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(l2); 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 = ERR_PTR(-EAGAIN); 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 (IS_ERR(ds)) 773 continue; 774 775 if (check_device && 776 nfs4_test_deviceid_unavailable(&mirror->mirror_ds->id_node)) { 777 // reinitialize the error state in case if this is the last iteration 778 ds = ERR_PTR(-EINVAL); 779 continue; 780 } 781 782 *best_idx = idx; 783 break; 784 } 785 786 return ds; 787 } 788 789 static struct nfs4_pnfs_ds * 790 ff_layout_choose_any_ds_for_read(struct pnfs_layout_segment *lseg, 791 u32 start_idx, u32 *best_idx) 792 { 793 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, false); 794 } 795 796 static struct nfs4_pnfs_ds * 797 ff_layout_choose_valid_ds_for_read(struct pnfs_layout_segment *lseg, 798 u32 start_idx, u32 *best_idx) 799 { 800 return ff_layout_choose_ds_for_read(lseg, start_idx, best_idx, true); 801 } 802 803 static struct nfs4_pnfs_ds * 804 ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, 805 u32 start_idx, u32 *best_idx) 806 { 807 struct nfs4_pnfs_ds *ds; 808 809 ds = ff_layout_choose_valid_ds_for_read(lseg, start_idx, best_idx); 810 if (!IS_ERR(ds)) 811 return ds; 812 return ff_layout_choose_any_ds_for_read(lseg, start_idx, best_idx); 813 } 814 815 static struct nfs4_pnfs_ds * 816 ff_layout_get_ds_for_read(struct nfs_pageio_descriptor *pgio, 817 u32 *best_idx) 818 { 819 struct pnfs_layout_segment *lseg = pgio->pg_lseg; 820 struct nfs4_pnfs_ds *ds; 821 822 ds = ff_layout_choose_best_ds_for_read(lseg, pgio->pg_mirror_idx, 823 best_idx); 824 if (!IS_ERR(ds) || !pgio->pg_mirror_idx) 825 return ds; 826 return ff_layout_choose_best_ds_for_read(lseg, 0, best_idx); 827 } 828 829 static void 830 ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, 831 struct nfs_page *req, 832 bool strict_iomode) 833 { 834 pnfs_put_lseg(pgio->pg_lseg); 835 pgio->pg_lseg = 836 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 837 req_offset(req), req->wb_bytes, IOMODE_READ, 838 strict_iomode, nfs_io_gfp_mask()); 839 if (IS_ERR(pgio->pg_lseg)) { 840 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 841 pgio->pg_lseg = NULL; 842 } 843 } 844 845 static void 846 ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, 847 struct nfs_page *req) 848 { 849 struct nfs_pgio_mirror *pgm; 850 struct nfs4_ff_layout_mirror *mirror; 851 struct nfs4_pnfs_ds *ds; 852 u32 ds_idx; 853 854 if (NFS_SERVER(pgio->pg_inode)->flags & 855 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR)) 856 pgio->pg_maxretrans = io_maxretrans; 857 retry: 858 pnfs_generic_pg_check_layout(pgio, req); 859 /* Use full layout for now */ 860 if (!pgio->pg_lseg) { 861 ff_layout_pg_get_read(pgio, req, false); 862 if (!pgio->pg_lseg) 863 goto out_nolseg; 864 } 865 if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { 866 ff_layout_pg_get_read(pgio, req, true); 867 if (!pgio->pg_lseg) 868 goto out_nolseg; 869 } 870 /* Reset wb_nio, since getting layout segment was successful */ 871 req->wb_nio = 0; 872 873 ds = ff_layout_get_ds_for_read(pgio, &ds_idx); 874 if (IS_ERR(ds)) { 875 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 876 goto out_mds; 877 pnfs_generic_pg_cleanup(pgio); 878 /* Sleep for 1 second before retrying */ 879 ssleep(1); 880 goto retry; 881 } 882 883 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx); 884 pgm = &pgio->pg_mirrors[0]; 885 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize; 886 887 pgio->pg_mirror_idx = ds_idx; 888 return; 889 out_nolseg: 890 if (pgio->pg_error < 0) { 891 if (pgio->pg_error != -EAGAIN) 892 return; 893 /* Retry getting layout segment if lower layer returned -EAGAIN */ 894 if (pgio->pg_maxretrans && req->wb_nio++ > pgio->pg_maxretrans) { 895 if (NFS_SERVER(pgio->pg_inode)->flags & NFS_MOUNT_SOFTERR) 896 pgio->pg_error = -ETIMEDOUT; 897 else 898 pgio->pg_error = -EIO; 899 return; 900 } 901 pgio->pg_error = 0; 902 /* Sleep for 1 second before retrying */ 903 ssleep(1); 904 goto retry; 905 } 906 out_mds: 907 trace_pnfs_mds_fallback_pg_init_read(pgio->pg_inode, 908 0, NFS4_MAX_UINT64, IOMODE_READ, 909 NFS_I(pgio->pg_inode)->layout, 910 pgio->pg_lseg); 911 pgio->pg_maxretrans = 0; 912 nfs_pageio_reset_read_mds(pgio); 913 } 914 915 static void 916 ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 917 struct nfs_page *req) 918 { 919 struct nfs4_ff_layout_mirror *mirror; 920 struct nfs_pgio_mirror *pgm; 921 struct nfs4_pnfs_ds *ds; 922 u32 i; 923 924 retry: 925 pnfs_generic_pg_check_layout(pgio, req); 926 if (!pgio->pg_lseg) { 927 pgio->pg_lseg = 928 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 929 req_offset(req), req->wb_bytes, 930 IOMODE_RW, false, nfs_io_gfp_mask()); 931 if (IS_ERR(pgio->pg_lseg)) { 932 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 933 pgio->pg_lseg = NULL; 934 return; 935 } 936 } 937 /* If no lseg, fall back to write through mds */ 938 if (pgio->pg_lseg == NULL) 939 goto out_mds; 940 941 /* Use a direct mapping of ds_idx to pgio mirror_idx */ 942 if (pgio->pg_mirror_count != FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)) 943 goto out_eagain; 944 945 for (i = 0; i < pgio->pg_mirror_count; i++) { 946 mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i); 947 ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, mirror, true); 948 if (IS_ERR(ds)) { 949 if (!ff_layout_no_fallback_to_mds(pgio->pg_lseg)) 950 goto out_mds; 951 pnfs_generic_pg_cleanup(pgio); 952 /* Sleep for 1 second before retrying */ 953 ssleep(1); 954 goto retry; 955 } 956 pgm = &pgio->pg_mirrors[i]; 957 pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize; 958 } 959 960 if (NFS_SERVER(pgio->pg_inode)->flags & 961 (NFS_MOUNT_SOFT|NFS_MOUNT_SOFTERR)) 962 pgio->pg_maxretrans = io_maxretrans; 963 return; 964 out_eagain: 965 pnfs_generic_pg_cleanup(pgio); 966 pgio->pg_error = -EAGAIN; 967 return; 968 out_mds: 969 trace_pnfs_mds_fallback_pg_init_write(pgio->pg_inode, 970 0, NFS4_MAX_UINT64, IOMODE_RW, 971 NFS_I(pgio->pg_inode)->layout, 972 pgio->pg_lseg); 973 pgio->pg_maxretrans = 0; 974 nfs_pageio_reset_write_mds(pgio); 975 pgio->pg_error = -EAGAIN; 976 } 977 978 static unsigned int 979 ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 