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