflen, "::1");
1921
1922         /*
1923          * RFC 4291, Section 2.2.3
1924          *
1925          * Special presentation address format for mapped v4
1926          * addresses.
1927          */
1928         if (ipv6_addr_v4mapped(addr))
1929                 return snprintf(buf, buflen, "::ffff:%pI4",
1930                                         &addr->s6_addr32[3]);
1931
1932         /*
1933          * RFC 4291, Section 2.2.1
1934          */
1935         return snprintf(buf, buflen, "%pI6c", addr);
1936 }
1937
1938 /* Derived from rpc_sockaddr2uaddr */
1939 static void
1940 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da)
1941 {
1942         struct sockaddr *sap = (struct sockaddr *)&da->da_addr;
1943         char portbuf[RPCBIND_MAXUADDRPLEN];
1944         char addrbuf[RPCBIND_MAXUADDRLEN];
1945         char *netid;
1946         unsigned short port;
1947         int len, netid_len;
1948         __be32 *p;
1949
1950         switch (sap->sa_family) {
1951         case AF_INET:
1952                 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0)
1953                         return;
1954                 port = ntohs(((struct sockaddr_in *)sap)->sin_port);
1955                 netid = "tcp";
1956                 netid_len = 3;
1957                 break;
1958         case AF_INET6:
1959                 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0)
1960                         return;
1961                 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port);
1962                 netid = "tcp6";
1963                 netid_len = 4;
1964                 break;
1965         default:
1966                 /* we only support tcp and tcp6 */
1967                 WARN_ON_ONCE(1);
1968                 return;
1969         }
1970
1971         snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff);
1972         len = strlcat(addrbuf, portbuf, sizeof(addrbuf));
1973
1974         p = xdr_reserve_space(xdr, 4 + netid_len);
1975         xdr_encode_opaque(p, netid, netid_len);
1976
1977         p = xdr_reserve_space(xdr, 4 + len);
1978         xdr_encode_opaque(p, addrbuf, len);
1979 }
1980
1981 static void
1982 ff_layout_encode_nfstime(struct xdr_stream *xdr,
1983                          ktime_t t)
1984 {
1985         struct timespec64 ts;
1986         __be32 *p;
1987
1988         p = xdr_reserve_space(xdr, 12);
1989         ts = ktime_to_timespec64(t);
1990         p = xdr_encode_hyper(p, ts.tv_sec);
1991         *p++ = cpu_to_be32(ts.tv_nsec);
1992 }
1993
1994 static void
1995 ff_layout_encode_io_latency(struct xdr_stream *xdr,
1996                             struct nfs4_ff_io_stat *stat)
1997 {
1998         __be32 *p;
1999
2000         p = xdr_reserve_space(xdr, 5 * 8);
2001         p = xdr_encode_hyper(p, stat->ops_requested);
2002         p = xdr_encode_hyper(p, stat->bytes_requested);
2003         p = xdr_encode_hyper(p, stat->ops_completed);
2004         p = xdr_encode_hyper(p, stat->bytes_completed);
2005         p = xdr_encode_hyper(p, stat->bytes_not_delivered);
2006         ff_layout_encode_nfstime(xdr, stat->total_busy_time);
2007         ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time);
2008 }
2009
2010 static void
2011 ff_layout_encode_layoutstats(struct xdr_stream *xdr,
2012                              struct nfs42_layoutstat_args *args,
2013                              struct nfs42_layoutstat_devinfo *devinfo)
2014 {
2015         struct nfs4_ff_layout_mirror *mirror = devinfo->layout_private;
2016         struct nfs4_pnfs_ds_addr *da;
2017         struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds;
2018         struct nfs_fh *fh = &mirror->fh_versions[0];
2019         __be32 *p, *start;
2020
2021         da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node);
2022         dprintk("%s: DS %s: encoding address %s\n",
2023                 __func__, ds->ds_remotestr, da->da_remotestr);
2024         /* layoutupdate length */
2025         start = xdr_reserve_space(xdr, 4);
2026         /* netaddr4 */
2027         ff_layout_encode_netaddr(xdr, da);
2028         /* nfs_fh4 */
2029         p = xdr_reserve_space(xdr, 4 + fh->size);
2030         xdr_encode_opaque(p, fh->data, fh->size);
2031         /* ff_io_latency4 read */
2032         spin_lock(&mirror->lock);
2033         ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat);
2034         /* ff_io_latency4 write */
2035         ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat);
2036         spin_unlock(&mirror->lock);
2037         /* nfstime4 */
2038         ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time));
2039         /* bool */
2040         p = xdr_reserve_space(xdr, 4);
2041         *p = cpu_to_be32(false);
2042
2043         *start = cpu_to_be32((xdr->p - start - 1) * 4);
2044 }
2045
2046 static int
2047 ff_layout_mirror_prepare_stats(struct nfs42_layoutstat_args *args,
2048                                struct pnfs_layout_hdr *lo,
2049                                int dev_limit)
2050 {
2051         struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo);
2052         struct nfs4_ff_layout_mirror *mirror;
2053         struct nfs4_deviceid_node *dev;
2054         struct nfs42_layoutstat_devinfo *devinfo;
2055         int i = 0;
2056
2057         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2058                 if (i >= dev_limit)