980 struct nfs_page *req) 981 { 982 if (!pgio->pg_lseg) { 983 pgio->pg_lseg = 984 pnfs_update_layout(pgio->pg_inode, nfs_req_openctx(req), 985 req_offset(req), req->wb_bytes, 986 IOMODE_RW, false, nfs_io_gfp_mask()); 987 if (IS_ERR(pgio->pg_lseg)) { 988 pgio->pg_error = PTR_ERR(pgio->pg_lseg); 989 pgio->pg_lseg = NULL; 990 goto out; 991 } 992 } 993 if (pgio->pg_lseg) 994 return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg); 995 996 trace_pnfs_mds_fallback_pg_get_mirror_count(pgio->pg_inode, 997 0, NFS4_MAX_UINT64, IOMODE_RW, 998 NFS_I(pgio->pg_inode)->layout, 999 pgio->pg_lseg); 1000 /* no lseg means that pnfs is not in use, so no mirroring here */ 1001 nfs_pageio_reset_write_mds(pgio); 1002 out: 1003 return 1; 1004 } 1005 1006 static u32 1007 ff_layout_pg_set_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx) 1008 { 1009 u32 old = desc->pg_mirror_idx; 1010 1011 desc->pg_mirror_idx = idx; 1012 return old; 1013 } 1014 1015 static struct nfs_pgio_mirror * 1016 ff_layout_pg_get_mirror_write(struct nfs_pageio_descriptor *desc, u32 idx) 1017 { 1018 return &desc->pg_mirrors[idx]; 1019 } 1020 1021 static const struct nfs_pageio_ops ff_layout_pg_read_ops = { 1022 .pg_init = ff_layout_pg_init_read, 1023 .pg_test = pnfs_generic_pg_test, 1024 .pg_doio = pnfs_generic_pg_readpages, 1025 .pg_cleanup = pnfs_generic_pg_cleanup, 1026 }; 1027 1028 static const struct nfs_pageio_ops ff_layout_pg_write_ops = { 1029 .pg_init = ff_layout_pg_init_write, 1030 .pg_test = pnfs_generic_pg_test, 1031 .pg_doio = pnfs_generic_pg_writepages, 1032 .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write, 1033 .pg_cleanup = pnfs_generic_pg_cleanup, 1034 .pg_get_mirror = ff_layout_pg_get_mirror_write, 1035 .pg_set_mirror = ff_layout_pg_set_mirror_write, 1036 }; 1037 1038 static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs) 1039 { 1040 struct rpc_task *task = &hdr->task; 1041 1042 pnfs_layoutcommit_inode(hdr->inode, false); 1043 1044 if (retry_pnfs) { 1045 dprintk("%s Reset task %5u for i/o through pNFS " 1046 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1047 hdr->task.tk_pid, 1048 hdr->inode->i_sb->s_id, 1049 (unsigned long long)NFS_FILEID(hdr->inode), 1050 hdr->args.count, 1051 (unsigned long long)hdr->args.offset); 1052 1053 hdr->completion_ops->reschedule_io(hdr); 1054 return; 1055 } 1056 1057 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1058 dprintk("%s Reset task %5u for i/o through MDS " 1059 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1060 hdr->task.tk_pid, 1061 hdr->inode->i_sb->s_id, 1062 (unsigned long long)NFS_FILEID(hdr->inode), 1063 hdr->args.count, 1064 (unsigned long long)hdr->args.offset); 1065 1066 trace_pnfs_mds_fallback_write_done(hdr->inode, 1067 hdr->args.offset, hdr->args.count, 1068 IOMODE_RW, NFS_I(hdr->inode)->layout, 1069 hdr->lseg); 1070 task->tk_status = pnfs_write_done_resend_to_mds(hdr); 1071 } 1072 } 1073 1074 static void ff_layout_resend_pnfs_read(struct nfs_pgio_header *hdr) 1075 { 1076 u32 idx = hdr->pgio_mirror_idx + 1; 1077 u32 new_idx = 0; 1078 struct nfs4_pnfs_ds *ds; 1079 1080 ds = ff_layout_choose_any_ds_for_read(hdr->lseg, idx, &new_idx); 1081 if (IS_ERR(ds)) 1082 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); 1083 else 1084 ff_layout_send_layouterror(hdr->lseg); 1085 pnfs_read_resend_pnfs(hdr, new_idx); 1086 } 1087 1088 static void ff_layout_reset_read(struct nfs_pgio_header *hdr) 1089 { 1090 struct rpc_task *task = &hdr->task; 1091 1092 pnfs_layoutcommit_inode(hdr->inode, false); 1093 pnfs_error_mark_layout_for_return(hdr->inode, hdr->lseg); 1094 1095 if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) { 1096 dprintk("%s Reset task %5u for i/o through MDS " 1097 "(req %s/%llu, %u bytes @ offset %llu)\n", __func__, 1098 hdr->task.tk_pid, 1099 hdr->inode->i_sb->s_id, 1100 (unsigned long long)NFS_FILEID(hdr->inode), 1101 hdr->args.count, 1102 (unsigned long long)hdr->args.offset); 1103 1104 trace_pnfs_mds_fallback_read_done(hdr->inode, 1105 hdr->args.offset, hdr->args.count, 1106 IOMODE_READ, NFS_I(hdr->inode)->layout, 1107 hdr->lseg); 1108 task->tk_status = pnfs_read_done_resend_to_mds(hdr); 1109 } 1110 } 1111 1112 static int ff_layout_async_handle_error_v4(struct rpc_task *task, 1113 u32 op_status, 1114 struct nfs4_state *state, 1115 struct nfs_client *clp, 1116 struct pnfs_layout_segment *lseg, 1117 u32 idx) 1118 { 1119 struct pnfs_layout_hdr *lo = lseg->pls_layout; 1120 struct inode *inode = lo->plh_inode; 1121 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1122 struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table; 1123 1124 switch (op_status) { 1125 case NFS4_OK: 1126 case NFS4ERR_NXIO: 1127 break; 1128 case NFSERR_PERM: 1129 if (!task->tk_xprt) 1130 break; 1131 xprt_force_disconnect(task->tk_xprt); 1132 goto out_retry; 1133 case NFS4ERR_BADSESSION: 1134 case NFS4ERR_BADSLOT: 1135 case NFS4ERR_BAD_HIGH_SLOT: 1136 case NFS4ERR_DEADSESSION: 1137 case NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1138 case NFS4ERR_SEQ_FALSE_RETRY: 1139 case NFS4ERR_SEQ_MISORDERED: 1140 dprintk("%s ERROR %d, Reset session. Exchangeid " 1141 "flags 0x%x\n", __func__, task->tk_status, 1142 clp->cl_exchange_flags); 1143 nfs4_schedule_session_recovery(clp->cl_session, task->tk_status); 1144 goto out_retry; 1145 case NFS4ERR_DELAY: 1146 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1147 fallthrough; 1148 case NFS4ERR_GRACE: 1149 rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX); 1150 goto out_retry; 1151 case NFS4ERR_RETRY_UNCACHED_REP: 1152 goto out_retry; 1153 /* Invalidate Layout errors */ 1154 case NFS4ERR_PNFS_NO_LAYOUT: 1155 case NFS4ERR_STALE: 1156 case NFS4ERR_BADHANDLE: 1157 case NFS4ERR_ISDIR: 1158 case NFS4ERR_FHEXPIRED: 1159 case NFS4ERR_WRONG_TYPE: 1160 dprintk("%s Invalid layout error %d\n", __func__, 1161 task->tk_status); 1162 /* 1163 * Destroy layout so new i/o will get a new layout. 1164 * Layout will not be destroyed until all current lseg 1165 * references are put. Mark layout as invalid to resend failed 1166 * i/o and all i/o waiting on the slot table to the MDS until 1167 * layout is destroyed and a new valid layout is obtained. 