1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Module for pnfs flexfile layout driver. 4 * 5 * Copyright (c) 2014, Primary Data, Inc. All rights reserved. 6 * 7 * Tao Peng <bergwolf@primarydata.com> 8 */ 9 10 #include <linux/nfs_fs.h> 11 #include <linux/nfs_mount.h> 12 #include <linux/nfs_page.h> 13 #include <linux/module.h> 14 #include <linux/file.h> 15 #include <linux/sched/mm.h> 16 17 #include <linux/sunrpc/metrics.h> 18 19 #include "flexfilelayout.h" 20 #include "../nfs4session.h" 21 #include "../nfs4idmap.h" 22 #include "../internal.h" 23 #include "../delegation.h" 24 #include "../nfs4trace.h" 25 #include "../iostat.h" 26 #include "../nfs.h" 27 #include "../nfs42.h" 28 29 #define NFSDBG_FACILITY NFSDBG_PNFS_LD 30 31 #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ) 32 #define FF_LAYOUTRETURN_MAXERR 20 33 34 enum nfs4_ff_op_type { 35 NFS4_FF_OP_LAYOUTSTATS, 36 NFS4_FF_OP_LAYOUTRETURN, 37 }; 38 39 static unsigned short io_maxretrans; 40 41 static const struct pnfs_commit_ops ff_layout_commit_ops; 42 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 43 struct nfs_pgio_header *hdr); 44 static int 45 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 46 struct nfs42_layoutstat_devinfo *devinfo, 47 int dev_limit, enum nfs4_ff_op_type type); 48 static void ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 49 const struct nfs42_layoutstat_devinfo *devinfo, 50 struct nfs4_ff_layout_ds_stripe *dss_info); 51 52 static struct pnfs_layout_hdr * 53 ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags) 54 { 55 struct nfs4_flexfile_layout *ffl; 56 57 ffl = kzalloc_obj(*ffl, gfp_flags); 58 if (ffl) { 59 pnfs_init_ds_commit_info(&ffl->commit_info); 60 INIT_LIST_HEAD(&ffl->error_list); 61 INIT_LIST_HEAD(&ffl->mirrors); 62 ffl->last_report_time = ktime_get(); 63 ffl->commit_info.ops = &ff_layout_commit_ops; 64 return &ffl->generic_hdr; 65 } else 66 return NULL; 67 } 68 69 static void 70 ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo) 71 { 72 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(lo); 73 struct nfs4_ff_layout_ds_err *err, *n; 74 75 list_for_each_entry_safe(err, n, &ffl->error_list, list) { 76 list_del(&err->list); 77 kfree(err); 78 } 79 kfree_rcu(ffl, generic_hdr.plh_rcu); 80 } 81 82 static int decode_pnfs_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid) 83 { 84 __be32 *p; 85 86 p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE); 87 if (unlikely(p == NULL)) 88 return -ENOBUFS; 89 stateid->type = NFS4_PNFS_DS_STATEID_TYPE; 90 memcpy(stateid->data, p, NFS4_STATEID_SIZE); 91 dprintk("%s: stateid id= [%x%x%x%x]\n", __func__, 92 p[0], p[1], p[2], p[3]); 93 return 0; 94 } 95 96 static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid) 97 { 98 __be32 *p; 99 100 p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE); 101 if (unlikely(!p)) 102 return -ENOBUFS; 103 memcpy(devid, p, NFS4_DEVICEID4_SIZE); 104 nfs4_print_deviceid(devid); 105 return 0; 106 } 107 108 static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh) 109 { 110 __be32 *p; 111 112 p = xdr_inline_decode(xdr, 4); 113 if (unlikely(!p)) 114 return -ENOBUFS; 115 fh->size = be32_to_cpup(p++); 116 if (fh->size > NFS_MAXFHSIZE) { 117 printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n", 118 fh->size); 119 return -EOVERFLOW; 120 } 121 /* fh.data */ 122 p = xdr_inline_decode(xdr, fh->size); 123 if (unlikely(!p)) 124 return -ENOBUFS; 125 memcpy(&fh->data, p, fh->size); 126 dprintk("%s: fh len %d\n", __func__, fh->size); 127 128 return 0; 129 } 130 131 /* 132 * Currently only stringified uids and gids are accepted. 133 * I.e., kerberos is not supported to the DSes, so no pricipals. 134 * 135 * That means that one common function will suffice, but when 136 * principals are added, this should be split to accomodate 137 * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid(). 138 */ 139 static int 140 decode_name(struct xdr_stream *xdr, u32 *id) 141 { 142 __be32 *p; 143 int len; 144 145 /* opaque_length(4)*/ 146 p = xdr_inline_decode(xdr, 4); 147 if (unlikely(!p)) 148 return -ENOBUFS; 149 len = be32_to_cpup(p++); 150 if (len < 0) 151 return -EINVAL; 152 153 dprintk("%s: len %u\n", __func__, len); 154 155 /* opaque body */ 156 p = xdr_inline_decode(xdr, len); 157 if (unlikely(!p)) 158 return -ENOBUFS; 159 160 if (!nfs_map_string_to_numeric((char *)p, len, id)) 161 return -EINVAL; 162 163 return 0; 164 } 165 166 static struct nfsd_file * 167 ff_local_open_fh(struct pnfs_layout_segment *lseg, u32 ds_idx, u32 dss_id, 168 struct nfs_client *clp, const struct cred *cred, 169 struct nfs_fh *fh, fmode_t mode) 170 { 171 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 172 struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx); 173 174 return nfs_local_open_fh(clp, cred, fh, &mirror->dss[dss_id].nfl, mode); 175 #else 176 return NULL; 177 #endif 178 } 179 180 static bool ff_dss_match_fh(const struct nfs4_ff_layout_ds_stripe *dss1, 181 const struct nfs4_ff_layout_ds_stripe *dss2) 182 { 183 int i, j; 184 185 if (dss1->fh_versions_cnt != dss2->fh_versions_cnt) 186 return false; 187 188 for (i = 0; i < dss1->fh_versions_cnt; i++) { 189 bool found_fh = false; 190 for (j = 0; j < dss2->fh_versions_cnt; j++) { 191 if (nfs_compare_fh(&dss1->fh_versions[i], 192 &dss2->fh_versions[j]) == 0) { 193 found_fh = true; 194 break; 195 } 196 } 197 if (!found_fh) 198 return false; 199 } 200 return true; 201 } 202 203 static bool ff_mirror_match_fh(const struct nfs4_ff_layout_mirror *m1, 204 const struct nfs4_ff_layout_mirror *m2) 205 { 206 u32 dss_id; 207 208 if (m1->dss_count != m2->dss_count) 209 return false; 210 211 for (dss_id = 0; dss_id < m1->dss_count; dss_id++) 212 if (!ff_dss_match_fh(&m1->dss[dss_id], &m2->dss[dss_id])) 213 return false; 214 215 return true; 216 } 217 218 static bool ff_mirror_match_devid(const struct nfs4_ff_layout_mirror *m1, 219 const struct nfs4_ff_layout_mirror *m2) 220 { 221 u32 dss_id; 222 223 if (m1->dss_count != m2->dss_count) 224 return false; 225 226 for (dss_id = 0; dss_id < m1->dss_count; dss_id++) 227 if (memcmp(&m1->dss[dss_id].devid, 228 &m2->dss[dss_id].devid, 229 sizeof(m1->dss[dss_id].devid)) != 0) 230 return false; 231 232 return true; 233 } 234 235 static struct nfs4_ff_layout_mirror * 236 ff_layout_add_mirror(struct pnfs_layout_hdr *lo, 237 struct nfs4_ff_layout_mirror *mirror) 238 { 239 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 240 struct nfs4_ff_layout_mirror *pos; 241 struct inode *inode = lo->plh_inode; 242 243 spin_lock(&inode->i_lock); 244 list_for_each_entry(pos, &ff_layout->mirrors, mirrors) { 245 if (!ff_mirror_match_devid(mirror, pos)) 246 continue; 247 if (!ff_mirror_match_fh(mirror, pos)) 248 continue; 249 if (refcount_inc_not_zero(&pos->ref)) { 250 spin_unlock(&inode->i_lock); 251 return pos; 252 } 253 } 254 list_add(&mirror->mirrors, &ff_layout->mirrors); 255 mirror->layout = lo; 256 spin_unlock(&inode->i_lock); 257 return mirror; 258 } 259 260 static void 261 ff_layout_remove_mirror(struct nfs4_ff_layout_mirror *mirror) 262 { 263 struct inode *inode; 264 if (mirror->layout == NULL) 265 return; 266 inode = mirror->layout->plh_inode; 267 spin_lock(&inode->i_lock); 268 list_del(&mirror->mirrors); 269 spin_unlock(&inode->i_lock); 270 mirror->layout = NULL; 271 } 272 273 static struct nfs4_ff_layout_mirror *ff_layout_alloc_mirror(u32 dss_count, 274 gfp_t gfp_flags) 275 { 276 struct nfs4_ff_layout_mirror *mirror; 277 278 mirror = kzalloc_obj(*mirror, gfp_flags); 279 if (mirror == NULL) 280 return NULL; 281 282 spin_lock_init(&mirror->lock); 283 refcount_set(&mirror->ref, 1); 284 INIT_LIST_HEAD(&mirror->mirrors); 285 286 mirror->dss_count = dss_count; 287 mirror->dss = 288 kzalloc_objs(struct nfs4_ff_layout_ds_stripe, dss_count, 289 gfp_flags); 290 if (mirror->dss == NULL) { 291 kfree(mirror); 292 return NULL; 293 } 294 295 for (u32 dss_id = 0; dss_id < mirror->dss_count; dss_id++) 296 nfs_localio_file_init(&mirror->dss[dss_id].nfl); 297 298 return mirror; 299 } 300 301 static void ff_layout_free_mirror(struct nfs4_ff_layout_mirror *mirror) 302 { 303 const struct cred *cred; 304 u32 dss_id; 305 306 ff_layout_remove_mirror(mirror); 307 308 for (dss_id = 0; dss_id < mirror->dss_count; dss_id++) { 309 kfree(mirror->dss[dss_id].fh_versions); 310 cred = rcu_access_pointer(mirror->dss[dss_id].ro_cred); 311 put_cred(cred); 312 cred = rcu_access_pointer(mirror->dss[dss_id].rw_cred); 313 put_cred(cred); 314 nfs_close_local_fh(&mirror->dss[dss_id].nfl); 315 nfs4_ff_layout_put_deviceid(mirror->dss[dss_id].mirror_ds); 316 } 317 318 kfree(mirror->dss); 319 kfree(mirror); 320 } 321 322 static void ff_layout_put_mirror(struct nfs4_ff_layout_mirror *mirror) 323 { 324 if (mirror != NULL && refcount_dec_and_test(&mirror->ref)) 325 ff_layout_free_mirror(mirror); 326 } 327 328 static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls) 329 { 330 u32 i; 331 332 for (i = 0; i < fls->mirror_array_cnt; i++) 333 ff_layout_put_mirror(fls->mirror_array[i]); 334 } 335 336 static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls) 337 { 338 if (fls) { 339 ff_layout_free_mirror_array(fls); 340 kfree(fls); 341 } 342 } 343 344 static bool 345 ff_lseg_match_mirrors(struct pnfs_layout_segment *l1, 346 struct pnfs_layout_segment *l2) 347 { 348 const struct nfs4_ff_layout_segment *fl1 = FF_LAYOUT_LSEG(l1); 349 const struct nfs4_ff_layout_segment *fl2 = FF_LAYOUT_LSEG(l2); 350 u32 i; 351 352 if (fl1->mirror_array_cnt != fl2->mirror_array_cnt) 353 return false; 354 for (i = 0; i < fl1->mirror_array_cnt; i++) { 355 if (fl1->mirror_array[i] != fl2->mirror_array[i]) 356 return false; 357 } 358 return true; 359 } 360 361 static bool 362 ff_lseg_range_is_after(const struct pnfs_layout_range *l1, 363 const struct pnfs_layout_range *l2) 364 { 365 u64 end1, end2; 366 367 if (l1->iomode != l2->iomode) 368 return l1->iomode != IOMODE_READ; 369 end1 = pnfs_calc_offset_end(l1->offset, l1->length); 370 end2 = pnfs_calc_offset_end(l2->offset, l2->length); 371 if (end1 < l2->offset) 372 return false; 373 if (end2 < l1->offset) 374 return true; 375 return l2->offset <= l1->offset; 376 } 377 378 static bool 379 ff_lseg_merge(struct pnfs_layout_segment *new, 380 struct pnfs_layout_segment *old) 381 { 382 u64 new_end, old_end; 383 384 if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags)) 385 return false; 386 if (new->pls_range.iomode != old->pls_range.iomode) 387 return false; 388 old_end = pnfs_calc_offset_end(old->pls_range.offset, 389 old->pls_range.length); 390 if (old_end < new->pls_range.offset) 391 return false; 392 new_end = pnfs_calc_offset_end(new->pls_range.offset, 393 new->pls_range.length); 394 if (new_end < old->pls_range.offset) 395 return false; 396 if (!ff_lseg_match_mirrors(new, old)) 397 return false; 398 399 /* Mergeable: copy info from 'old' to 'new' */ 400 if (new_end < old_end) 401 new_end = old_end; 402 if (new->pls_range.offset < old->pls_range.offset) 403 new->pls_range.offset = old->pls_range.offset; 404 new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset, 405 new_end); 406 if (test_bit(NFS_LSEG_ROC, &old->pls_flags)) 407 set_bit(NFS_LSEG_ROC, &new->pls_flags); 408 return true; 409 } 410 411 static void 412 ff_layout_add_lseg(struct