2059                         break;
2060                 if (!mirror->mirror_ds)
2061                         continue;
2062                 /* mirror refcount put in cleanup_layoutstats */
2063                 if (!atomic_inc_not_zero(&mirror->ref))
2064                         continue;
2065                 dev = &mirror->mirror_ds->id_node;
2066                 devinfo = &args->devinfo[i];
2067                 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE);
2068                 devinfo->offset = 0;
2069                 devinfo->length = NFS4_MAX_UINT64;
2070                 devinfo->read_count = mirror->read_stat.io_stat.ops_completed;
2071                 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed;
2072                 devinfo->write_count = mirror->write_stat.io_stat.ops_completed;
2073                 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed;
2074                 devinfo->layout_type = LAYOUT_FLEX_FILES;
2075                 devinfo->layoutstats_encode = ff_layout_encode_layoutstats;
2076                 devinfo->layout_private = mirror;
2077
2078                 i++;
2079         }
2080         return i;
2081 }
2082
2083 static int
2084 ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args)
2085 {
2086         struct nfs4_flexfile_layout *ff_layout;
2087         struct nfs4_ff_layout_mirror *mirror;
2088         int dev_count = 0;
2089
2090         spin_lock(&args->inode->i_lock);
2091         ff_layout = FF_LAYOUT_FROM_HDR(NFS_I(args->inode)->layout);
2092         list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) {
2093                 if (atomic_read(&mirror->ref) != 0)
2094                         dev_count ++;
2095         }
2096         spin_unlock(&args->inode->i_lock);
2097         /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */
2098         if (dev_count > PNFS_LAYOUTSTATS_MAXDEV) {
2099                 dprintk("%s: truncating devinfo to limit (%d:%d)\n",
2100                         __func__, dev_count, PNFS_LAYOUTSTATS_MAXDEV);
2101                 dev_count = PNFS_LAYOUTSTATS_MAXDEV;
2102         }
2103         args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), GFP_NOIO);
2104         if (!args->devinfo)
2105                 return -ENOMEM;
2106
2107         spin_lock(&args->inode->i_lock);
2108         args->num_dev = ff_layout_mirror_prepare_stats(args,
2109                         &ff_layout->generic_hdr, dev_count);
2110         spin_unlock(&args->inode->i_lock);
2111
2112         return 0;
2113 }
2114
2115 static void
2116 ff_layout_cleanup_layoutstats(struct nfs42_layoutstat_data *data)
2117 {
2118         struct nfs4_ff_layout_mirror *mirror;
2119         int i;
2120
2121         for (i = 0; i < data->args.num_dev; i++) {
2122                 mirror = data->args.devinfo[i].layout_private;
2123                 data->args.devinfo[i].layout_private = NULL;
2124                 ff_layout_put_mirror(mirror);
2125         }
2126 }
2127
2128 static struct pnfs_layoutdriver_type flexfilelayout_type = {
2129         .id                     = LAYOUT_FLEX_FILES,
2130         .name                   = "LAYOUT_FLEX_FILES",
2131         .owner                  = THIS_MODULE,
2132         .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
2133         .free_layout_hdr        = ff_layout_free_layout_hdr,
2134         .alloc_lseg             = ff_layout_alloc_lseg,
2135         .free_lseg              = ff_layout_free_lseg,
2136         .add_lseg               = ff_layout_add_lseg,
2137         .pg_read_ops            = &ff_layout_pg_read_ops,
2138         .pg_write_ops           = &ff_layout_pg_write_ops,
2139         .get_ds_info            = ff_layout_get_ds_info,
2140         .free_deviceid_node     = ff_layout_free_deviceid_node,
2141         .mark_request_commit    = pnfs_layout_mark_request_commit,
2142         .clear_request_commit   = pnfs_generic_clear_request_commit,
2143         .scan_commit_lists      = pnfs_generic_scan_commit_lists,
2144         .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
2145         .commit_pagelist        = ff_layout_commit_pagelist,
2146         .read_pagelist          = ff_layout_read_pagelist,
2147         .write_pagelist         = ff_layout_write_pagelist,
2148         .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
2149         .encode_layoutreturn    = ff_layout_encode_layoutreturn,
2150         .sync                   = pnfs_nfs_generic_sync,
2151         .prepare_layoutstats    = ff_layout_prepare_layoutstats,
2152         .cleanup_layoutstats    = ff_layout_cleanup_layoutstats,
2153 };
2154
2155 static int __init nfs4flexfilelayout_init(void)
2156 {
2157         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
2158                __func__);
2159         return pnfs_register_layoutdriver(&flexfilelayout_type);
2160 }
2161
2162 static void __exit nfs4flexfilelayout_exit(void)
2163 {
2164         printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
2165                __func__);
2166         pnfs_unregister_layoutdriver(&flexfilelayout_type);
2167 }
2168
2169 MODULE_ALIAS("nfs-layouttype4-4");
2170
2171 MODULE_LICENSE("GPL");
2172 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
2173
2174 module_init(nfs4flexfilelayout_init);
2175 module_exit(nfs4flexfilelayout_exit);