1168 */ 1169 pnfs_destroy_layout(NFS_I(inode)); 1170 rpc_wake_up(&tbl->slot_tbl_waitq); 1171 goto reset; 1172 default: 1173 break; 1174 } 1175 1176 switch (task->tk_status) { 1177 /* RPC connection errors */ 1178 case -ENETDOWN: 1179 case -ENETUNREACH: 1180 if (test_bit(NFS_CS_NETUNREACH_FATAL, &clp->cl_flags)) 1181 return -NFS4ERR_FATAL_IOERROR; 1182 fallthrough; 1183 case -ECONNREFUSED: 1184 case -EHOSTDOWN: 1185 case -EHOSTUNREACH: 1186 case -EIO: 1187 case -ETIMEDOUT: 1188 case -EPIPE: 1189 case -EPROTO: 1190 case -ENODEV: 1191 dprintk("%s DS connection error %d\n", __func__, 1192 task->tk_status); 1193 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1194 &devid->deviceid); 1195 rpc_wake_up(&tbl->slot_tbl_waitq); 1196 break; 1197 default: 1198 break; 1199 } 1200 1201 if (ff_layout_avoid_mds_available_ds(lseg)) 1202 return -NFS4ERR_RESET_TO_PNFS; 1203 reset: 1204 dprintk("%s Retry through MDS. Error %d\n", __func__, 1205 task->tk_status); 1206 return -NFS4ERR_RESET_TO_MDS; 1207 1208 out_retry: 1209 task->tk_status = 0; 1210 return -EAGAIN; 1211 } 1212 1213 /* Retry all errors through either pNFS or MDS except for -EJUKEBOX */ 1214 static int ff_layout_async_handle_error_v3(struct rpc_task *task, 1215 u32 op_status, 1216 struct nfs_client *clp, 1217 struct pnfs_layout_segment *lseg, 1218 u32 idx) 1219 { 1220 struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx); 1221 1222 switch (op_status) { 1223 case NFS_OK: 1224 case NFSERR_NXIO: 1225 break; 1226 case NFSERR_PERM: 1227 if (!task->tk_xprt) 1228 break; 1229 xprt_force_disconnect(task->tk_xprt); 1230 goto out_retry; 1231 case NFSERR_ACCES: 1232 case NFSERR_BADHANDLE: 1233 case NFSERR_FBIG: 1234 case NFSERR_IO: 1235 case NFSERR_NOSPC: 1236 case NFSERR_ROFS: 1237 case NFSERR_STALE: 1238 goto out_reset_to_pnfs; 1239 case NFSERR_JUKEBOX: 1240 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1241 goto out_retry; 1242 default: 1243 break; 1244 } 1245 1246 switch (task->tk_status) { 1247 /* File access problems. Don't mark the device as unavailable */ 1248 case -EACCES: 1249 case -ESTALE: 1250 case -EISDIR: 1251 case -EBADHANDLE: 1252 case -ELOOP: 1253 case -ENOSPC: 1254 break; 1255 case -EJUKEBOX: 1256 nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY); 1257 goto out_retry; 1258 case -ENETDOWN: 1259 case -ENETUNREACH: 1260 if (test_bit(NFS_CS_NETUNREACH_FATAL, &clp->cl_flags)) 1261 return -NFS4ERR_FATAL_IOERROR; 1262 fallthrough; 1263 default: 1264 dprintk("%s DS connection error %d\n", __func__, 1265 task->tk_status); 1266 nfs4_delete_deviceid(devid->ld, devid->nfs_client, 1267 &devid->deviceid); 1268 } 1269 out_reset_to_pnfs: 1270 /* FIXME: Need to prevent infinite looping here. */ 1271 return -NFS4ERR_RESET_TO_PNFS; 1272 out_retry: 1273 task->tk_status = 0; 1274 rpc_restart_call_prepare(task); 1275 rpc_delay(task, NFS_JUKEBOX_RETRY_TIME); 1276 return -EAGAIN; 1277 } 1278 1279 static int ff_layout_async_handle_error(struct rpc_task *task, 1280 u32 op_status, 1281 struct nfs4_state *state, 1282 struct nfs_client *clp, 1283 struct pnfs_layout_segment *lseg, 1284 u32 idx) 1285 { 1286 int vers = clp->cl_nfs_mod->rpc_vers->number; 1287 1288 if (task->tk_status >= 0) { 1289 ff_layout_mark_ds_reachable(lseg, idx); 1290 return 0; 1291 } 1292 1293 /* Handle the case of an invalid layout segment */ 1294 if (!pnfs_is_valid_lseg(lseg)) 1295 return -NFS4ERR_RESET_TO_PNFS; 1296 1297 switch (vers) { 1298 case 3: 1299 return ff_layout_async_handle_error_v3(task, op_status, clp, 1300 lseg, idx); 1301 case 4: 1302 return ff_layout_async_handle_error_v4(task, op_status, state, 1303 clp, lseg, idx); 1304 default: 1305 /* should never happen */ 1306 WARN_ON_ONCE(1); 1307 return 0; 1308 } 1309 } 1310 1311 static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg, 1312 u32 idx, u64 offset, u64 length, 1313 u32 *op_status, int opnum, int error) 1314 { 1315 struct nfs4_ff_layout_mirror *mirror; 1316 u32 status = *op_status; 1317 int err; 1318 1319 if (status == 0) { 1320 switch (error) { 1321 case -ETIMEDOUT: 1322 case -EPFNOSUPPORT: 1323 case -EPROTONOSUPPORT: 1324 case -EOPNOTSUPP: 1325 case -EINVAL: 1326 case -ECONNREFUSED: 1327 case -ECONNRESET: 1328 case -EHOSTDOWN: 1329 case -EHOSTUNREACH: 1330 case -ENETDOWN: 1331 case -ENETUNREACH: 1332 case -EADDRINUSE: 1333 case -ENOBUFS: 1334 case -EPIPE: 1335 case -EPERM: 1336 case -EPROTO: 1337 case -ENODEV: 1338 *op_status = status = NFS4ERR_NXIO; 1339 break; 1340 case -EACCES: 1341 *op_status = status = NFS4ERR_ACCESS; 1342 break; 1343 default: 1344 return; 1345 } 1346 } 1347 1348 mirror = FF_LAYOUT_COMP(lseg, idx); 1349 err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout), 1350 mirror, offset, length, status, opnum, 1351 nfs_io_gfp_mask()); 1352 1353 switch (status) { 1354 case NFS4ERR_DELAY: 1355 case NFS4ERR_GRACE: 1356 case NFS4ERR_PERM: 1357 break; 1358 case NFS4ERR_NXIO: 1359 ff_layout_mark_ds_unreachable(lseg, idx); 1360 /* 1361 * Don't return the layout if this is a read and we still 1362 * have layouts to try 1363 */ 1364 if (opnum == OP_READ) 1365 break; 1366 fallthrough; 1367 default: 1368 pnfs_error_mark_layout_for_return(lseg->pls_layout->plh_inode, 1369 lseg); 1370 } 1371 1372 dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status); 1373 } 1374 1375 /* NFS_PROTO call done callback routines */ 1376 static int ff_layout_read_done_cb(struct rpc_task *task, 1377 struct nfs_pgio_header *hdr) 1378 { 1379 int err; 1380 1381 if (task->tk_status < 0) { 1382 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1383 hdr->args.offset, hdr->args.count, 1384 &hdr->res.op_status, OP_READ, 1385 task->tk_status); 1386 trace_ff_layout_read_error(hdr, task->tk_status); 1387 } 1388 1389 err = ff_layout_async_handle_error(task, hdr->res.op_status, 1390 hdr->args.context->state, 1391 hdr->ds_clp, hdr->lseg, 1392 hdr->pgio_mirror_idx); 1393 1394 trace_nfs4_pnfs_read(hdr, err); 1395 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1396 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1397 switch (err) { 1398 case -NFS4ERR_RESET_TO_PNFS: 1399 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1400 return task->tk_status; 1401 case -NFS4ERR_RESET_TO_MDS: 1402 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1403 return task->tk_status; 1404 case -EAGAIN: 1405 goto out_eagain; 1406 case -NFS4ERR_FATAL_IOERROR: 1407 task->tk_status = -EIO; 1408 return 0; 1409 } 1410 1411 return 0; 1412 out_eagain: 1413 rpc_restart_call_prepare(task); 1414 return -EAGAIN; 1415 } 1416 1417 static bool 1418 ff_layout_need_layoutcommit(struct pnfs_layout_segment *lseg) 1419 { 1420 return !(FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_LAYOUTCOMMIT); 1421 } 1422 1423 /* 1424 * We reference the rpc_cred of the first WRITE that triggers the need for 1425 * a LAYOUTCOMMIT, and use it to send the layoutcommit compound. 1426 * rfc5661 is not clear about which credential should be used. 1427 * 1428 * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so 1429 * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 1430 * we always send layoutcommit after DS writes. 1431 */ 1432 static void 1433 ff_layout_set_layoutcommit(struct inode *inode, 1434 struct pnfs_layout_segment *lseg, 1435 loff_t end_offset) 1436 { 1437 if (!ff_layout_need_layoutcommit(lseg)) 1438 return; 1439 1440 pnfs_set_layoutcommit(inode, lseg, end_offset); 1441 dprintk("%s inode %lu pls_end_pos %llu\n", __func__, inode->i_ino, 1442 (unsigned long long) NFS_I(inode)->layout->plh_lwb); 1443 } 1444 1445 static void ff_layout_read_record_layoutstats_start(struct rpc_task *task, 1446 struct nfs_pgio_header *hdr) 1447 { 1448 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1449 return; 1450 nfs4_ff_layout_stat_io_start_read(hdr->inode, 1451 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1452 hdr->args.count, 1453 task->tk_start); 1454 } 1455 1456 static void ff_layout_read_record_layoutstats_done(struct rpc_task *task, 1457 struct nfs_pgio_header *hdr) 1458 { 1459 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1460 return; 1461 nfs4_ff_layout_stat_io_end_read(task, 1462 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1463 hdr->args.count, 1464 hdr->res.count); 1465 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1466 } 1467 1468 static int ff_layout_read_prepare_common(struct rpc_task *task, 1469 struct nfs_pgio_header *hdr) 1470 { 1471 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1472 rpc_exit(task, -EIO); 1473 return -EIO; 1474 } 1475 1476 if (!pnfs_is_valid_lseg(hdr->lseg)) { 1477 rpc_exit(task, -EAGAIN); 1478 return -EAGAIN; 1479 } 1480 1481 ff_layout_read_record_layoutstats_start(task, hdr); 1482 return 0; 1483 } 1484 1485 /* 1486 * Call ops for the async read/write cases 1487 * In the case of dense layouts, the offset needs to be reset to its 1488 * original value. 1489 */ 1490 static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data) 1491 { 1492 struct nfs_pgio_header *hdr = data; 1493 1494 if (ff_layout_read_prepare_common(task, hdr)) 1495 return; 1496 1497 rpc_call_start(task); 1498 } 1499 1500 static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data) 1501 { 1502 struct nfs_pgio_header *hdr = data; 1503 1504 if (nfs4_setup_sequence(hdr->ds_clp, 1505 &hdr->args.seq_args, 1506 &hdr->res.seq_res, 1507 task)) 1508 return; 1509 1510 ff_layout_read_prepare_common(task, hdr); 1511 } 1512 1513 static void ff_layout_read_call_done(struct rpc_task *task, void *data) 1514 { 1515 struct nfs_pgio_header *hdr = data; 1516 1517 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1518 task->tk_status == 0) { 1519 nfs4_sequence_done(task, &hdr->res.seq_res); 1520 return; 1521 } 1522 1523 /* Note this may cause RPC to be resent */ 1524 hdr->mds_ops->rpc_call_done(task, hdr); 1525 } 1526 1527 static void ff_layout_read_count_stats(struct rpc_task *task, void *data) 1528 { 1529 struct nfs_pgio_header *hdr = data; 1530 1531 ff_layout_read_record_layoutstats_done(task, hdr); 1532 rpc_count_iostats_metrics(task, 1533 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]); 1534 } 1535 1536 static void ff_layout_read_release(void *data) 1537 { 1538 struct nfs_pgio_header *hdr = data; 1539 1540 ff_layout_read_record_layoutstats_done(&hdr->task, hdr); 1541 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) 1542 ff_layout_resend_pnfs_read(hdr); 1543 else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1544 ff_layout_reset_read(hdr); 1545 pnfs_generic_rw_release(data); 1546 } 1547 1548 1549 static int ff_layout_write_done_cb(struct rpc_task *task, 1550 struct nfs_pgio_header *hdr) 1551 { 1552 loff_t end_offs = 0; 1553 int err; 1554 1555 if (task->tk_status < 0) { 1556 ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx, 1557 hdr->args.offset, hdr->args.count, 1558 &hdr->res.op_status, OP_WRITE, 1559 task->tk_status); 1560 trace_ff_layout_write_error(hdr, task->tk_status); 1561 } 1562 1563 err = ff_layout_async_handle_error(task, hdr->res.op_status, 1564 hdr->args.context->state, 1565 hdr->ds_clp, hdr->lseg, 1566 hdr->pgio_mirror_idx); 1567 1568 trace_nfs4_pnfs_write(hdr, err); 1569 clear_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1570 clear_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1571 switch (err) { 1572 case -NFS4ERR_RESET_TO_PNFS: 1573 set_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags); 1574 return task->tk_status; 1575 case -NFS4ERR_RESET_TO_MDS: 1576 set_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags); 1577 return task->tk_status; 1578 case -EAGAIN: 1579 return -EAGAIN; 1580 case -NFS4ERR_FATAL_IOERROR: 1581 task->tk_status = -EIO; 1582 return 0; 1583 } 1584 1585 if (hdr->res.verf->committed == NFS_FILE_SYNC || 1586 hdr->res.verf->committed == NFS_DATA_SYNC) 1587 end_offs = hdr->mds_offset + (loff_t)hdr->res.count; 1588 1589 /* Note: if the write is unstable, don't set end_offs until commit */ 1590 ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); 1591 1592 /* zero out fattr since we don't care DS attr at all */ 1593 hdr->fattr.valid = 0; 1594 if (task->tk_status >= 0) 1595 nfs_writeback_update_inode(hdr); 1596 1597 return 0; 1598 } 1599 1600 static int ff_layout_commit_done_cb(struct rpc_task *task, 1601 struct nfs_commit_data *data) 1602 { 1603 int err; 1604 1605 if (task->tk_status < 0) { 1606 ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index, 1607 data->args.offset, data->args.count, 1608 &data->res.op_status, OP_COMMIT, 1609 task->tk_status); 