pnfs_layout_hdr *lo, 413 struct pnfs_layout_segment *lseg, 414 struct list_head *free_me) 415 { 416 pnfs_generic_layout_insert_lseg(lo, lseg, 417 ff_lseg_range_is_after, 418 ff_lseg_merge, 419 free_me); 420 } 421 422 static u32 ff_mirror_efficiency_sum(const struct nfs4_ff_layout_mirror *mirror) 423 { 424 u32 dss_id, sum = 0; 425 426 for (dss_id = 0; dss_id < mirror->dss_count; dss_id++) 427 sum += mirror->dss[dss_id].efficiency; 428 429 return sum; 430 } 431 432 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls) 433 { 434 int i, j; 435 436 for (i = 0; i < fls->mirror_array_cnt - 1; i++) { 437 for (j = i + 1; j < fls->mirror_array_cnt; j++) 438 if (ff_mirror_efficiency_sum(fls->mirror_array[i]) < 439 ff_mirror_efficiency_sum(fls->mirror_array[j])) 440 swap(fls->mirror_array[i], 441 fls->mirror_array[j]); 442 } 443 } 444 445 static struct pnfs_layout_segment * 446 ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh, 447 struct nfs4_layoutget_res *lgr, 448 gfp_t gfp_flags) 449 { 450 struct pnfs_layout_segment *ret; 451 struct nfs4_ff_layout_segment *fls = NULL; 452 struct xdr_stream stream; 453 struct xdr_buf buf; 454 struct folio *scratch; 455 u64 stripe_unit; 456 u32 mirror_array_cnt; 457 __be32 *p; 458 int i, rc; 459 struct nfs4_ff_layout_ds_stripe *dss_info; 460 461 dprintk("--> %s\n", __func__); 462 scratch = folio_alloc(gfp_flags, 0); 463 if (!scratch) 464 return ERR_PTR(-ENOMEM); 465 466 xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages, 467 lgr->layoutp->len); 468 xdr_set_scratch_folio(&stream, scratch); 469 470 /* stripe unit and mirror_array_cnt */ 471 rc = -EIO; 472 p = xdr_inline_decode(&stream, 8 + 4); 473 if (!p) 474 goto out_err_free; 475 476 p = xdr_decode_hyper(p, &stripe_unit); 477 mirror_array_cnt = be32_to_cpup(p++); 478 dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__, 479 stripe_unit, mirror_array_cnt); 480 481 if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT || 482 mirror_array_cnt == 0) 483 goto out_err_free; 484 485 rc = -ENOMEM; 486 fls = kzalloc_flex(*fls, mirror_array, mirror_array_cnt, gfp_flags); 487 if (!fls) 488 goto out_err_free; 489 490 fls->mirror_array_cnt = mirror_array_cnt; 491 fls->stripe_unit = stripe_unit; 492 493 u32 dss_count = 0; 494 for (i = 0; i < fls->mirror_array_cnt; i++) { 495 struct nfs4_ff_layout_mirror *mirror; 496 struct cred *kcred; 497 const struct cred __rcu *cred; 498 kuid_t uid; 499 kgid_t gid; 500 u32 fh_count, id; 501 int j, dss_id; 502 503 rc = -EIO; 504 p = xdr_inline_decode(&stream, 4); 505 if (!p) 506 goto out_err_free; 507 508 // Ensure all mirrors have same stripe count. 509 if (dss_count == 0) 510 dss_count = be32_to_cpup(p); 511 else if (dss_count != be32_to_cpup(p)) 512 goto out_err_free; 513 514 if (dss_count > NFS4_FLEXFILE_LAYOUT_MAX_STRIPE_CNT || 515 dss_count == 0) 516 goto out_err_free; 517 518 if (dss_count > 1 && stripe_unit == 0) 519 goto out_err_free; 520 521 fls->mirror_array[i] = ff_layout_alloc_mirror(dss_count, gfp_flags); 522 if (fls->mirror_array[i] == NULL) { 523 rc = -ENOMEM; 524 goto out_err_free; 525 } 526 527 for (dss_id = 0; dss_id < dss_count; dss_id++) { 528 dss_info = &fls->mirror_array[i]->dss[dss_id]; 529 dss_info->mirror = fls->mirror_array[i]; 530 531 /* deviceid */ 532 rc = decode_deviceid(&stream, &dss_info->devid); 533 if (rc) 534 goto out_err_free; 535 536 /* efficiency */ 537 rc = -EIO; 538 p = xdr_inline_decode(&stream, 4); 539 if (!p) 540 goto out_err_free; 541 dss_info->efficiency = be32_to_cpup(p); 542 543 /* stateid */ 544 rc = decode_pnfs_stateid(&stream, &dss_info->stateid); 545 if (rc) 546 goto out_err_free; 547 548 /* fh */ 549 rc = -EIO; 550 p = xdr_inline_decode(&stream, 4); 551 if (!p) 552 goto out_err_free; 553 fh_count = be32_to_cpup(p); 554 555 dss_info->fh_versions = 556 kzalloc_objs(struct nfs_fh, fh_count, gfp_flags); 557 if (dss_info->fh_versions == NULL) { 558 rc = -ENOMEM; 559 goto out_err_free; 560 } 561 562 for (j = 0; j < fh_count; j++) { 563 rc = decode_nfs_fh(&stream, 564 &dss_info->fh_versions[j]); 565 if (rc) 566 goto out_err_free; 567 } 568 569 dss_info->fh_versions_cnt = fh_count; 570 571 /* user */ 572 rc = decode_name(&stream, &id); 573 if (rc) 574 goto out_err_free; 575 576 uid = make_kuid(&init_user_ns, id); 577 578 /* group */ 579 rc = decode_name(&stream, &id); 580 if (rc) 581 goto out_err_free; 582 583 gid = make_kgid(&init_user_ns, id); 584 585 if (gfp_flags & __GFP_FS) 586 kcred = prepare_kernel_cred(&init_task); 587 else { 588 unsigned int nofs_flags = memalloc_nofs_save(); 589 590 kcred = prepare_kernel_cred(&init_task); 591 memalloc_nofs_restore(nofs_flags); 592 } 593 rc = -ENOMEM; 594 if (!kcred) 595 goto out_err_free; 596 kcred->fsuid = uid; 597 kcred->fsgid = gid; 598 cred = RCU_INITIALIZER(kcred); 599 600 if (lgr->range.iomode == IOMODE_READ) 601 rcu_assign_pointer(dss_info->ro_cred, cred); 602 else 603 rcu_assign_pointer(dss_info->rw_cred, cred); 604 } 605 606 mirror = ff_layout_add_mirror(lh, fls->mirror_array[i]); 607 if (mirror != fls->mirror_array[i]) { 608 for (dss_id = 0; dss_id < dss_count; dss_id++) { 609 dss_info = &fls->mirror_array[i]->dss[dss_id]; 610 /* swap cred ptrs so free_mirror will clean up old */ 611 if (lgr->range.iomode == IOMODE_READ) { 612 cred = xchg(&mirror->dss[dss_id].ro_cred, 613 dss_info->ro_cred); 614 rcu_assign_pointer(dss_info->ro_cred, cred); 615 } else { 616 cred = xchg(&mirror->dss[dss_id].rw_cred, 617 dss_info->rw_cred); 618 rcu_assign_pointer(dss_info->rw_cred, cred); 619 } 620 } 621 ff_layout_free_mirror(fls->mirror_array[i]); 622 fls->mirror_array[i] = mirror; 623 } 624 625 dprintk("%s: iomode %s uid %u gid %u\n", __func__, 626 lgr->range.iomode == IOMODE_READ ? "READ" : "RW", 627 from_kuid(&init_user_ns, uid), 628 from_kgid(&init_user_ns, gid)); 629 } 630 631 p = xdr_inline_decode(&stream, 4); 632 if (!p) 633 goto out_sort_mirrors; 634 fls->flags = be32_to_cpup(p); 635 636 p = xdr_inline_decode(&stream, 4); 637 if (!p) 638 goto out_sort_mirrors; 639 for (i=0; i < fls->mirror_array_cnt; i++) 640 fls->mirror_array[i]->report_interval = be32_to_cpup(p); 641 642 out_sort_mirrors: 643 ff_layout_sort_mirrors(fls); 644 ret = &fls->generic_hdr; 645 dprintk("<-- %s (success)\n", __func__); 646 out_free_page: 647 folio_put(scratch); 648 return ret; 649 out_err_free: 650 _ff_layout_free_lseg(fls); 651 ret = ERR_PTR(rc); 652 dprintk("<-- %s (%d)\n", __func__, rc); 653 goto out_free_page; 654 } 655 656 static void 657 ff_layout_free_lseg(struct pnfs_layout_segment *lseg) 658 { 659 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 660 661 dprintk("--> %s\n", __func__); 662 663 if (lseg->pls_range.iomode == IOMODE_RW) { 664 struct nfs4_flexfile_layout *ffl; 665 struct inode *inode; 666 667 ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout); 668 inode = ffl->generic_hdr.plh_inode; 669 spin_lock(&inode->i_lock); 670 pnfs_generic_ds_cinfo_release_lseg(&ffl->commit_info, lseg); 671 spin_unlock(&inode->i_lock); 672 } 673 _ff_layout_free_lseg(fls); 674 } 675 676 static u32 calc_commit_idx(struct pnfs_layout_segment *lseg, 677 u32 mirror_idx, u32 dss_id) 678 { 679 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 680 681 return (mirror_idx * flseg->mirror_array[0]->dss_count) + dss_id; 682 } 683 684 static u32 calc_mirror_idx_from_commit(struct pnfs_layout_segment *lseg, 685 u32 commit_index) 686 { 687 return commit_index / FF_LAYOUT_LSEG(lseg)->mirror_array[0]->dss_count; 688 } 689 690 static u32 calc_dss_id_from_commit(struct pnfs_layout_segment *lseg, 691 u32 commit_index) 692 { 693 return commit_index % FF_LAYOUT_LSEG(lseg)->mirror_array[0]->dss_count; 694 } 695 696 static void 697 nfs4_ff_start_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 698 { 699 /* first IO request? */ 700 if (atomic_inc_return(&timer->n_ops) == 1) { 701 timer->start_time = now; 702 } 703 } 704 705 static ktime_t 706 nfs4_ff_end_busy_timer(struct nfs4_ff_busy_timer *timer, ktime_t now) 707 { 708 ktime_t start; 709 710 if (atomic_dec_return(&timer->n_ops) < 0) 711 WARN_ON_ONCE(1); 712 713 start = timer->start_time; 714 timer->start_time = now; 715 return ktime_sub(now, start); 716 } 717 718 static bool 719 nfs4_ff_layoutstat_start_io(struct nfs4_ff_layout_mirror *mirror, 720 u32 dss_id, 721 struct nfs4_ff_layoutstat *layoutstat, 722 ktime_t now) 723 { 724 s64 report_interval = FF_LAYOUTSTATS_REPORT_INTERVAL; 725 struct nfs4_flexfile_layout *ffl = FF_LAYOUT_FROM_HDR(mirror->layout); 726 727 nfs4_ff_start_busy_timer(&layoutstat->busy_timer, now); 728 if (!mirror->dss[dss_id].start_time) 729 mirror->dss[dss_id].start_time = now; 730 if (mirror->report_interval != 0) 731 report_interval = (s64)mirror->report_interval * 1000LL; 732 else if (layoutstats_timer != 0) 733 report_interval = (s64)layoutstats_timer * 1000LL; 734 if (ktime_to_ms(ktime_sub(now, ffl->last_report_time)) >= 735 report_interval) { 736 ffl->last_report_time = now; 737 return true; 738 } 739 740 return false; 741 } 742 743 static void 744 nfs4_ff_layout_stat_io_update_requested(struct nfs4_ff_layoutstat *layoutstat, 745 __u64 requested) 746 { 747 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 748 749 iostat->ops_requested++; 750 iostat->bytes_requested += requested; 751 } 752 753 static void 754 nfs4_ff_layout_stat_io_update_completed(struct nfs4_ff_layoutstat *layoutstat, 755 __u64 requested, 756 __u64 completed, 757 ktime_t time_completed, 758 ktime_t time_started) 759 { 760 struct nfs4_ff_io_stat *iostat = &layoutstat->io_stat; 761 ktime_t completion_time = ktime_sub(time_completed, time_started); 762 ktime_t timer; 763 764 iostat->ops_completed++; 765 iostat->bytes_completed += completed; 766 iostat->bytes_not_delivered += requested - completed; 767 768 timer = nfs4_ff_end_busy_timer(&layoutstat->busy_timer, time_completed); 769 iostat->total_busy_time = 770 ktime_add(iostat->total_busy_time, timer); 771 iostat->aggregate_completion_time = 772 ktime_add(iostat->aggregate_completion_time, 773 completion_time); 774 } 775 776 static void 777 nfs4_ff_layout_stat_io_start_read(struct inode *inode, 778 struct nfs4_ff_layout_mirror *mirror, 779 u32 dss_id, 780 __u64 requested, ktime_t now) 781 { 782 bool report; 783 784 spin_lock(&mirror->lock); 785 report = nfs4_ff_layoutstat_start_io( 786 mirror, dss_id, &mirror->dss[dss_id].read_stat, now); 787 nfs4_ff_layout_stat_io_update_requested( 788 &mirror->dss[dss_id].read_stat, requested); 789 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 790 spin_unlock(&mirror->lock); 791 792 if (report) 793 pnfs_report_layoutstat(inode, nfs_io_gfp_mask()); 794 } 795 796 static void 797 nfs4_ff_layout_stat_io_end_read(struct rpc_task *task, 798 struct nfs4_ff_layout_mirror *mirror, 799 u32 dss_id, 800 __u64 requested, 801 __u64 completed) 802 { 803 spin_lock(&mirror->lock); 804 nfs4_ff_layout_stat_io_update_completed(&mirror->dss[dss_id].read_stat, 805 requested, completed, 806 ktime_get(), task->tk_start); 807 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 808 spin_unlock(&mirror->lock); 809 } 810 811 static void 812 nfs4_ff_layout_stat_io_start_write(struct inode *inode, 813 struct nfs4_ff_layout_mirror *mirror, 814 u32 dss_id, 815 __u64 requested, ktime_t now) 816 { 817 bool report; 818 819 spin_lock(&mirror->lock); 820 report = nfs4_ff_layoutstat_start_io( 821 mirror, 822 dss_id, 823 &mirror->dss[dss_id].write_stat, 824 now); 825 nfs4_ff_layout_stat_io_update_requested( 826 &mirror->dss[dss_id].write_stat, 827 requested); 828 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 829 spin_unlock(&mirror->lock); 830 831 if (report) 832 pnfs_report_layoutstat(inode, nfs_io_gfp_mask()); 833 } 834 835 static void 836 nfs4_ff_layout_stat_io_end_write(struct rpc_task *task, 837 struct nfs4_ff_layout_mirror *mirror, 838 u32 dss_id, 839 __u64 requested, 840 __u64 completed, 841 enum nfs3_stable_how committed) 842 { 843 if (committed == NFS_UNSTABLE) 844 requested = completed = 0; 845 846 spin_lock(&mirror->lock); 847 nfs4_ff_layout_stat_io_update_completed(&mirror->dss[dss_id].write_stat, 848 requested, completed, ktime_get(), task->tk_start); 849 set_bit(NFS4_FF_MIRROR_STAT_AVAIL, &mirror->flags); 850 spin_unlock(&mirror->lock); 851 } 852 853 static void 854 ff_layout_mark_ds_unreachable(struct pnfs_layout_segment *lseg, u32 idx, u32 dss_id) 855 { 856 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx, dss_id); 857 858 if (devid) 859 nfs4_mark_deviceid_unavailable(devid); 860 } 861 862 static void 863 ff_layout_mark_ds_reachable(struct pnfs_layout_segment *lseg, u32 idx, u32 dss_id) 864 { 865 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx, dss_id); 866 867 if (devid) 868 nfs4_mark_deviceid_available(devid); 869 } 870 871 static struct nfs4_pnfs_ds * 872 ff_layout_choose_ds_for_read(struct pnfs_layout_segment *lseg, 873 u32 start_idx, u32 *best_idx, 874 u32 offset, u32 *dss_id, 875 bool check_device) 876 { 877 struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg); 878 struct nfs4_ff_layout_mirror *mirror; 879 struct nfs4_pnfs_ds *ds = ERR_PTR(-EAGAIN); 880 u32 idx; 881 882 /* mirrors are initially sorted by efficiency */ 883 for (idx = start_idx; idx < fls->mirror_array_cnt; idx++) { 884 mirror = FF_LAYOUT_COMP(lseg, idx); 885 *dss_id = nfs4_ff_layout_calc_dss_id( 886 fls->stripe_unit, 887 fls->mirror_array[idx]->dss_count, 888 offset); 889 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, *dss_id, false); 890 if (IS_ERR(ds)) 891 continue; 892 893 if (check_device && 894 nfs4_test_deviceid_unavailable(&mirror->dss[*dss_id].mirror_ds->id_node)) { 895 // reinitialize the error state in case if this is the last iteration 896 ds = ERR_PTR(-EINVAL); 897 continue; 898 } 899 900 *best_idx = idx; 901 break; 902 } 903 904 return ds; 905 } 906 907 static struct nfs4_pnfs_ds * 908 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg, 909 u32 start_idx, u32 *best_idx, 910 u32 offset, u32 *dss_id) 911 { 912 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, 913 offset, dss_id, false); 914 } 915 916 static struct nfs4_pnfs_ds * 917 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg, 918 u32 start_idx, u32 *best_idx, 919 u32 offset, u32 *dss_id) 920 { 921 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, 922 offset, dss_id, true); 923 } 924 925 static struct nfs4_pnfs_ds * 926 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, 927 u32 start_idx, u32 *best_idx, 928 u32 offset, u32 *dss_id) 929 { 930 struct nfs4_pnfs_ds *ds; 931 932 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx, 933 offset, dss_id); 934 if (!IS_ERR(ds)) 935 return ds; 936 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx, 937 offset, dss_id); 938 } 939 940 static struct nfs4_pnfs_ds * 941 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio, 942 u32 *best_idx, 943 u32 offset, 944 u32 *dss_id) 945 { 946 struct pnfs_layout_segment *lseg = pgio->pg_lseg; 947 struct nfs4_pnfs_ds *ds; 948 949 ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx, 950 best_idx, offset, dss_id); 951 if (!IS_ERR(ds) || !pgio->pg_mirror_idx) 952 return ds; 953 return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx, 954 offset, dss_id); 955 } 956 957 static void 958 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, 959 struct nfs_page *req, 960 bool strict_iomode) 961 { 962 pnfs_put_lseg(pgio->pg_lseg); 963 pgio->pg_lseg = 964 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 965 req_offset(req), req->wb_bytes, IOMODE_READ, 966 strict_iomode, nfs_io_gfp_mask()); 967 if (IS_ERR(pgio->pg_lseg)) { 968 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 969 pgio->pg_lseg = NULL; 970 } 971 } 972 973 static bool 974 ff_layout_lseg_is_striped(const struct nfs4_ff_layout_segment *fls) 975 { 976 return fls->mirror_array[0]->dss_count > 1; 977 } 978 979 /* 980 * ff_layout_pg_test(). Called by nfs_can_coalesce_requests() 981 * 982 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number 983 * of bytes (maximum @req->wb_bytes) that can be coalesced. 984 */ 985 static size_t 986 ff_layout_pg_test(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, 987 struct nfs_page *req) 988 { 989 unsigned int size; 990 u64 p_stripe, r_stripe; 991 u32 stripe_offset; 992 u64 segment_offset = pgio->pg_lseg->pls_range.offset; 993 u32 stripe_unit = FF_LAYOUT_LSEG(pgio->pg_lseg)->stripe_unit; 994 995 /* calls nfs_generic_pg_test */ 996 size = pnfs_generic_pg_test(pgio, prev, req); 997 if (!size) 998 return 0; 999 else if (!ff_layout_lseg_is_striped(FF_LAYOUT_LSEG(pgio->pg_lseg))) 1000 return size; 1001 1002 /* see if req and prev are in the same stripe */ 1003 if (prev) { 1004 p_stripe = (u64)req_offset(prev) - segment_offset; 1005 r_stripe = (u64)req_offset(req) - segment_offset; 1006 do_div(p_stripe, stripe_unit); 1007 do_div(r_stripe, stripe_unit); 1008 1009 if (p_stripe != r_stripe) 1010 return 0; 1011 } 1012 1013 /* calculate remaining bytes in the current stripe */ 1014 div_u64_rem((u64)req_offset(req) - segment_offset, 1015 stripe_unit, 1016 &stripe_offset); 1017 WARN_ON_ONCE(stripe_offset > stripe_unit); 1018 if (stripe_offset >= stripe_unit) 1019 return 0; 1020 return min(stripe_unit - (unsigned int)stripe_offset, size); 1021 } 1022 1023 static void 1024 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, 1025 struct nfs_page *req) 1026 { 1027 struct nfs_pgio_mirror *pgm; 1028 struct nfs4_ff_layout_mirror *mirror; 1029 struct nfs4_pnfs_ds *ds; 1030 u32 ds_idx, dss_id; 1031 1032 if (NFS_SERVER(pgio->pg_inode)->flags & 1033 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR)) 1034 pgio->pg_maxretrans = io_maxretrans; 1035 retry: 1036 pnfs_generic_pg_check_layout(pgio, req); 1037 /* Use full layout for now */ 1038 if (!pgio->pg_lseg) { 1039 ff_layout_pg_get_read(pgio, req, false); 1040 if (!pgio->pg_lseg) 1041 goto out_nolseg; 1042 } 1043 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { 1044 ff_layout_pg_get_read(pgio, req, true); 1045 if (!pgio->pg_lseg) 1046 goto out_nolseg; 1047 } 1048 /* Reset wb_nio, since getting layout segment was successful */ 1049 req->wb_nio = 0; 1050 1051 ds = ff_layout_get_ds_for_read(pgio, &ds_idx, 1052 req_offset(req), &dss_id); 1053 if (IS_ERR(ds)) { 1054 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 1055 goto out_mds; 1056 pnfs_generic_pg_cleanup(pgio); 1057 /* Sleep for 1 second before retrying */ 1058 ssleep(1); 1059 goto retry; 1060 } 1061 1062 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx); 1063 pgm = &pgio->pg_mirrors[0]; 1064 pgm->pg_bsize = mirror->dss[dss_id].mirror_ds->ds_versions[0].rsize; 1065 1066 pgio->pg_mirror_idx = ds_idx; 1067 return; 1068 out_nolseg: 1069 if (pgio->pg_error < 0) { 1070 if (pgio->pg_error != -EAGAIN) 1071 return; 1072 /* Retry getting layout segment if lower layer returned -EAGAIN */ 1073 if (pgio->pg_maxretrans && req->wb_nio++ > pgio->pg_maxretrans) { 1074 if (NFS_SERVER(pgio->pg_inode)->flags & NFS_MOUNT_SOFTERR) 1075 pgio->pg_error = -ETIMEDOUT; 1076 else 1077 pgio->pg_error = -EIO; 1078 return; 1079 } 1080 pgio->pg_error = 0; 1081 /* Sleep for 1 second before retrying */ 1082 ssleep(1); 1083 goto retry; 1084 } 1085 out_mds: 1086 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode, 1087 0, NFS4_MAX_UINT64, IOMODE_READ, 1088 NFS_I(pgio->pg_inode)->layout, 1089 pgio->pg_lseg); 1090 pgio->pg_maxretrans = 0; 1091 nfs_pageio_reset_read_mds(pgio); 1092 } 1093 1094 static void 1095 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 1096 struct nfs_page *req) 1097 { 1098 struct nfs4_ff_layout_mirror *mirror; 1099 struct nfs_pgio_mirror *pgm; 1100 struct nfs4_pnfs_ds *ds; 1101 u32 i, dss_id; 1102 1103 retry: 1104 pnfs_generic_pg_check_layout(pgio, req); 1105 if (!pgio->pg_lseg) { 1106 pgio->pg_lseg = 1107 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 1108 req_offset(req), req->wb_bytes, 1109 IOMODE_RW, false, nfs_io_gfp_mask()); 1110 if (IS_ERR(pgio->pg_lseg)) { 1111 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 1112 pgio->pg_lseg = NULL; 1113 return; 1114 } 1115 } 1116 /* If no lseg, fall back to write through mds */ 1117 if (pgio->pg_lseg == NULL) 1118 goto out_mds; 1119 1120 /* Use a direct mapping of ds_idx to pgio mirror_idx */ 1121 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)) 1122 goto out_eagain; 1123 1124 for (i = 0; i < pgio->pg_mirror_count; i++) { 1125 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); 1126 dss_id = nfs4_ff_layout_calc_dss_id( 1127 FF_LAYOUT_LSEG(pgio->pg_lseg)->stripe_unit, 1128 mirror->dss_count, 1129 req_offset(req)); 1130 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, 1131 dss_id, true); 1132 if (IS_ERR(ds)) { 1133 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 1134 goto out_mds; 1135 pnfs_generic_pg_cleanup(pgio); 1136 /* Sleep for 1 second before retrying */ 1137 ssleep(1); 1138 goto retry; 1139 } 1140 pgm = &pgio->pg_mirrors[i]; 1141 pgm->pg_bsize = mirror->dss[dss_id].mirror_ds->ds_versions[0].wsize; 1142 } 1143 1144 if (NFS_SERVER(pgio->pg_inode)->flags & 1145 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR)) 1146 pgio->pg_maxretrans = io_maxretrans; 1147 return; 1148 out_eagain: 1149 pnfs_generic_pg_cleanup(pgio); 1150 pgio->pg_error = -EAGAIN; 1151 return; 1152 out_mds: 1153 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode, 1154 0, NFS4_MAX_UINT64, IOMODE_RW, 1155 NFS_I(pgio->pg_inode)->layout, 1156 pgio->pg_lseg); 1157 pgio->pg_maxretrans = 0; 1158 nfs_pageio_reset_write_mds(pgio); 1159 pgio->pg_error = -EAGAIN; 1160 } 1161 1162 static unsigned int 1163 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 1164 struct nfs_page *req) 1165 { 1166 if (!pgio->pg_lseg) { 1167 pgio->pg_lseg = 1168 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 1169 req_offset(req), req->wb_bytes, 1170 IOMODE_RW, false, nfs_io_gfp_mask()); 1171 if (IS_ERR(pgio->pg_lseg)) { 1172 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 1173 pgio->pg_lseg = NULL; 1174 goto out; 1175 } 1176 } 1177 if (pgio->pg_lseg) 1178 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg); 1179 1180 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode, 1181 0, NFS4_MAX_UINT64, IOMODE_RW, 1182 NFS_I(pgio->pg_inode)->layout, 1183 pgio->pg_lseg); 1184 /* no lseg means that pnfs is not in use, so no mirroring here */ 1185 nfs_pageio_reset_write_mds(pgio); 1186 out: 1187 return 1; 1188 } 1189 1190 static u32 1191 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx) 1192 { 1193 u32 old = desc->pg_mirror_idx; 1194 1195 desc->pg_mirror_idx = idx; 1196 return old; 1197 } 1198 1199 static struct nfs_pgio_mirror * 1200 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx) 1201 { 1202 return &desc->pg_mirrors[idx]; 1203 } 1204 1205 static const struct nfs_pageio_ops ff_layout_pg_read_ops = { 1206 .pg_init = ff_layout_pg_init_read, 1207 .pg_test = ff_layout_pg_test, 1208 .pg_doio = pnfs_generic_pg_readpages, 1209 .pg_cleanup = pnfs_generic_pg_cleanup, 1210 }; 1211 1212 static const struct nfs_pageio_ops ff_layout_pg_write_ops = { 1213 .pg_init = ff_layout_pg_init_write, 1214 .pg_test = ff_layout_pg_test, 1215 .pg_doio = pnfs_generic_pg_writepages, 1216 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write, 1217 .pg_cleanup = pnfs_generic_pg_cleanup, 1218 .pg_get_mirror = ff_layout_pg_get_mirror_write, 1219 .pg_set_mirror = ff_layout_pg_set_mirror_write, 1220 }; 1221 1222 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) 1223 { 1224 struct rpc_task *task = &hdr->task; 1225 1226 pnfs_layoutcommit_inode(hdr->inode, false); 1227 1228 if (retry_pnfs) { 1229 dprintk("%s Reset task %5u for i/o through pNFS " 1230 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1231 hdr->task.tk_pid, 1232 hdr->inode->i_sb->s_id, 1233 (unsigned long long)NFS_FILEID(hdr->inode), 1234 hdr->args.count, 1235 (unsigned long long)hdr->args.offset); 1236 1237 hdr->completion_ops->reschedule_io(hdr); 1238 return; 1239 } 1240 1241 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1242 dprintk("%s Reset task %5u for i/o through MDS " 1243 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1244 hdr->task.tk_pid, 1245 hdr->inode->i_sb->s_id, 1246 (unsigned long long)NFS_FILEID(hdr->inode), 1247 hdr->args.count, 1248 (unsigned long long)hdr->args.offset); 1249 1250 trace_pnfs_mds_fallback_write_done(hdr->inode, 1251 hdr->args.offset, hdr->args.count, 1252 IOMODE_RW, NFS_I(hdr->inode)->layout, 1253 hdr->lseg); 1254 task->tk_status = pnfs_write_done_resend_to_mds(hdr); 1255 } 1256 } 1257 1258 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr) 1259 { 1260 u32 idx = hdr->pgio_mirror_idx + 1; 1261 u32 new_idx = 0; 1262 u32 dss_id = 0; 1263 struct nfs4_pnfs_ds *ds; 1264 1265 ds = ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx, 1266 hdr->args.offset, &dss_id); 1267 if (IS_ERR(ds)) 1268 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); 1269 else 1270 ff_layout_send_layouterror(hdr->lseg); 1271 pnfs_read_resend_pnfs(hdr, new_idx); 1272 } 1273 1274 static void ff_layout_reset_read(struct nfs_pgio_header *hdr) 1275 { 1276 struct rpc_task *task = &hdr->task; 1277 1278 pnfs_layoutcommit_inode(hdr->inode, false); 1279 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); 1280 1281 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1282 dprintk("%s Reset task %5u for i/o through MDS " 1283 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1284 hdr->task.tk_pid, 1285 hdr->inode->i_sb->s_id, 1286 (unsigned long long)NFS_FILEID(hdr->inode), 1287 hdr->args.count, 1288 (unsigned long long)hdr->args.offset); 1289 1290 trace_pnfs_mds_fallback_read_done(hdr->inode, 1291 hdr->args.offset, hdr->args.count, 1292 IOMODE_READ, NFS_I(hdr->inode)->layout, 1293 hdr->lseg); 1294 task->tk_status = pnfs_read_done_resend_to_mds(hdr); 1295 } 1296 } 1297 1298 static int ff_layout_async_handle_error_v4(struct rpc_task *task, 1299 u32 op_status, 1300 struct nfs4_state *state, 1301 struct nfs_client *clp, 1302 struct pnfs_layout_segment *lseg, 1303 u32 idx, u32 dss_id) 1304 { 1305 struct pnfs_layout_hdr *lo = lseg->pls_layout; 1306 struct inode *inode = lo->plh_inode; 1307 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx, dss_id); 1308 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; 1309 1310 switch (op_status) { 1311 case NFS4_OK: 1312 case NFS4ERR_NXIO: 1313 break; 1314 case NFSERR_PERM: 1315 if (!task->tk_xprt) 1316 break; 1317 xprt_force_disconnect(task->tk_xprt); 1318 goto out_retry; 1319 case NFS4ERR_BADSESSION: 1320 case NFS4ERR_BADSLOT: 1321 case NFS4ERR_BAD_HIGH_SLOT: 1322 case NFS4ERR_DEADSESSION: 1323 case NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1324 case NFS4ERR_SEQ_FALSE_RETRY: 1325 case NFS4ERR_SEQ_MISORDERED: 1326 dprintk("%s ERROR %d, Reset session. Exchangeid " 1327 "flags 0x%x\n", __func__, task->tk_status, 1328 clp->cl_exchange_flags); 1329 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); 1330 goto out_retry; 1331 case NFS4ERR_DELAY: 1332 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1333 fallthrough; 1334 case NFS4ERR_GRACE: 1335 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX); 1336 goto out_retry; 1337 case NFS4ERR_RETRY_UNCACHED_REP: 1338 goto out_retry; 1339 /* Invalidate Layout errors */ 1340 case NFS4ERR_PNFS_NO_LAYOUT: 1341 case NFS4ERR_STALE: 1342 case NFS4ERR_BADHANDLE: 1343 case NFS4ERR_ISDIR: 1344 case NFS4ERR_FHEXPIRED: 1345 case NFS4ERR_WRONG_TYPE: 1346 dprintk("%s Invalid layout error %d\n", __func__, 1347 task->tk_status); 1348 /* 1349 * Destroy layout so new i/o will get a new layout. 1350 * Layout will not be destroyed until all current lseg 1351 * references are put. Mark layout as invalid to resend failed 1352 * i/o and all i/o waiting on the slot table to the MDS until 1353 * layout is destroyed and a new valid layout is obtained. 1354 */ 1355 pnfs_destroy_layout(NFS_I(inode)); 1356 rpc_wake_up(&tbl->slot_tbl_waitq); 1357 goto reset; 1358 default: 1359 break; 1360 } 1361 1362 switch (task->tk_status) { 1363 /* RPC connection errors */ 1364 case -ENETDOWN: 1365 case -ENETUNREACH: 1366 if (test_bit(NFS_CS_NETUNREACH_FATAL, &clp->cl_flags)) 1367 return -NFS4ERR_FATAL_IOERROR; 1368 fallthrough; 1369 case -ECONNREFUSED: 1370 case -EHOSTDOWN: 1371 case -EHOSTUNREACH: 1372 case -EIO: 1373 case -ETIMEDOUT: 1374 case -EPIPE: 1375 case -EPROTO: 1376 case -ENODEV: 1377 dprintk("%s DS connection error %d\n", __func__, 1378 task->tk_status); 1379 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1380 &devid->deviceid); 1381 rpc_wake_up(&tbl->slot_tbl_waitq); 1382 break; 1383 default: 1384 break; 1385 } 1386 1387 if (ff_layout_avoid_mds_available_ds(lseg)) 1388 return -NFS4ERR_RESET_TO_PNFS; 1389 reset: 1390 dprintk("%s Retry through MDS. Error %d\n", __func__, 1391 task->tk_status); 1392 return -NFS4ERR_RESET_TO_MDS; 1393 1394 out_retry: 1395 task->tk_status = 0; 1396 return -EAGAIN; 1397 } 1398 1399 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */ 1400 static int ff_layout_async_handle_error_v3(struct rpc_task *task, 1401 u32 op_status, 1402 struct nfs_client *clp, 1403 struct pnfs_layout_segment *lseg, 1404 u32 idx, u32 dss_id) 1405 { 1406 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx, dss_id); 1407 1408 switch (op_status) { 1409 case NFS_OK: 1410 case NFSERR_NXIO: 1411 break; 1412 case NFSERR_PERM: 1413 if (!task->tk_xprt) 1414 break; 1415 xprt_force_disconnect(task->tk_xprt); 1416 goto out_retry; 1417 case NFSERR_ACCES: 1418 case NFSERR_BADHANDLE: 1419 case NFSERR_FBIG: 1420 case NFSERR_IO: 1421 case NFSERR_NOSPC: 1422 case NFSERR_ROFS: 1423 case NFSERR_STALE: 1424 goto out_reset_to_pnfs; 1425 case NFSERR_JUKEBOX: 1426 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1427 goto out_retry; 1428 default: 1429 break; 1430 } 1431 1432 switch (task->tk_status) { 1433 /* File access problems. Don't mark the device as unavailable */ 1434 case -EACCES: 1435 case -ESTALE: 1436 case -EISDIR: 1437 case -EBADHANDLE: 1438 case -ELOOP: 1439 case -ENOSPC: 1440 break; 1441 case -EJUKEBOX: 1442 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1443 goto out_retry; 1444 case -ENETDOWN: 1445 case -ENETUNREACH: 1446 if (test_bit(NFS_CS_NETUNREACH_FATAL, &clp->cl_flags)) 1447 return -NFS4ERR_FATAL_IOERROR; 1448 fallthrough; 1449 default: 1450 dprintk("%s DS connection error %d\n", __func__, 1451 task->tk_status); 1452 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1453 &devid->deviceid); 1454 } 1455 out_reset_to_pnfs: 1456 /* FIXME: Need to prevent infinite looping here. */ 1457 return -NFS4ERR_RESET_TO_PNFS; 1458 out_retry: 1459 task->tk_status = 0; 1460 rpc_restart_call_prepare(task); 1461 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME); 1462 return -EAGAIN; 1463 } 1464 1465 static int ff_layout_async_handle_error(struct rpc_task *task, 1466 u32 op_status, 1467 struct nfs4_state *state, 1468 struct nfs_client *clp, 1469 struct pnfs_layout_segment *lseg, 1470 u32 idx, u32 dss_id) 1471 { 1472 int vers = clp->cl_nfs_mod->rpc_vers->number; 1473 1474 if (task->tk_status >= 0) { 1475 ff_layout_mark_ds_reachable(lseg, idx, dss_id); 1476 return 0; 1477 } 1478 1479 /* Handle the case of an invalid layout segment */ 1480 if (!pnfs_is_valid_lseg(lseg)) 1481 return -NFS4ERR_RESET_TO_PNFS; 1482 1483 switch (vers) { 1484 case 3: 1485 return ff_layout_async_handle_error_v3(task, op_status, clp, 1486 lseg, idx, dss_id); 1487 case 4: 1488 return ff_layout_async_handle_error_v4(task, op_status, state, 1489 clp, lseg, idx, dss_id); 1490 default: 1491 /* should never happen */ 1492 WARN_ON_ONCE(1); 1493 return 0; 1494 } 1495 } 1496 1497 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, 1498 u32 idx, u32 dss_id, u64 offset, u64 length, 1499 u32 *op_status, int opnum, int error) 1500 { 1501 struct nfs4_ff_layout_mirror *mirror; 1502 u32 status = *op_status; 1503 int err; 1504 1505 if (status == 0) { 1506 switch (error) { 1507 case -ETIMEDOUT: 1508 case -EPFNOSUPPORT: 1509 case -EPROTONOSUPPORT: 1510 case -EOPNOTSUPP: 1511 case -EINVAL: 1512 case -ECONNREFUSED: 1513 case -ECONNRESET: 1514 case -EHOSTDOWN: 1515 case -EHOSTUNREACH: 1516 case -ENETDOWN: 1517 case -ENETUNREACH: 1518 case -EADDRINUSE: 1519 case -ENOBUFS: 1520 case -EPIPE: 1521 case -EPERM: 1522 case -EPROTO: 1523 case -ENODEV: 1524 *op_status = status = NFS4ERR_NXIO; 1525 break; 1526 case -EACCES: 1527 *op_status = status = NFS4ERR_ACCESS; 1528 break; 1529 default: 1530 return; 1531 } 1532 } 1533 1534 mirror = FF_LAYOUT_COMP(lseg, idx); 1535 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 1536 mirror, dss_id, offset, length, status, opnum, 1537 nfs_io_gfp_mask()); 1538 1539 switch (status) { 1540 case NFS4ERR_DELAY: 1541 case NFS4ERR_GRACE: 1542 case NFS4ERR_PERM: 1543 break; 1544 case NFS4ERR_NXIO: 1545 ff_layout_mark_ds_unreachable(lseg, idx, dss_id); 1546 /* 1547 * Don't return the layout if this is a read and we still 1548 * have layouts to try 1549 */ 1550 if (opnum == OP_READ) 1551 break; 1552 fallthrough; 1553 default: 1554 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, 1555 lseg); 1556 } 1557 1558 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status); 1559 } 1560 1561 /* NFS_PROTO call done callback routines */ 1562 static int ff_layout_read_done_cb(struct rpc_task *task, 1563 struct nfs_pgio_header *hdr) 1564 { 1565 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(hdr->lseg); 1566 u32 dss_id = nfs4_ff_layout_calc_dss_id( 1567 flseg->stripe_unit, 1568 flseg->mirror_array[hdr->pgio_mirror_idx]->dss_count, 1569 hdr->args.offset); 1570 int err; 1571 1572 if (task->tk_status < 0) { 1573 ff_layout_io_track_ds_error(hdr->lseg, 1574 hdr->pgio_mirror_idx, dss_id, 1575 hdr->args.offset, hdr->args.count, 1576 &hdr->res.op_status, OP_READ, 1577 task->tk_status); 1578 trace_ff_layout_read_error(hdr, task->tk_status); 1579 } 1580 1581 err = ff_layout_async_handle_error(task, hdr->res.op_status, 1582 hdr->args.context->state, 1583 hdr->ds_clp, hdr->lseg, 1584 hdr->pgio_mirror_idx, 1585 dss_id); 1586 1587 trace_nfs4_pnfs_read(hdr, err); 1588 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1589 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1590 switch (err) { 1591 case -NFS4ERR_RESET_TO_PNFS: 1592 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1593 return task->tk_status; 1594 case -NFS4ERR_RESET_TO_MDS: 1595 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1596 return task->tk_status; 1597 case -EAGAIN: 1598 goto out_eagain; 1599 case -NFS4ERR_FATAL_IOERROR: 1600 task->tk_status = -EIO; 1601 return 0; 1602 } 1603 1604 return 0; 1605 out_eagain: 1606 rpc_restart_call_prepare(task); 1607 return -EAGAIN; 1608 } 1609 1610 static bool 1611 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg) 1612 { 1613 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT); 1614 } 1615 1616 /* 1617 * We reference the rpc_cred of the first WRITE that triggers the need for 1618 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. 