1610 trace_ff_layout_commit_error(data, task->tk_status); 1611 } 1612 1613 err = ff_layout_async_handle_error(task, data->res.op_status, 1614 NULL, data->ds_clp, data->lseg, 1615 data->ds_commit_index); 1616 1617 trace_nfs4_pnfs_commit_ds(data, err); 1618 switch (err) { 1619 case -NFS4ERR_RESET_TO_PNFS: 1620 pnfs_generic_prepare_to_resend_writes(data); 1621 return -EAGAIN; 1622 case -NFS4ERR_RESET_TO_MDS: 1623 pnfs_generic_prepare_to_resend_writes(data); 1624 return -EAGAIN; 1625 case -EAGAIN: 1626 rpc_restart_call_prepare(task); 1627 return -EAGAIN; 1628 case -NFS4ERR_FATAL_IOERROR: 1629 task->tk_status = -EIO; 1630 return 0; 1631 } 1632 1633 ff_layout_set_layoutcommit(data->inode, data->lseg, data->lwb); 1634 1635 return 0; 1636 } 1637 1638 static void ff_layout_write_record_layoutstats_start(struct rpc_task *task, 1639 struct nfs_pgio_header *hdr) 1640 { 1641 if (test_and_set_bit(NFS_IOHDR_STAT, &hdr->flags)) 1642 return; 1643 nfs4_ff_layout_stat_io_start_write(hdr->inode, 1644 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1645 hdr->args.count, 1646 task->tk_start); 1647 } 1648 1649 static void ff_layout_write_record_layoutstats_done(struct rpc_task *task, 1650 struct nfs_pgio_header *hdr) 1651 { 1652 if (!test_and_clear_bit(NFS_IOHDR_STAT, &hdr->flags)) 1653 return; 1654 nfs4_ff_layout_stat_io_end_write(task, 1655 FF_LAYOUT_COMP(hdr->lseg, hdr->pgio_mirror_idx), 1656 hdr->args.count, hdr->res.count, 1657 hdr->res.verf->committed); 1658 set_bit(NFS_LSEG_LAYOUTRETURN, &hdr->lseg->pls_flags); 1659 } 1660 1661 static int ff_layout_write_prepare_common(struct rpc_task *task, 1662 struct nfs_pgio_header *hdr) 1663 { 1664 if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) { 1665 rpc_exit(task, -EIO); 1666 return -EIO; 1667 } 1668 1669 if (!pnfs_is_valid_lseg(hdr->lseg)) { 1670 rpc_exit(task, -EAGAIN); 1671 return -EAGAIN; 1672 } 1673 1674 ff_layout_write_record_layoutstats_start(task, hdr); 1675 return 0; 1676 } 1677 1678 static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data) 1679 { 1680 struct nfs_pgio_header *hdr = data; 1681 1682 if (ff_layout_write_prepare_common(task, hdr)) 1683 return; 1684 1685 rpc_call_start(task); 1686 } 1687 1688 static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data) 1689 { 1690 struct nfs_pgio_header *hdr = data; 1691 1692 if (nfs4_setup_sequence(hdr->ds_clp, 1693 &hdr->args.seq_args, 1694 &hdr->res.seq_res, 1695 task)) 1696 return; 1697 1698 ff_layout_write_prepare_common(task, hdr); 1699 } 1700 1701 static void ff_layout_write_call_done(struct rpc_task *task, void *data) 1702 { 1703 struct nfs_pgio_header *hdr = data; 1704 1705 if (test_bit(NFS_IOHDR_REDO, &hdr->flags) && 1706 task->tk_status == 0) { 1707 nfs4_sequence_done(task, &hdr->res.seq_res); 1708 return; 1709 } 1710 1711 /* Note this may cause RPC to be resent */ 1712 hdr->mds_ops->rpc_call_done(task, hdr); 1713 } 1714 1715 static void ff_layout_write_count_stats(struct rpc_task *task, void *data) 1716 { 1717 struct nfs_pgio_header *hdr = data; 1718 1719 ff_layout_write_record_layoutstats_done(task, hdr); 1720 rpc_count_iostats_metrics(task, 1721 &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]); 1722 } 1723 1724 static void ff_layout_write_release(void *data) 1725 { 1726 struct nfs_pgio_header *hdr = data; 1727 1728 ff_layout_write_record_layoutstats_done(&hdr->task, hdr); 1729 if (test_bit(NFS_IOHDR_RESEND_PNFS, &hdr->flags)) { 1730 ff_layout_send_layouterror(hdr->lseg); 1731 ff_layout_reset_write(hdr, true); 1732 } else if (test_bit(NFS_IOHDR_RESEND_MDS, &hdr->flags)) 1733 ff_layout_reset_write(hdr, false); 1734 pnfs_generic_rw_release(data); 1735 } 1736 1737 static void ff_layout_commit_record_layoutstats_start(struct rpc_task *task, 1738 struct nfs_commit_data *cdata) 1739 { 1740 if (test_and_set_bit(NFS_IOHDR_STAT, &cdata->flags)) 1741 return; 1742 nfs4_ff_layout_stat_io_start_write(cdata->inode, 1743 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1744 0, task->tk_start); 1745 } 1746 1747 static void ff_layout_commit_record_layoutstats_done(struct rpc_task *task, 1748 struct nfs_commit_data *cdata) 1749 { 1750 struct nfs_page *req; 1751 __u64 count = 0; 1752 1753 if (!test_and_clear_bit(NFS_IOHDR_STAT, &cdata->flags)) 1754 return; 1755 1756 if (task->tk_status == 0) { 1757 list_for_each_entry(req, &cdata->pages, wb_list) 1758 count += req->wb_bytes; 1759 } 1760 nfs4_ff_layout_stat_io_end_write(task, 1761 FF_LAYOUT_COMP(cdata->lseg, cdata->ds_commit_index), 1762 count, count, NFS_FILE_SYNC); 1763 set_bit(NFS_LSEG_LAYOUTRETURN, &cdata->lseg->pls_flags); 1764 } 1765 1766 static int ff_layout_commit_prepare_common(struct rpc_task *task, 1767 struct nfs_commit_data *cdata) 1768 { 1769 if (!pnfs_is_valid_lseg(cdata->lseg)) { 1770 rpc_exit(task, -EAGAIN); 1771 return -EAGAIN; 1772 } 1773 1774 ff_layout_commit_record_layoutstats_start(task, cdata); 1775 return 0; 1776 } 1777 1778 static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data) 1779 { 1780 if (ff_layout_commit_prepare_common(task, data)) 1781 return; 1782 1783 rpc_call_start(task); 1784 } 1785 1786 static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data) 1787 { 1788 struct nfs_commit_data *wdata = data; 1789 1790 if (nfs4_setup_sequence(wdata->ds_clp, 1791 &wdata->args.seq_args, 1792 &wdata->res.seq_res, 1793 task)) 1794 return; 1795 ff_layout_commit_prepare_common(task, data); 1796 } 1797 1798 static void ff_layout_commit_done(struct rpc_task *task, void *data) 1799 { 1800 pnfs_generic_write_commit_done(task, data); 1801 } 1802 1803 static void ff_layout_commit_count_stats(struct rpc_task *task, void *data) 1804 { 1805 struct nfs_commit_data *cdata = data; 1806 1807 ff_layout_commit_record_layoutstats_done(task, cdata); 1808 rpc_count_iostats_metrics(task, 1809 &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]); 1810 } 1811 1812 static void ff_layout_commit_release(void *data) 1813 { 1814 struct nfs_commit_data *cdata = data; 1815 1816 ff_layout_commit_record_layoutstats_done(&cdata->task, cdata); 1817 pnfs_generic_commit_release(data); 1818 } 1819 1820 static const struct rpc_call_ops ff_layout_read_call_ops_v3 = { 1821 .rpc_call_prepare = ff_layout_read_prepare_v3, 