1619 * rfc5661 is not clear about which credential should be used. 1620 * 1621 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so 1622 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 1623 * we always send layoutcommit after DS writes. 1624 */ 1625 static void 1626 ff_layout_set_layoutcommit(struct inode *inode, 1627 struct pnfs_layout_segment *lseg, 1628 loff_t end_offset) 1629 { 1630 if (!ff_layout_need_layoutcommit(lseg)) 1631 return; 1632 1633 pnfs_set_layoutcommit(inode, lseg, end_offset); 1634 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino, 1635 (unsigned long long) NFS_I(inode)->layout->plh_lwb); 1636 } 1637 1638 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task, 1639 struct nfs_pgio_header *hdr) 1640 { 1641 struct nfs4_ff_layout_mirror *mirror; 1642 u32 dss_id; 1643 1644 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1645 return; 1646 1647 mirror = FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx); 1648 dss_id = nfs4_ff_layout_calc_dss_id( 1649 FF_LAYOUT_LSEG(hdr->lseg)->stripe_unit, 1650 mirror->dss_count, 1651 hdr->args.offset); 1652 1653 nfs4_ff_layout_stat_io_start_read( 1654 hdr->inode, 1655 mirror, 1656 dss_id, 1657 hdr->args.count, 1658 task->tk_start); 1659 } 1660 1661 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 1662 struct nfs_pgio_header *hdr) 1663 { 1664 struct nfs4_ff_layout_mirror *mirror; 1665 u32 dss_id; 1666 1667 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1668 return; 1669 1670 mirror = FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx); 1671 dss_id = nfs4_ff_layout_calc_dss_id( 1672 FF_LAYOUT_LSEG(hdr->lseg)->stripe_unit, 1673 mirror->dss_count, 1674 hdr->args.offset); 1675 1676 nfs4_ff_layout_stat_io_end_read( 1677 task, 1678 mirror, 1679 dss_id, 1680 hdr->args.count, 1681 hdr->res.count); 1682 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1683 } 1684 1685 static int ff_layout_read_prepare_common(struct rpc_task *task, 1686 struct nfs_pgio_header *hdr) 1687 { 1688 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1689 rpc_exit(task, -EIO); 1690 return -EIO; 1691 } 1692 1693 if (!pnfs_is_valid_lseg(hdr->lseg)) { 1694 rpc_exit(task, -EAGAIN); 1695 return -EAGAIN; 1696 } 1697 1698 ff_layout_read_record_layoutstats_start(task, hdr); 1699 return 0; 1700 } 1701 1702 /* 1703 * Call ops for the async read/write cases 1704 * In the case of dense layouts, the offset needs to be reset to its 1705 * original value. 1706 */ 1707 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data) 1708 { 1709 struct nfs_pgio_header *hdr = data; 1710 1711 if (ff_layout_read_prepare_common(task, hdr)) 1712 return; 1713 1714 rpc_call_start(task); 1715 } 1716 1717 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data) 1718 { 1719 struct nfs_pgio_header *hdr = data; 1720 1721 if (nfs4_setup_sequence(hdr->ds_clp, 1722 &hdr->args.seq_args, 1723 &hdr->res.seq_res, 1724 task)) 1725 return; 1726 1727 ff_layout_read_prepare_common(task, hdr); 1728 } 1729 1730 static void ff_layout_read_call_done(struct rpc_task *task, void *data) 1731 { 1732 struct nfs_pgio_header *hdr = data; 1733 1734 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1735 task->tk_status == 0) { 1736 nfs4_sequence_done(task, &hdr->res.seq_res); 1737 return; 1738 } 1739 1740 /* Note this may cause RPC to be resent */ 1741 hdr->mds_ops->rpc_call_done(task, hdr); 1742 } 1743 1744 static void ff_layout_read_count_stats(struct rpc_task *task, void *data) 1745 { 1746 struct nfs_pgio_header *hdr = data; 1747 1748 ff_layout_read_record_layoutstats_done(task, hdr); 1749 rpc_count_iostats_metrics(task, 1750 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]); 1751 } 1752 1753 static void ff_layout_read_release(void *data) 1754 { 1755 struct nfs_pgio_header *hdr = data; 1756 1757 ff_layout_read_record_layoutstats_done(&hdr->task, hdr); 1758 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) 1759 ff_layout_resend_pnfs_read(hdr); 1760 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1761 ff_layout_reset_read(hdr); 1762 pnfs_generic_rw_release(data); 1763 } 1764 1765 1766 static int ff_layout_write_done_cb(struct rpc_task *task, 1767 struct nfs_pgio_header *hdr) 1768 { 1769 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(hdr->lseg); 1770 u32 dss_id = nfs4_ff_layout_calc_dss_id( 1771 flseg->stripe_unit, 1772 flseg->mirror_array[hdr->pgio_mirror_idx]->dss_count, 1773 hdr->args.offset); 1774 loff_t end_offs = 0; 1775 int err; 1776 1777 if (task->tk_status < 0) { 1778 ff_layout_io_track_ds_error(hdr->lseg, 1779 hdr->pgio_mirror_idx, dss_id, 1780 hdr->args.offset, hdr->args.count, 1781 &hdr->res.op_status, OP_WRITE, 1782 task->tk_status); 1783 trace_ff_layout_write_error(hdr, task->tk_status); 1784 } 1785 1786 err = ff_layout_async_handle_error(task, hdr->res.op_status, 1787 hdr->args.context->state, 1788 hdr->ds_clp, hdr->lseg, 1789 hdr->pgio_mirror_idx, 1790 dss_id); 1791 1792 trace_nfs4_pnfs_write(hdr, err); 1793 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1794 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1795 switch (err) { 1796 case -NFS4ERR_RESET_TO_PNFS: 1797 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1798 return task->tk_status; 1799 case -NFS4ERR_RESET_TO_MDS: 1800 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1801 return task->tk_status; 1802 case -EAGAIN: 1803 return -EAGAIN; 1804 case -NFS4ERR_FATAL_IOERROR: 1805 task->tk_status = -EIO; 1806 return 0; 1807 } 1808 1809 if (hdr->res.verf->committed == NFS_FILE_SYNC || 1810 hdr->res.verf->committed == NFS_DATA_SYNC) 1811 end_offs = hdr->mds_offset + (loff_t)hdr->res.count; 1812 1813 /* Note: if the write is unstable, don't set end_offs until commit */ 1814 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); 1815 1816 /* zero out fattr since we don't care DS attr at all */ 1817 hdr->fattr.valid = 0; 1818 if (task->tk_status >= 0) 1819 nfs_writeback_update_inode(hdr); 1820 1821 return 0; 1822 } 1823 1824 static int ff_layout_commit_done_cb(struct rpc_task *task, 1825 struct nfs_commit_data *data) 1826 { 1827 int err; 1828 u32 idx = calc_mirror_idx_from_commit(data->lseg, data->ds_commit_index); 1829 u32 dss_id = calc_dss_id_from_commit(data->lseg, data->ds_commit_index); 1830 1831 if (task->tk_status < 0) { 1832 ff_layout_io_track_ds_error(data->lseg, idx, dss_id, 1833 data->args.offset, data->args.count, 1834 &data->res.op_status, OP_COMMIT, 1835 task->tk_status); 1836 trace_ff_layout_commit_error(data, task->tk_status); 1837 } 1838 1839 err = ff_layout_async_handle_error(task, data->res.op_status, 1840 NULL, data->ds_clp, data->lseg, idx, 1841 dss_id); 1842 1843 trace_nfs4_pnfs_commit_ds(data, err); 1844 switch (err) { 1845 case -NFS4ERR_RESET_TO_PNFS: 1846 pnfs_generic_prepare_to_resend_writes(data); 1847 return -EAGAIN; 1848 case -NFS4ERR_RESET_TO_MDS: 1849 pnfs_generic_prepare_to_resend_writes(data); 1850 return -EAGAIN; 1851 case -EAGAIN: 1852 rpc_restart_call_prepare(task); 1853 return -EAGAIN; 1854 case -NFS4ERR_FATAL_IOERROR: 1855 task->tk_status = -EIO; 1856 return 0; 1857 } 1858 1859 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb); 1860 return 0; 1861 } 1862 1863 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task, 1864 struct nfs_pgio_header *hdr) 1865 { 1866 struct nfs4_ff_layout_mirror *mirror; 1867 u32 dss_id; 1868 1869 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1870 return; 1871 1872 mirror = FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx); 1873 dss_id = nfs4_ff_layout_calc_dss_id( 1874 FF_LAYOUT_LSEG(hdr->lseg)->stripe_unit, 1875 mirror->dss_count, 1876 hdr->args.offset); 1877 1878 nfs4_ff_layout_stat_io_start_write( 1879 hdr->inode, 1880 mirror, 1881 dss_id, 1882 hdr->args.count, 1883 task->tk_start); 1884 } 1885 1886 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task, 1887 struct nfs_pgio_header *hdr) 1888 { 1889 struct nfs4_ff_layout_mirror *mirror; 1890 u32 dss_id; 1891 1892 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1893 return; 1894 1895 mirror = FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx); 1896 dss_id = nfs4_ff_layout_calc_dss_id( 1897 FF_LAYOUT_LSEG(hdr->lseg)->stripe_unit, 1898 mirror->dss_count, 1899 hdr->args.offset); 1900 1901 nfs4_ff_layout_stat_io_end_write( 1902 task, 1903 mirror, 1904 dss_id, 1905 hdr->args.count, 1906 hdr->res.count, 1907 hdr->res.verf->committed); 1908 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1909 } 1910 1911 static int ff_layout_write_prepare_common(struct rpc_task *task, 1912 struct nfs_pgio_header *hdr) 1913 { 1914 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1915 rpc_exit(task, -EIO); 1916 return -EIO; 1917 } 1918 1919 if (!pnfs_is_valid_lseg(hdr->lseg)) { 1920 rpc_exit(task, -EAGAIN); 1921 return -EAGAIN; 1922 } 1923 1924 ff_layout_write_record_layoutstats_start(task, hdr); 1925 return 0; 1926 } 1927 1928 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data) 1929 { 1930 struct nfs_pgio_header *hdr = data; 1931 1932 if (ff_layout_write_prepare_common(task, hdr)) 1933 return; 1934 1935 rpc_call_start(task); 1936 } 1937 1938 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data) 1939 { 1940 struct nfs_pgio_header *hdr = data; 1941 1942 if (nfs4_setup_sequence(hdr->ds_clp, 1943 &hdr->args.seq_args, 1944 &hdr->res.seq_res, 1945 task)) 1946 return; 1947 1948 ff_layout_write_prepare_common(task, hdr); 1949 } 1950 1951 static void ff_layout_write_call_done(struct rpc_task *task, void *data) 1952 { 1953 struct nfs_pgio_header *hdr = data; 1954 1955 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1956 task->tk_status == 0) { 1957 nfs4_sequence_done(task, &hdr->res.seq_res); 1958 return; 1959 } 1960 1961 /* Note this may cause RPC to be resent */ 1962 hdr->mds_ops->rpc_call_done(task, hdr); 1963 } 1964 1965 static void ff_layout_write_count_stats(struct rpc_task *task, void *data) 1966 { 1967 struct nfs_pgio_header *hdr = data; 1968 1969 ff_layout_write_record_layoutstats_done(task, hdr); 1970 rpc_count_iostats_metrics(task, 