1822 .rpc_call_done = ff_layout_read_call_done, 1823 .rpc_count_stats = ff_layout_read_count_stats, 1824 .rpc_release = ff_layout_read_release, 1825 }; 1826 1827 static const struct rpc_call_ops ff_layout_read_call_ops_v4 = { 1828 .rpc_call_prepare = ff_layout_read_prepare_v4, 1829 .rpc_call_done = ff_layout_read_call_done, 1830 .rpc_count_stats = ff_layout_read_count_stats, 1831 .rpc_release = ff_layout_read_release, 1832 }; 1833 1834 static const struct rpc_call_ops ff_layout_write_call_ops_v3 = { 1835 .rpc_call_prepare = ff_layout_write_prepare_v3, 1836 .rpc_call_done = ff_layout_write_call_done, 1837 .rpc_count_stats = ff_layout_write_count_stats, 1838 .rpc_release = ff_layout_write_release, 1839 }; 1840 1841 static const struct rpc_call_ops ff_layout_write_call_ops_v4 = { 1842 .rpc_call_prepare = ff_layout_write_prepare_v4, 1843 .rpc_call_done = ff_layout_write_call_done, 1844 .rpc_count_stats = ff_layout_write_count_stats, 1845 .rpc_release = ff_layout_write_release, 1846 }; 1847 1848 static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = { 1849 .rpc_call_prepare = ff_layout_commit_prepare_v3, 1850 .rpc_call_done = ff_layout_commit_done, 1851 .rpc_count_stats = ff_layout_commit_count_stats, 1852 .rpc_release = ff_layout_commit_release, 1853 }; 1854 1855 static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = { 1856 .rpc_call_prepare = ff_layout_commit_prepare_v4, 1857 .rpc_call_done = ff_layout_commit_done, 1858 .rpc_count_stats = ff_layout_commit_count_stats, 1859 .rpc_release = ff_layout_commit_release, 1860 }; 1861 1862 static enum pnfs_try_status 1863 ff_layout_read_pagelist(struct nfs_pgio_header *hdr) 1864 { 1865 struct pnfs_layout_segment *lseg = hdr->lseg; 1866 struct nfs4_pnfs_ds *ds; 1867 struct rpc_clnt *ds_clnt; 1868 struct nfsd_file *localio; 1869 struct nfs4_ff_layout_mirror *mirror; 1870 const struct cred *ds_cred; 1871 loff_t offset = hdr->args.offset; 1872 u32 idx = hdr->pgio_mirror_idx; 1873 int vers; 1874 struct nfs_fh *fh; 1875 bool ds_fatal_error = false; 1876 1877 dprintk("--> %s ino %lu pgbase %u req %zu@%llu\n", 1878 __func__, hdr->inode->i_ino, 1879 hdr->args.pgbase, (size_t)hdr->args.count, offset); 1880 1881 mirror = FF_LAYOUT_COMP(lseg, idx); 1882 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, false); 1883 if (IS_ERR(ds)) { 1884 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds)); 1885 goto out_failed; 1886 } 1887 1888 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 1889 hdr->inode); 1890 if (IS_ERR(ds_clnt)) 1891 goto out_failed; 1892 1893 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred); 1894 if (!ds_cred) 1895 goto out_failed; 1896 1897 vers = nfs4_ff_layout_ds_version(mirror); 1898 1899 dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__, 1900 ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), vers); 1901 1902 hdr->pgio_done_cb = ff_layout_read_done_cb; 1903 refcount_inc(&ds->ds_clp->cl_count); 1904 hdr->ds_clp = ds->ds_clp; 1905 fh = nfs4_ff_layout_select_ds_fh(mirror); 1906 if (fh) 1907 hdr->args.fh = fh; 1908 1909 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid); 1910 1911 /* 1912 * Note that if we ever decide to split across DSes, 1913 * then we may need to handle dense-like offsets. 1914 */ 1915 hdr->args.offset = offset; 1916 hdr->mds_offset = offset; 1917 1918 /* Start IO accounting for local read */ 1919 localio = ff_local_open_fh(lseg, idx, ds->ds_clp, ds_cred, fh, FMODE_READ); 1920 if (localio) { 1921 hdr->task.tk_start = ktime_get(); 1922 ff_layout_read_record_layoutstats_start(&hdr->task, hdr); 1923 } 1924 1925 /* Perform an asynchronous read to ds */ 1926 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 1927 vers == 3 ? &ff_layout_read_call_ops_v3 : 1928 &ff_layout_read_call_ops_v4, 1929 0, RPC_TASK_SOFTCONN, localio); 1930 put_cred(ds_cred); 1931 return PNFS_ATTEMPTED; 1932 1933 out_failed: 1934 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error) 1935 return PNFS_TRY_AGAIN; 1936 trace_pnfs_mds_fallback_read_pagelist(hdr->inode, 1937 hdr->args.offset, hdr->args.count, 1938 IOMODE_READ, NFS_I(hdr->inode)->layout, lseg); 1939 return PNFS_NOT_ATTEMPTED; 1940 } 1941 1942 /* Perform async writes. */ 1943 static enum pnfs_try_status 1944 ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync) 1945 { 1946 struct pnfs_layout_segment *lseg = hdr->lseg; 1947 struct nfs4_pnfs_ds *ds; 1948 struct rpc_clnt *ds_clnt; 1949 struct nfsd_file *localio; 1950 struct nfs4_ff_layout_mirror *mirror; 1951 const struct cred *ds_cred; 1952 loff_t offset = hdr->args.offset; 1953 int vers; 1954 struct nfs_fh *fh; 1955 u32 idx = hdr->pgio_mirror_idx; 1956 bool ds_fatal_error = false; 1957 1958 mirror = FF_LAYOUT_COMP(lseg, idx); 1959 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true); 1960 if (IS_ERR(ds)) { 1961 ds_fatal_error = nfs_error_is_fatal(PTR_ERR(ds)); 1962 goto out_failed; 1963 } 1964 1965 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 1966 hdr->inode); 1967 if (IS_ERR(ds_clnt)) 1968 goto out_failed; 1969 1970 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, hdr->cred); 1971 if (!ds_cred) 1972 goto out_failed; 1973 1974 vers = nfs4_ff_layout_ds_version(mirror); 1975 1976 dprintk("%s ino %lu sync %d req %zu@%llu DS: %s cl_count %d vers %d\n", 1977 __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count, 1978 offset, ds->ds_remotestr, refcount_read(&ds->ds_clp->cl_count), 1979 vers); 1980 1981 hdr->pgio_done_cb = ff_layout_write_done_cb; 1982 refcount_inc(&ds->ds_clp->cl_count); 1983 hdr->ds_clp = ds->ds_clp; 1984 hdr->ds_commit_idx = idx; 1985 fh = nfs4_ff_layout_select_ds_fh(mirror); 1986 if (fh) 1987 hdr->args.fh = fh; 1988 1989 nfs4_ff_layout_select_ds_stateid(mirror, &hdr->args.stateid); 1990 1991 /* 1992 * Note that if we ever decide to split across DSes, 1993 * then we may need to handle dense-like offsets. 