1971 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]); 1972 } 1973 1974 static void ff_layout_write_release(void *data) 1975 { 1976 struct nfs_pgio_header *hdr = data; 1977 1978 ff_layout_write_record_layoutstats_done(&hdr->task, hdr); 1979 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) { 1980 ff_layout_send_layouterror(hdr->lseg); 1981 ff_layout_reset_write(hdr, true); 1982 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1983 ff_layout_reset_write(hdr, false); 1984 pnfs_generic_rw_release(data); 1985 } 1986 1987 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task, 1988 struct nfs_commit_data *cdata) 1989 { 1990 u32 idx, dss_id; 1991 1992 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags)) 1993 return; 1994 1995 idx = calc_mirror_idx_from_commit(cdata->lseg, cdata->ds_commit_index); 1996 dss_id = calc_dss_id_from_commit(cdata->lseg, cdata->ds_commit_index); 1997 nfs4_ff_layout_stat_io_start_write(cdata->inode, 1998 FF_LAYOUT_COMP(cdata->lseg, idx), 1999 dss_id, 2000 0, task->tk_start); 2001 } 2002 2003 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task, 2004 struct nfs_commit_data *cdata) 2005 { 2006 struct nfs_page *req; 2007 __u64 count = 0; 2008 u32 idx, dss_id; 2009 2010 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags)) 2011 return; 2012 2013 if (task->tk_status == 0) { 2014 list_for_each_entry(req, &cdata->pages, wb_list) 2015 count += req->wb_bytes; 2016 } 2017 2018 idx = calc_mirror_idx_from_commit(cdata->lseg, cdata->ds_commit_index); 2019 dss_id = calc_dss_id_from_commit(cdata->lseg, cdata->ds_commit_index); 2020 nfs4_ff_layout_stat_io_end_write(task, 2021 FF_LAYOUT_COMP(cdata->lseg, idx), 2022 dss_id, 2023 count, count, NFS_FILE_SYNC); 2024 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags); 2025 } 2026 2027 static int ff_layout_commit_prepare_common(struct rpc_task *task, 2028 struct nfs_commit_data *cdata) 2029 { 2030 if (!pnfs_is_valid_lseg(cdata->lseg)) { 2031 rpc_exit(task, -EAGAIN); 2032 return -EAGAIN; 2033 } 2034 2035 ff_layout_commit_record_layoutstats_start(task, cdata); 2036 return 0; 2037 } 2038 2039 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data) 2040 { 2041 if (ff_layout_commit_prepare_common(task, data)) 2042 return; 2043 2044 rpc_call_start(task); 2045 } 2046 2047 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data) 2048 { 2049 struct nfs_commit_data *wdata = data; 2050 2051 if (nfs4_setup_sequence(wdata->ds_clp, 2052 &wdata->args.seq_args, 2053 &wdata->res.seq_res, 2054 task)) 2055 return; 2056 ff_layout_commit_prepare_common(task, data); 2057 } 2058 2059 static void ff_layout_commit_done(struct rpc_task *task, void *data) 2060 { 2061 pnfs_generic_write_commit_done(task, data); 2062 } 2063 2064 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data) 2065 { 2066 struct nfs_commit_data *cdata = data; 2067 2068 ff_layout_commit_record_layoutstats_done(task, cdata); 2069 rpc_count_iostats_metrics(task, 2070 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]); 2071 } 2072 2073 static void ff_layout_commit_release(void *data) 2074 { 2075 struct nfs_commit_data *cdata = data; 2076 2077 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata); 2078 pnfs_generic_commit_release(data); 2079 } 2080 2081 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = { 2082 .rpc_call_prepare = ff_layout_read_prepare_v3, 2083 .rpc_call_done = ff_layout_read_call_done, 2084 .rpc_count_stats = ff_layout_read_count_stats, 2085 .rpc_release = ff_layout_read_release, 2086 }; 2087 2088 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = { 2089 .rpc_call_prepare = ff_layout_read_prepare_v4, 2090 .rpc_call_done = ff_layout_read_call_done, 2091 .rpc_count_stats = ff_layout_read_count_stats, 2092 .rpc_release = ff_layout_read_release, 2093 }; 2094 2095 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = { 2096 .rpc_call_prepare = ff_layout_write_prepare_v3, 2097 .rpc_call_done = ff_layout_write_call_done, 2098 .rpc_count_stats = ff_layout_write_count_stats, 2099 .rpc_release = ff_layout_write_release, 2100 }; 2101 2102 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = { 2103 .rpc_call_prepare = ff_layout_write_prepare_v4, 2104 .rpc_call_done = ff_layout_write_call_done, 2105 .rpc_count_stats = ff_layout_write_count_stats, 2106 .rpc_release = ff_layout_write_release, 2107 }; 2108 2109 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = { 2110 .rpc_call_prepare = ff_layout_commit_prepare_v3, 2111 .rpc_call_done = ff_layout_commit_done, 2112 .rpc_count_stats = ff_layout_commit_count_stats, 2113 .rpc_release = ff_layout_commit_release, 2114 }; 2115 2116 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = { 2117 .rpc_call_prepare = ff_layout_commit_prepare_v4, 2118 .rpc_call_done = ff_layout_commit_done, 2119 .rpc_count_stats = ff_layout_commit_count_stats, 2120 .rpc_release = ff_layout_commit_release, 2121 }; 2122 2123 static enum pnfs_try_status 2124 ff_layout_read_pagelist(struct nfs_pgio_header *hdr) 2125 { 2126 struct pnfs_layout_segment *lseg = hdr->lseg; 2127 struct nfs4_pnfs_ds *ds; 2128 struct rpc_clnt *ds_clnt; 2129 struct nfsd_file *localio; 2130 struct nfs4_ff_layout_mirror *mirror; 2131 const struct cred *ds_cred; 2132 loff_t offset = hdr->args.offset; 2133 u32 idx = hdr->pgio_mirror_idx; 2134 int vers; 2135 struct nfs_fh *fh; 2136 u32 dss_id; 2137 bool ds_fatal_error = false; 2138 2139 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n", 2140 __func__, hdr->inode->i_ino, 2141 hdr->args.pgbase, (size_t)hdr->args.count, offset); 2142 2143 mirror = FF_LAYOUT_COMP(lseg, idx); 2144 dss_id = nfs4_ff_layout_calc_dss_id( 2145 FF_LAYOUT_LSEG(lseg)->stripe_unit, 2146 mirror->dss_count, 2147 offset); 2148 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, dss_id, false); 2149 if (IS_ERR(ds)) { 2150 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds)); 2151 goto out_failed; 2152 } 2153 2154 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 2155 hdr->inode, dss_id); 2156 if (IS_ERR(ds_clnt)) 2157 goto out_failed; 2158 2159 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred, dss_id); 2160 if (!ds_cred) 2161 goto out_failed; 2162 2163 vers = nfs4_ff_layout_ds_version(mirror, dss_id); 2164 2165 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__, 2166 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers); 2167 2168 hdr->pgio_done_cb = ff_layout_read_done_cb; 2169 refcount_inc(&ds->ds_clp->cl_count); 2170 hdr->ds_clp = ds->ds_clp; 2171 fh = nfs4_ff_layout_select_ds_fh(mirror, dss_id); 2172 if (fh) 2173 hdr->args.fh = fh; 2174 2175 nfs4_ff_layout_select_ds_stateid(mirror, dss_id, &hdr->args.stateid); 2176 2177 /* 2178 * Note that if we ever decide to split across DSes, 2179 * then we may need to handle dense-like offsets. 2180 */ 2181 hdr->args.offset = offset; 2182 hdr->mds_offset = offset; 2183 2184 /* Start IO accounting for local read */ 2185 localio = ff_local_open_fh(lseg, idx, dss_id, ds->ds_clp, ds_cred, fh, 2186 FMODE_READ); 2187 if (localio) { 2188 hdr->task.tk_start = ktime_get(); 2189 ff_layout_read_record_layoutstats_start(&hdr->task, hdr); 2190 } 2191 2192 /* Perform an asynchronous read to ds */ 2193 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 2194 vers == 3 ? &ff_layout_read_call_ops_v3 : 2195 &ff_layout_read_call_ops_v4, 2196 0, RPC_TASK_SOFTCONN, localio); 2197 put_cred(ds_cred); 2198 return PNFS_ATTEMPTED; 2199 2200 out_failed: 2201 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error) 2202 return PNFS_TRY_AGAIN; 2203 trace_pnfs_mds_fallback_read_pagelist(hdr->inode, 2204 hdr->args.offset, hdr->args.count, 2205 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg); 2206 return PNFS_NOT_ATTEMPTED; 2207 } 2208 2209 /* Perform async writes. */ 2210 static enum pnfs_try_status 2211 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync) 2212 { 2213 struct pnfs_layout_segment *lseg = hdr->lseg; 2214 struct nfs4_pnfs_ds *ds; 2215 struct rpc_clnt *ds_clnt; 2216 struct nfsd_file *localio; 2217 struct nfs4_ff_layout_mirror *mirror; 2218 const struct cred *ds_cred; 2219 loff_t offset = hdr->args.offset; 2220 int vers; 2221 struct nfs_fh *fh; 2222 u32 idx = hdr->pgio_mirror_idx; 2223 u32 dss_id; 2224 bool ds_fatal_error = false; 2225 2226 mirror = FF_LAYOUT_COMP(lseg, idx); 2227 dss_id = nfs4_ff_layout_calc_dss_id( 2228 FF_LAYOUT_LSEG(lseg)->stripe_unit, 2229 mirror->dss_count, 2230 offset); 2231 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, dss_id, true); 2232 if (IS_ERR(ds)) { 2233 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds)); 2234 goto out_failed; 2235 } 2236 2237 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 2238 hdr->inode, dss_id); 2239 if (IS_ERR(ds_clnt)) 2240 goto out_failed; 2241 2242 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred, dss_id); 2243 if (!ds_cred) 2244 goto out_failed; 2245 2246 vers = nfs4_ff_layout_ds_version(mirror, dss_id); 2247 2248 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n", 2249 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, 2250 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), 2251 vers); 2252 2253 hdr->pgio_done_cb = ff_layout_write_done_cb; 2254 refcount_inc(&ds->ds_clp->cl_count); 2255 hdr->ds_clp = ds->ds_clp; 2256 hdr->ds_commit_idx = calc_commit_idx(lseg, idx, dss_id); 2257 fh = nfs4_ff_layout_select_ds_fh(mirror, dss_id); 2258 if (fh) 2259 hdr->args.fh = fh; 2260 2261 nfs4_ff_layout_select_ds_stateid(mirror, dss_id, &hdr->args.stateid); 2262 2263 /* 2264 * Note that if we ever decide to split across DSes, 2265 * then we may need to handle dense-like offsets. 2266 */ 2267 hdr->args.offset = offset; 2268 2269 /* Start IO accounting for local write */ 2270 localio = ff_local_open_fh(lseg, idx, dss_id, ds->ds_clp, ds_cred, fh, 2271 FMODE_READ|FMODE_WRITE); 2272 if (localio) { 2273 hdr->task.tk_start = ktime_get(); 2274 ff_layout_write_record_layoutstats_start(&hdr->task, hdr); 2275 } 2276 2277 /* Perform an asynchronous write */ 2278 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 2279 vers == 3 ? &ff_layout_write_call_ops_v3 : 2280 &ff_layout_write_call_ops_v4, 2281 sync, RPC_TASK_SOFTCONN, localio); 2282 put_cred(ds_cred); 2283 return PNFS_ATTEMPTED; 2284 2285 out_failed: 2286 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error) 2287 return PNFS_TRY_AGAIN; 2288 trace_pnfs_mds_fallback_write_pagelist(hdr->inode, 2289 hdr->args.offset, hdr->args.count, 2290 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg); 2291 return PNFS_NOT_ATTEMPTED; 2292 } 2293 2294 static struct nfs_fh * 2295 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i, u32 dss_id) 2296 { 2297 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2298 2299 /* FIXME: Assume that there is only one NFS version available 2300 * for the DS. 2301 */ 2302 return &flseg->mirror_array[i]->dss[dss_id].fh_versions[0]; 2303 } 2304 2305 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how) 2306 { 2307 struct pnfs_layout_segment *lseg = data->lseg; 2308 struct nfs4_pnfs_ds *ds; 2309 struct rpc_clnt *ds_clnt; 2310 struct nfsd_file *localio; 2311 struct nfs4_ff_layout_mirror *mirror; 2312 const struct cred *ds_cred; 2313 u32 idx, dss_id; 2314 int vers, ret; 2315 struct nfs_fh *fh; 2316 2317 if (!lseg || !(pnfs_is_valid_lseg(lseg) || 2318 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))) 2319 goto out_err; 2320 2321 idx = calc_mirror_idx_from_commit(lseg, data->ds_commit_index); 2322 mirror = FF_LAYOUT_COMP(lseg, idx); 2323 dss_id = calc_dss_id_from_commit(lseg, data->ds_commit_index); 2324 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, dss_id, true); 2325 if (IS_ERR(ds)) 2326 goto out_err; 2327 2328 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 2329 data->inode, dss_id); 2330 if (IS_ERR(ds_clnt)) 2331 goto out_err; 2332 2333 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred, dss_id); 2334 if (!ds_cred) 2335 goto out_err; 2336 2337 vers = nfs4_ff_layout_ds_version(mirror, dss_id); 2338 2339 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__, 2340 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count), 2341 vers); 2342 data->commit_done_cb = ff_layout_commit_done_cb; 2343 data->cred = ds_cred; 2344 refcount_inc(&ds->ds_clp->cl_count); 2345 data->ds_clp = ds->ds_clp; 2346 fh = select_ds_fh_from_commit(lseg, idx, dss_id); 2347 if (fh) 2348 data->args.fh = fh; 2349 2350 /* Start IO accounting for local commit */ 2351 localio = ff_local_open_fh(lseg, idx, dss_id, ds->ds_clp, ds_cred, fh, 2352 FMODE_READ|FMODE_WRITE); 2353 if (localio) { 2354 data->task.tk_start = ktime_get(); 2355 ff_layout_commit_record_layoutstats_start(&data->task, data); 2356 } 2357 2358 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops, 2359 vers == 3 ? &ff_layout_commit_call_ops_v3 : 2360 &ff_layout_commit_call_ops_v4, 2361 how, RPC_TASK_SOFTCONN, localio); 2362 put_cred(ds_cred); 2363 return ret; 2364 out_err: 2365 pnfs_generic_prepare_to_resend_writes(data); 2366 pnfs_generic_commit_release(data); 2367 return -EAGAIN; 2368 } 2369 2370 static int 2371 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 2372 int how, struct nfs_commit_info *cinfo) 2373 { 2374 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, 2375 ff_layout_initiate_commit); 2376 } 2377 2378 static bool ff_layout_match_rw(const struct rpc_task *task, 2379 const struct nfs_pgio_header *hdr, 2380 const struct pnfs_layout_segment *lseg) 2381 { 2382 return hdr->lseg == lseg; 2383 } 2384 2385 static bool ff_layout_match_commit(const struct rpc_task *task, 2386 const struct nfs_commit_data *cdata, 2387 const struct pnfs_layout_segment *lseg) 2388 { 2389 return cdata->lseg == lseg; 2390 } 2391 2392 static bool ff_layout_match_io(const struct rpc_task *task, const void *data) 2393 { 2394 const struct rpc_call_ops *ops = task->tk_ops; 2395 2396 if (ops == &ff_layout_read_call_ops_v3 || 2397 ops == &ff_layout_read_call_ops_v4 || 2398 ops == &ff_layout_write_call_ops_v3 || 2399 ops == &ff_layout_write_call_ops_v4) 2400 return ff_layout_match_rw(task, task->tk_calldata, data); 2401 if (ops == &ff_layout_commit_call_ops_v3 || 2402 ops == &ff_layout_commit_call_ops_v4) 2403 return ff_layout_match_commit(task, task->tk_calldata, data); 2404 return false; 2405 } 2406 2407 static void ff_layout_cancel_io(struct pnfs_layout_segment *lseg) 2408 { 2409 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2410 struct nfs4_ff_layout_mirror *mirror; 2411 struct nfs4_ff_layout_ds *mirror_ds; 2412 struct nfs4_pnfs_ds *ds; 2413 struct nfs_client *ds_clp; 2414 struct rpc_clnt *clnt; 2415 u32 idx, dss_id; 2416 2417 for (idx = 0; idx < flseg->mirror_array_cnt; idx++) { 2418 mirror = flseg->mirror_array[idx]; 2419 for (dss_id = 0; dss_id < mirror->dss_count; dss_id++) { 2420 mirror_ds = mirror->dss[dss_id].mirror_ds; 2421 if (IS_ERR_OR_NULL(mirror_ds)) 2422 continue; 2423 ds = mirror->dss[dss_id].mirror_ds->ds; 2424 if (!ds) 2425 continue; 2426 ds_clp = ds->ds_clp; 2427 if (!ds_clp) 2428 continue; 2429 clnt = ds_clp->cl_rpcclient; 2430 if (!clnt) 2431 continue; 2432 if (!rpc_cancel_tasks(clnt, -EAGAIN, 2433 ff_layout_match_io, lseg)) 2434 continue; 2435 rpc_clnt_disconnect(clnt); 2436 } 2437 } 2438 } 2439 2440 static struct pnfs_ds_commit_info * 2441 ff_layout_get_ds_info(struct inode *inode) 2442 { 2443 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; 2444 2445 if (layout == NULL) 2446 return NULL; 2447 2448 return &FF_LAYOUT_FROM_HDR(layout)->commit_info; 2449 } 2450 2451 static void 2452 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo, 2453 struct pnfs_layout_segment *lseg) 2454 { 2455 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2456 struct inode *inode = lseg->pls_layout->plh_inode; 2457 struct pnfs_commit_array *array, *new; 2458 u32 size = flseg->mirror_array_cnt * flseg->mirror_array[0]->dss_count; 2459 2460 new = pnfs_alloc_commit_array(size, 2461 nfs_io_gfp_mask()); 2462 if (new) { 2463 spin_lock(&inode->i_lock); 2464 array = pnfs_add_commit_array(fl_cinfo, new, lseg); 2465 spin_unlock(&inode->i_lock); 2466 if (array != new) 2467 pnfs_free_commit_array(new); 2468 } 2469 } 2470 2471 static void 2472 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo, 2473 struct inode *inode) 2474 { 2475 spin_lock(&inode->i_lock); 2476 pnfs_generic_ds_cinfo_destroy(fl_cinfo); 2477 spin_unlock(&inode->i_lock); 2478 } 2479 2480 static void 2481 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d) 2482 { 2483 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds, 2484 id_node)); 2485 } 2486 2487 static int ff_layout_encode_ioerr(struct xdr_stream *xdr, 2488 const struct nfs4_layoutreturn_args *args, 2489 const struct nfs4_flexfile_layoutreturn_args *ff_args) 2490 { 2491 __be32 *start; 2492 2493 start = xdr_reserve_space(xdr, 4); 2494 if (unlikely(!start)) 2495 return -E2BIG; 2496 2497 *start = cpu_to_be32(ff_args->num_errors); 2498 /* This assume we always return _ALL_ layouts */ 2499 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors); 2500 } 2501 2502 static void 2503 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, 2504 const nfs4_stateid *stateid, 2505 const struct nfs42_layoutstat_devinfo *devinfo) 2506 { 2507 __be32 *p; 2508 2509 p = xdr_reserve_space(xdr, 8 + 8); 2510 p = xdr_encode_hyper(p, devinfo->offset); 2511 p = xdr_encode_hyper(p, devinfo->length); 2512 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE); 2513 p = xdr_reserve_space(xdr, 4*8); 2514 p = xdr_encode_hyper(p, devinfo->read_count); 2515 p = xdr_encode_hyper(p, devinfo->read_bytes); 2516 p = xdr_encode_hyper(p, devinfo->write_count); 2517 p = xdr_encode_hyper(p, devinfo->write_bytes); 2518 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE); 2519 } 2520 2521 static void 2522 ff_layout_encode_ff_iostat(struct xdr_stream *xdr, 2523 const nfs4_stateid *stateid, 2524 const struct nfs42_layoutstat_devinfo *devinfo) 2525 { 2526 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo); 2527 ff_layout_encode_ff_layoutupdate(xdr, devinfo, 2528 devinfo->ld_private.data); 2529 } 2530 2531 /* report nothing for now */ 2532 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr, 2533 const struct nfs4_layoutreturn_args *args, 2534 struct nfs4_flexfile_layoutreturn_args *ff_args) 2535 { 2536 __be32 *p; 2537 int i; 2538 2539 p = xdr_reserve_space(xdr, 4); 2540 *p = cpu_to_be32(ff_args->num_dev); 2541 for (i = 0; i < ff_args->num_dev; i++) 2542 ff_layout_encode_ff_iostat(xdr, 2543 &args->layout->plh_stateid, 2544 &ff_args->devinfo[i]); 2545 } 2546 2547 static void 2548 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, 2549 unsigned int num_entries) 2550 { 2551 unsigned int i; 2552 2553 for (i = 0; i < num_entries; i++) { 2554 if (!devinfo[i].ld_private.ops) 2555 continue; 2556 if (!devinfo[i].ld_private.ops->free) 2557 continue; 2558 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private); 2559 } 2560 } 2561 2562 static struct nfs4_deviceid_node * 2563 ff_layout_alloc_deviceid_node(struct nfs_server *server, 2564 struct pnfs_device *pdev, gfp_t gfp_flags) 2565 { 2566 struct nfs4_ff_layout_ds *dsaddr; 2567 2568 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags); 2569 if (!dsaddr) 2570 return NULL; 2571 return &dsaddr->id_node; 2572 } 2573 2574 static void 2575 ff_layout_encode_layoutreturn(struct xdr_stream *xdr, 2576 const void *voidargs, 2577 const struct nfs4_xdr_opaque_data *ff_opaque) 2578 { 2579 const struct nfs4_layoutreturn_args *args = voidargs; 2580 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data; 2581 struct xdr_buf tmp_buf = { 2582 .head = { 2583 [0] = { 2584 .iov_base = page_address(ff_args->pages[0]), 2585 }, 2586 }, 2587 .buflen = PAGE_SIZE, 2588 }; 2589 struct xdr_stream tmp_xdr; 2590 __be32 *start; 2591 2592 dprintk("%s: Begin\n", __func__); 2593 2594 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL); 2595 2596 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args); 2597 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args); 2598 2599 start = xdr_reserve_space(xdr, 4); 2600 *start = cpu_to_be32(tmp_buf.len); 2601 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len); 2602 2603 dprintk("%s: Return\n", __func__); 2604 } 2605 2606 static void 2607 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args) 2608 { 2609 struct nfs4_flexfile_layoutreturn_args *ff_args; 2610 2611 if (!args->data) 2612 return; 2613 ff_args = args->data; 2614 args->data = NULL; 2615 2616 ff_layout_free_ds_ioerr(&ff_args->errors); 2617 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev); 2618 2619 put_page(ff_args->pages[0]); 2620 kfree(ff_args); 2621 } 2622 2623 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = { 2624 .encode = ff_layout_encode_layoutreturn, 2625 .free = ff_layout_free_layoutreturn, 2626 }; 2627 2628 static int 2629 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args) 2630 { 2631 struct nfs4_flexfile_layoutreturn_args *ff_args; 2632 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout); 2633 2634 ff_args = kmalloc_obj(*ff_args, nfs_io_gfp_mask()); 2635 if (!ff_args) 2636 goto out_nomem; 2637 ff_args->pages[0] = alloc_page(nfs_io_gfp_mask()); 2638 if (!ff_args->pages[0]) 2639 goto out_nomem_free; 2640 2641 INIT_LIST_HEAD(&ff_args->errors); 2642 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout, 2643 &args->range, &ff_args->errors, 2644 FF_LAYOUTRETURN_MAXERR); 2645 2646 spin_lock(&args->inode->i_lock); 2647 ff_args->num_dev = ff_layout_mirror_prepare_stats( 2648 &ff_layout->generic_hdr, &ff_args->devinfo[0], 2649 ARRAY_SIZE(ff_args->devinfo), NFS4_FF_OP_LAYOUTRETURN); 2650 spin_unlock(&args->inode->i_lock); 2651 2652 args->ld_private->ops = &layoutreturn_ops; 2653 args->ld_private->data = ff_args; 2654 return 0; 2655 out_nomem_free: 2656 kfree(ff_args); 2657 out_nomem: 2658 return -ENOMEM; 2659 } 2660 2661 #ifdef CONFIG_NFS_V4_2 2662 void 2663 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg) 2664 { 2665 struct pnfs_layout_hdr *lo = lseg->pls_layout; 2666 struct nfs42_layout_error *errors; 2667 LIST_HEAD(head); 2668 2669 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR)) 2670 return; 2671 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1); 2672 if (list_empty(&head)) 2673 return; 2674 2675 errors = kmalloc_objs(*errors, NFS42_LAYOUTERROR_MAX, nfs_io_gfp_mask()); 2676 if (errors != NULL) { 2677 const struct nfs4_ff_layout_ds_err *pos; 2678 size_t n = 0; 2679 2680 list_for_each_entry(pos, &head, list) { 2681 errors[n].offset = pos->offset; 2682 errors[n].length = pos->length; 2683 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid); 2684 errors[n].errors[0].dev_id = pos->deviceid; 2685 errors[n].errors[0].status = pos->status; 2686 errors[n].errors[0].opnum = pos->opnum; 2687 n++; 2688 if (!list_is_last(&pos->list, &head) && 2689 n < NFS42_LAYOUTERROR_MAX) 2690 continue; 2691 if (nfs42_proc_layouterror(lseg, errors, n) < 0) 2692 break; 2693 n = 0; 2694 } 2695 kfree(errors); 2696 } 2697 ff_layout_free_ds_ioerr(&head); 2698 } 2699 #else 2700 void 2701 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg) 2702 { 2703 } 2704 #endif 2705 2706 static int 2707 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen) 2708 { 2709 const struct sockaddr_in *sin = (struct sockaddr_in *)sap; 2710 2711 return snprintf(buf, buflen, "%pI4", &sin->sin_addr); 2712 } 2713 2714 static size_t 2715 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, 2716 const int buflen) 2717 { 2718 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; 2719 const struct in6_addr *addr = &sin6->sin6_addr; 2720 2721 /* 2722 * RFC 4291, Section 2.2.2 2723 * 2724 * Shorthanded ANY address 2725 */ 2726 if (ipv6_addr_any(addr)) 2727 return snprintf(buf, buflen, "::"); 2728 2729 /* 2730 * RFC 4291, Section 2.2.2 2731 * 2732 * Shorthanded loopback address 2733 */ 2734 if (ipv6_addr_loopback(addr)) 2735 return snprintf(buf, buflen, "::1"); 2736 2737 /* 2738 * RFC 4291, Section 2.2.3 2739 * 2740 * Special presentation address format for mapped v4 2741 * addresses. 