1994 */ 1995 hdr->args.offset = offset; 1996 1997 /* Start IO accounting for local write */ 1998 localio = ff_local_open_fh(lseg, idx, ds->ds_clp, ds_cred, fh, 1999 FMODE_READ|FMODE_WRITE); 2000 if (localio) { 2001 hdr->task.tk_start = ktime_get(); 2002 ff_layout_write_record_layoutstats_start(&hdr->task, hdr); 2003 } 2004 2005 /* Perform an asynchronous write */ 2006 nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops, 2007 vers == 3 ? &ff_layout_write_call_ops_v3 : 2008 &ff_layout_write_call_ops_v4, 2009 sync, RPC_TASK_SOFTCONN, localio); 2010 put_cred(ds_cred); 2011 return PNFS_ATTEMPTED; 2012 2013 out_failed: 2014 if (ff_layout_avoid_mds_available_ds(lseg) && !ds_fatal_error) 2015 return PNFS_TRY_AGAIN; 2016 trace_pnfs_mds_fallback_write_pagelist(hdr->inode, 2017 hdr->args.offset, hdr->args.count, 2018 IOMODE_RW, NFS_I(hdr->inode)->layout, lseg); 2019 return PNFS_NOT_ATTEMPTED; 2020 } 2021 2022 static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i) 2023 { 2024 return i; 2025 } 2026 2027 static struct nfs_fh * 2028 select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i) 2029 { 2030 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2031 2032 /* FIXME: Assume that there is only one NFS version available 2033 * for the DS. 2034 */ 2035 return &flseg->mirror_array[i]->fh_versions[0]; 2036 } 2037 2038 static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how) 2039 { 2040 struct pnfs_layout_segment *lseg = data->lseg; 2041 struct nfs4_pnfs_ds *ds; 2042 struct rpc_clnt *ds_clnt; 2043 struct nfsd_file *localio; 2044 struct nfs4_ff_layout_mirror *mirror; 2045 const struct cred *ds_cred; 2046 u32 idx; 2047 int vers, ret; 2048 struct nfs_fh *fh; 2049 2050 if (!lseg || !(pnfs_is_valid_lseg(lseg) || 2051 test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags))) 2052 goto out_err; 2053 2054 idx = calc_ds_index_from_commit(lseg, data->ds_commit_index); 2055 mirror = FF_LAYOUT_COMP(lseg, idx); 2056 ds = nfs4_ff_layout_prepare_ds(lseg, mirror, true); 2057 if (IS_ERR(ds)) 2058 goto out_err; 2059 2060 ds_clnt = nfs4_ff_find_or_create_ds_client(mirror, ds->ds_clp, 2061 data->inode); 2062 if (IS_ERR(ds_clnt)) 2063 goto out_err; 2064 2065 ds_cred = ff_layout_get_ds_cred(mirror, &lseg->pls_range, data->cred); 2066 if (!ds_cred) 2067 goto out_err; 2068 2069 vers = nfs4_ff_layout_ds_version(mirror); 2070 2071 dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__, 2072 data->inode->i_ino, how, refcount_read(&ds->ds_clp->cl_count), 2073 vers); 2074 data->commit_done_cb = ff_layout_commit_done_cb; 2075 data->cred = ds_cred; 2076 refcount_inc(&ds->ds_clp->cl_count); 2077 data->ds_clp = ds->ds_clp; 2078 fh = select_ds_fh_from_commit(lseg, data->ds_commit_index); 2079 if (fh) 2080 data->args.fh = fh; 2081 2082 /* Start IO accounting for local commit */ 2083 localio = ff_local_open_fh(lseg, idx, ds->ds_clp, ds_cred, fh, 2084 FMODE_READ|FMODE_WRITE); 2085 if (localio) { 2086 data->task.tk_start = ktime_get(); 2087 ff_layout_commit_record_layoutstats_start(&data->task, data); 2088 } 2089 2090 ret = nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops, 2091 vers == 3 ? &ff_layout_commit_call_ops_v3 : 2092 &ff_layout_commit_call_ops_v4, 2093 how, RPC_TASK_SOFTCONN, localio); 2094 put_cred(ds_cred); 2095 return ret; 2096 out_err: 2097 pnfs_generic_prepare_to_resend_writes(data); 2098 pnfs_generic_commit_release(data); 2099 return -EAGAIN; 2100 } 2101 2102 static int 2103 ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages, 2104 int how, struct nfs_commit_info *cinfo) 2105 { 2106 return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo, 2107 ff_layout_initiate_commit); 2108 } 2109 2110 static bool ff_layout_match_rw(const struct rpc_task *task, 2111 const struct nfs_pgio_header *hdr, 2112 const struct pnfs_layout_segment *lseg) 2113 { 2114 return hdr->lseg == lseg; 2115 } 2116 2117 static bool ff_layout_match_commit(const struct rpc_task *task, 2118 const struct nfs_commit_data *cdata, 2119 const struct pnfs_layout_segment *lseg) 2120 { 2121 return cdata->lseg == lseg; 2122 } 2123 2124 static bool ff_layout_match_io(const struct rpc_task *task, const void *data) 2125 { 2126 const struct rpc_call_ops *ops = task->tk_ops; 2127 2128 if (ops == &ff_layout_read_call_ops_v3 || 2129 ops == &ff_layout_read_call_ops_v4 || 2130 ops == &ff_layout_write_call_ops_v3 || 2131 ops == &ff_layout_write_call_ops_v4) 2132 return ff_layout_match_rw(task, task->tk_calldata, data); 2133 if (ops == &ff_layout_commit_call_ops_v3 || 2134 ops == &ff_layout_commit_call_ops_v4) 2135 return ff_layout_match_commit(task, task->tk_calldata, data); 2136 return false; 2137 } 2138 2139 static void ff_layout_cancel_io(struct pnfs_layout_segment *lseg) 2140 { 2141 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2142 struct nfs4_ff_layout_mirror *mirror; 2143 struct nfs4_ff_layout_ds *mirror_ds; 2144 struct nfs4_pnfs_ds *ds; 2145 struct nfs_client *ds_clp; 2146 struct rpc_clnt *clnt; 2147 u32 idx; 2148 2149 for (idx = 0; idx < flseg->mirror_array_cnt; idx++) { 2150 mirror = flseg->mirror_array[idx]; 2151 mirror_ds = mirror->mirror_ds; 2152 if (IS_ERR_OR_NULL(mirror_ds)) 2153 continue; 2154 ds = mirror->mirror_ds->ds; 2155 if (!ds) 2156 continue; 2157 ds_clp = ds->ds_clp; 2158 if (!ds_clp) 2159 continue; 2160 clnt = ds_clp->cl_rpcclient; 2161 if (!clnt) 2162 continue; 2163 if (!rpc_cancel_tasks(clnt, -EAGAIN, ff_layout_match_io, lseg)) 2164 continue; 2165 rpc_clnt_disconnect(clnt); 2166 } 2167 } 2168 2169 static struct pnfs_ds_commit_info * 2170 ff_layout_get_ds_info(struct inode *inode) 2171 { 2172 struct pnfs_layout_hdr *layout = NFS_I(inode)->layout; 2173 2174 if (layout == NULL) 2175 return NULL; 2176 2177 return &FF_LAYOUT_FROM_HDR(layout)->commit_info; 2178 } 2179 2180 static void 2181 ff_layout_setup_ds_info(struct pnfs_ds_commit_info *fl_cinfo, 2182 struct pnfs_layout_segment *lseg) 2183 { 2184 struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg); 2185 struct inode *inode = lseg->pls_layout->plh_inode; 2186 struct pnfs_commit_array *array, *new; 2187 2188 new = pnfs_alloc_commit_array(flseg->mirror_array_cnt, 2189 nfs_io_gfp_mask()); 2190 if (new) { 2191 spin_lock(&inode->i_lock); 2192 array = pnfs_add_commit_array(fl_cinfo, new, lseg); 2193 spin_unlock(&inode->i_lock); 2194 if (array != new) 2195 pnfs_free_commit_array(new); 2196 } 2197 } 2198 2199 static void 2200 ff_layout_release_ds_info(struct pnfs_ds_commit_info *fl_cinfo, 2201 struct inode *inode) 2202 { 2203 spin_lock(&inode->i_lock); 2204 pnfs_generic_ds_cinfo_destroy(fl_cinfo); 