2742 */ 2743 if (ipv6_addr_v4mapped(addr)) 2744 return snprintf(buf, buflen, "::ffff:%pI4", 2745 &addr->s6_addr32[3]); 2746 2747 /* 2748 * RFC 4291, Section 2.2.1 2749 */ 2750 return snprintf(buf, buflen, "%pI6c", addr); 2751 } 2752 2753 /* Derived from rpc_sockaddr2uaddr */ 2754 static void 2755 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da) 2756 { 2757 struct sockaddr *sap = (struct sockaddr *)&da->da_addr; 2758 char portbuf[RPCBIND_MAXUADDRPLEN]; 2759 char addrbuf[RPCBIND_MAXUADDRLEN]; 2760 unsigned short port; 2761 int len, netid_len; 2762 __be32 *p; 2763 2764 switch (sap->sa_family) { 2765 case AF_INET: 2766 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0) 2767 return; 2768 port = ntohs(((struct sockaddr_in *)sap)->sin_port); 2769 break; 2770 case AF_INET6: 2771 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0) 2772 return; 2773 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port); 2774 break; 2775 default: 2776 WARN_ON_ONCE(1); 2777 return; 2778 } 2779 2780 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff); 2781 len = strlcat(addrbuf, portbuf, sizeof(addrbuf)); 2782 2783 netid_len = strlen(da->da_netid); 2784 p = xdr_reserve_space(xdr, 4 + netid_len); 2785 xdr_encode_opaque(p, da->da_netid, netid_len); 2786 2787 p = xdr_reserve_space(xdr, 4 + len); 2788 xdr_encode_opaque(p, addrbuf, len); 2789 } 2790 2791 static void 2792 ff_layout_encode_nfstime(struct xdr_stream *xdr, 2793 ktime_t t) 2794 { 2795 struct timespec64 ts; 2796 __be32 *p; 2797 2798 p = xdr_reserve_space(xdr, 12); 2799 ts = ktime_to_timespec64(t); 2800 p = xdr_encode_hyper(p, ts.tv_sec); 2801 *p++ = cpu_to_be32(ts.tv_nsec); 2802 } 2803 2804 static void 2805 ff_layout_encode_io_latency(struct xdr_stream *xdr, 2806 struct nfs4_ff_io_stat *stat) 2807 { 2808 __be32 *p; 2809 2810 p = xdr_reserve_space(xdr, 5 * 8); 2811 p = xdr_encode_hyper(p, stat->ops_requested); 2812 p = xdr_encode_hyper(p, stat->bytes_requested); 2813 p = xdr_encode_hyper(p, stat->ops_completed); 2814 p = xdr_encode_hyper(p, stat->bytes_completed); 2815 p = xdr_encode_hyper(p, stat->bytes_not_delivered); 2816 ff_layout_encode_nfstime(xdr, stat->total_busy_time); 2817 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time); 2818 } 2819 2820 static void 2821 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 2822 const struct nfs42_layoutstat_devinfo *devinfo, 2823 struct nfs4_ff_layout_ds_stripe *dss_info) 2824 { 2825 struct nfs4_pnfs_ds_addr *da; 2826 struct nfs4_pnfs_ds *ds = dss_info->mirror_ds->ds; 2827 struct nfs_fh *fh = &dss_info->fh_versions[0]; 2828 __be32 *p; 2829 2830 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node); 2831 dprintk("%s: DS %s: encoding address %s\n", 2832 __func__, ds->ds_remotestr, da->da_remotestr); 2833 /* netaddr4 */ 2834 ff_layout_encode_netaddr(xdr, da); 2835 /* nfs_fh4 */ 2836 p = xdr_reserve_space(xdr, 4 + fh->size); 2837 xdr_encode_opaque(p, fh->data, fh->size); 2838 /* ff_io_latency4 read */ 2839 spin_lock(&dss_info->mirror->lock); 2840 ff_layout_encode_io_latency(xdr, 2841 &dss_info->read_stat.io_stat); 2842 /* ff_io_latency4 write */ 2843 ff_layout_encode_io_latency(xdr, 2844 &dss_info->write_stat.io_stat); 2845 spin_unlock(&dss_info->mirror->lock); 2846 /* nfstime4 */ 2847 ff_layout_encode_nfstime(xdr, 2848 ktime_sub(ktime_get(), 2849 dss_info->start_time)); 2850 /* bool */ 2851 p = xdr_reserve_space(xdr, 4); 2852 *p = cpu_to_be32(false); 2853 } 2854 2855 static void 2856 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, 2857 const struct nfs4_xdr_opaque_data *opaque) 2858 { 2859 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque, 2860 struct nfs42_layoutstat_devinfo, ld_private); 2861 __be32 *start; 2862 2863 /* layoutupdate length */ 2864 start = xdr_reserve_space(xdr, 4); 2865 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data); 2866 2867 *start = cpu_to_be32((xdr->p - start - 1) * 4); 2868 } 2869 2870 static void 2871 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque) 2872 { 2873 struct nfs4_ff_layout_ds_stripe *dss_info = opaque->data; 2874 struct nfs4_ff_layout_mirror *mirror = dss_info->mirror; 2875 2876 ff_layout_put_mirror(mirror); 2877 } 2878 2879 static const struct nfs4_xdr_opaque_ops layoutstat_ops = { 2880 .encode = ff_layout_encode_layoutstats, 2881 .free = ff_layout_free_layoutstats, 2882 }; 2883 2884 static int 2885 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 2886 struct nfs42_layoutstat_devinfo *devinfo, 2887 int dev_limit, enum nfs4_ff_op_type type) 2888 { 2889 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 2890 struct nfs4_ff_layout_mirror *mirror; 2891 struct nfs4_ff_layout_ds_stripe *dss_info; 2892 struct nfs4_deviceid_node *dev; 2893 int i = 0, dss_id; 2894 2895 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) { 2896 for (dss_id = 0; dss_id < mirror->dss_count; ++dss_id) { 2897 dss_info = &mirror->dss[dss_id]; 2898 if (i >= dev_limit) 2899 break; 2900 if (IS_ERR_OR_NULL(dss_info->mirror_ds)) 2901 continue; 2902 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, 2903 &mirror->flags) && 2904 type != NFS4_FF_OP_LAYOUTRETURN) 2905 continue; 2906 /* mirror refcount put in cleanup_layoutstats */ 2907 if (!refcount_inc_not_zero(&mirror->ref)) 2908 continue; 2909 dev = &dss_info->mirror_ds->id_node; 2910 memcpy(&devinfo->dev_id, 2911 &dev->deviceid, 2912 NFS4_DEVICEID4_SIZE); 2913 devinfo->offset = 0; 2914 devinfo->length = NFS4_MAX_UINT64; 2915 spin_lock(&mirror->lock); 2916 devinfo->read_count = 2917 dss_info->read_stat.io_stat.ops_completed; 2918 devinfo->read_bytes = 2919 dss_info->read_stat.io_stat.bytes_completed; 2920 devinfo->write_count = 2921 dss_info->write_stat.io_stat.ops_completed; 2922 devinfo->write_bytes = 2923 dss_info->write_stat.io_stat.bytes_completed; 2924 spin_unlock(&mirror->lock); 2925 devinfo->layout_type = LAYOUT_FLEX_FILES; 2926 devinfo->ld_private.ops = &layoutstat_ops; 2927 devinfo->ld_private.data = &mirror->dss[dss_id]; 2928 2929 devinfo++; 2930 i++; 2931 } 2932 } 2933 return i; 2934 } 2935 2936 static int ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args) 2937 { 2938 struct pnfs_layout_hdr *lo; 2939 struct nfs4_flexfile_layout *ff_layout; 2940 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV; 2941 2942 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */ 2943 args->devinfo = kmalloc_objs(*args->devinfo, dev_count, 2944 nfs_io_gfp_mask()); 2945 if (!args->devinfo) 2946 return -ENOMEM; 2947 2948 spin_lock(&args->inode->i_lock); 2949 lo = NFS_I(args->inode)->layout; 2950 if (lo && pnfs_layout_is_valid(lo)) { 2951 ff_layout = FF_LAYOUT_FROM_HDR(lo); 2952 args->num_dev = ff_layout_mirror_prepare_stats( 2953 &ff_layout->generic_hdr, &args->devinfo[0], dev_count, 2954 NFS4_FF_OP_LAYOUTSTATS); 2955 } else 2956 args->num_dev = 0; 2957 spin_unlock(&args->inode->i_lock); 2958 if (!args->num_dev) { 2959 kfree(args->devinfo); 2960 args->devinfo = NULL; 2961 return -ENOENT; 2962 } 2963 2964 return 0; 2965 } 2966 2967 static int 2968 ff_layout_set_layoutdriver(struct nfs_server *server, 2969 const struct nfs_fh *dummy) 2970 { 2971 #if IS_ENABLED(CONFIG_NFS_V4_2) 2972 server->caps |= NFS_CAP_LAYOUTSTATS | NFS_CAP_REBOOT_LAYOUTRETURN; 2973 #endif 2974 return 0; 2975 } 2976 2977 static const struct pnfs_commit_ops ff_layout_commit_ops = { 2978 .setup_ds_info = ff_layout_setup_ds_info, 2979 .release_ds_info = ff_layout_release_ds_info, 2980 .mark_request_commit = pnfs_layout_mark_request_commit, 2981 .clear_request_commit = pnfs_generic_clear_request_commit, 2982 .scan_commit_lists = pnfs_generic_scan_commit_lists, 2983 .recover_commit_reqs = pnfs_generic_recover_commit_reqs, 2984 .commit_pagelist = ff_layout_commit_pagelist, 2985 }; 2986 2987 static struct pnfs_layoutdriver_type flexfilelayout_type = { 2988 .id = LAYOUT_FLEX_FILES, 2989 .name = "LAYOUT_FLEX_FILES", 2990 .owner = THIS_MODULE, 2991 .flags = PNFS_LAYOUTGET_ON_OPEN, 2992 .max_layoutget_response = 4096, /* 1 page or so... */ 2993 .set_layoutdriver = ff_layout_set_layoutdriver, 2994 .alloc_layout_hdr = ff_layout_alloc_layout_hdr, 2995 .free_layout_hdr = ff_layout_free_layout_hdr, 2996 .alloc_lseg = ff_layout_alloc_lseg, 2997 .free_lseg = ff_layout_free_lseg, 2998 .add_lseg = ff_layout_add_lseg, 2999 .pg_read_ops = &ff_layout_pg_read_ops, 3000 .pg_write_ops = &ff_layout_pg_write_ops, 3001 .get_ds_info = ff_layout_get_ds_info, 3002 .free_deviceid_node = ff_layout_free_deviceid_node, 3003 .read_pagelist = ff_layout_read_pagelist, 3004 .write_pagelist = ff_layout_write_pagelist, 3005 .alloc_deviceid_node = ff_layout_alloc_deviceid_node, 3006 .prepare_layoutreturn = ff_layout_prepare_layoutreturn, 3007 .sync = pnfs_nfs_generic_sync, 3008 .prepare_layoutstats = ff_layout_prepare_layoutstats, 3009 .cancel_io = ff_layout_cancel_io, 3010 }; 3011 3012 static int __init nfs4flexfilelayout_init(void) 3013 { 3014 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n", 3015 __func__); 3016 return pnfs_register_layoutdriver(&flexfilelayout_type); 3017 } 3018 3019 static void __exit nfs4flexfilelayout_exit(void) 3020 { 3021 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n", 3022 __func__); 3023 pnfs_unregister_layoutdriver(&flexfilelayout_type); 3024 } 3025 3026 MODULE_ALIAS("nfs-layouttype4-4"); 3027 3028 MODULE_LICENSE("GPL"); 3029 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver"); 3030 3031 module_init(nfs4flexfilelayout_init); 3032 module_exit(nfs4flexfilelayout_exit); 3033 3034 module_param(io_maxretrans, ushort, 0644); 3035 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client " 3036 "retries an I/O request before returning an error. "); 3037