2205 spin_unlock(&inode->i_lock); 2206 } 2207 2208 static void 2209 ff_layout_free_deviceid_node(struct nfs4_deviceid_node *d) 2210 { 2211 nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds, 2212 id_node)); 2213 } 2214 2215 static int ff_layout_encode_ioerr(struct xdr_stream *xdr, 2216 const struct nfs4_layoutreturn_args *args, 2217 const struct nfs4_flexfile_layoutreturn_args *ff_args) 2218 { 2219 __be32 *start; 2220 2221 start = xdr_reserve_space(xdr, 4); 2222 if (unlikely(!start)) 2223 return -E2BIG; 2224 2225 *start = cpu_to_be32(ff_args->num_errors); 2226 /* This assume we always return _ALL_ layouts */ 2227 return ff_layout_encode_ds_ioerr(xdr, &ff_args->errors); 2228 } 2229 2230 static void 2231 ff_layout_encode_ff_iostat_head(struct xdr_stream *xdr, 2232 const nfs4_stateid *stateid, 2233 const struct nfs42_layoutstat_devinfo *devinfo) 2234 { 2235 __be32 *p; 2236 2237 p = xdr_reserve_space(xdr, 8 + 8); 2238 p = xdr_encode_hyper(p, devinfo->offset); 2239 p = xdr_encode_hyper(p, devinfo->length); 2240 encode_opaque_fixed(xdr, stateid->data, NFS4_STATEID_SIZE); 2241 p = xdr_reserve_space(xdr, 4*8); 2242 p = xdr_encode_hyper(p, devinfo->read_count); 2243 p = xdr_encode_hyper(p, devinfo->read_bytes); 2244 p = xdr_encode_hyper(p, devinfo->write_count); 2245 p = xdr_encode_hyper(p, devinfo->write_bytes); 2246 encode_opaque_fixed(xdr, devinfo->dev_id.data, NFS4_DEVICEID4_SIZE); 2247 } 2248 2249 static void 2250 ff_layout_encode_ff_iostat(struct xdr_stream *xdr, 2251 const nfs4_stateid *stateid, 2252 const struct nfs42_layoutstat_devinfo *devinfo) 2253 { 2254 ff_layout_encode_ff_iostat_head(xdr, stateid, devinfo); 2255 ff_layout_encode_ff_layoutupdate(xdr, devinfo, 2256 devinfo->ld_private.data); 2257 } 2258 2259 /* report nothing for now */ 2260 static void ff_layout_encode_iostats_array(struct xdr_stream *xdr, 2261 const struct nfs4_layoutreturn_args *args, 2262 struct nfs4_flexfile_layoutreturn_args *ff_args) 2263 { 2264 __be32 *p; 2265 int i; 2266 2267 p = xdr_reserve_space(xdr, 4); 2268 *p = cpu_to_be32(ff_args->num_dev); 2269 for (i = 0; i < ff_args->num_dev; i++) 2270 ff_layout_encode_ff_iostat(xdr, 2271 &args->layout->plh_stateid, 2272 &ff_args->devinfo[i]); 2273 } 2274 2275 static void 2276 ff_layout_free_iostats_array(struct nfs42_layoutstat_devinfo *devinfo, 2277 unsigned int num_entries) 2278 { 2279 unsigned int i; 2280 2281 for (i = 0; i < num_entries; i++) { 2282 if (!devinfo[i].ld_private.ops) 2283 continue; 2284 if (!devinfo[i].ld_private.ops->free) 2285 continue; 2286 devinfo[i].ld_private.ops->free(&devinfo[i].ld_private); 2287 } 2288 } 2289 2290 static struct nfs4_deviceid_node * 2291 ff_layout_alloc_deviceid_node(struct nfs_server *server, 2292 struct pnfs_device *pdev, gfp_t gfp_flags) 2293 { 2294 struct nfs4_ff_layout_ds *dsaddr; 2295 2296 dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags); 2297 if (!dsaddr) 2298 return NULL; 2299 return &dsaddr->id_node; 2300 } 2301 2302 static void 2303 ff_layout_encode_layoutreturn(struct xdr_stream *xdr, 2304 const void *voidargs, 2305 const struct nfs4_xdr_opaque_data *ff_opaque) 2306 { 2307 const struct nfs4_layoutreturn_args *args = voidargs; 2308 struct nfs4_flexfile_layoutreturn_args *ff_args = ff_opaque->data; 2309 struct xdr_buf tmp_buf = { 2310 .head = { 2311 [0] = { 2312 .iov_base = page_address(ff_args->pages[0]), 2313 }, 2314 }, 2315 .buflen = PAGE_SIZE, 2316 }; 2317 struct xdr_stream tmp_xdr; 2318 __be32 *start; 2319 2320 dprintk("%s: Begin\n", __func__); 2321 2322 xdr_init_encode(&tmp_xdr, &tmp_buf, NULL, NULL); 2323 2324 ff_layout_encode_ioerr(&tmp_xdr, args, ff_args); 2325 ff_layout_encode_iostats_array(&tmp_xdr, args, ff_args); 2326 2327 start = xdr_reserve_space(xdr, 4); 2328 *start = cpu_to_be32(tmp_buf.len); 2329 xdr_write_pages(xdr, ff_args->pages, 0, tmp_buf.len); 2330 2331 dprintk("%s: Return\n", __func__); 2332 } 2333 2334 static void 2335 ff_layout_free_layoutreturn(struct nfs4_xdr_opaque_data *args) 2336 { 2337 struct nfs4_flexfile_layoutreturn_args *ff_args; 2338 2339 if (!args->data) 2340 return; 2341 ff_args = args->data; 2342 args->data = NULL; 2343 2344 ff_layout_free_ds_ioerr(&ff_args->errors); 2345 ff_layout_free_iostats_array(ff_args->devinfo, ff_args->num_dev); 2346 2347 put_page(ff_args->pages[0]); 2348 kfree(ff_args); 2349 } 2350 2351 static const struct nfs4_xdr_opaque_ops layoutreturn_ops = { 2352 .encode = ff_layout_encode_layoutreturn, 2353 .free = ff_layout_free_layoutreturn, 2354 }; 2355 2356 static int 2357 ff_layout_prepare_layoutreturn(struct nfs4_layoutreturn_args *args) 2358 { 2359 struct nfs4_flexfile_layoutreturn_args *ff_args; 2360 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(args->layout); 2361 2362 ff_args = kmalloc(sizeof(*ff_args), nfs_io_gfp_mask()); 2363 if (!ff_args) 2364 goto out_nomem; 2365 ff_args->pages[0] = alloc_page(nfs_io_gfp_mask()); 2366 if (!ff_args->pages[0]) 2367 goto out_nomem_free; 2368 2369 INIT_LIST_HEAD(&ff_args->errors); 2370 ff_args->num_errors = ff_layout_fetch_ds_ioerr(args->layout, 2371 &args->range, &ff_args->errors, 2372 FF_LAYOUTRETURN_MAXERR); 2373 2374 spin_lock(&args->inode->i_lock); 2375 ff_args->num_dev = ff_layout_mirror_prepare_stats( 2376 &ff_layout->generic_hdr, &ff_args->devinfo[0], 2377 ARRAY_SIZE(ff_args->devinfo), NFS4_FF_OP_LAYOUTRETURN); 2378 spin_unlock(&args->inode->i_lock); 2379 2380 args->ld_private->ops = &layoutreturn_ops; 2381 args->ld_private->data = ff_args; 2382 return 0; 2383 out_nomem_free: 2384 kfree(ff_args); 2385 out_nomem: 2386 return -ENOMEM; 2387 } 2388 2389 #ifdef CONFIG_NFS_V4_2 2390 void 2391 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg) 2392 { 2393 struct pnfs_layout_hdr *lo = lseg->pls_layout; 2394 struct nfs42_layout_error *errors; 2395 LIST_HEAD(head); 2396 2397 if (!nfs_server_capable(lo->plh_inode, NFS_CAP_LAYOUTERROR)) 2398 return; 2399 ff_layout_fetch_ds_ioerr(lo, &lseg->pls_range, &head, -1); 2400 if (list_empty(&head)) 2401 return; 2402 2403 errors = kmalloc_array(NFS42_LAYOUTERROR_MAX, sizeof(*errors), 2404 nfs_io_gfp_mask()); 2405 if (errors != NULL) { 2406 const struct nfs4_ff_layout_ds_err *pos; 2407 size_t n = 0; 2408 2409 list_for_each_entry(pos, &head, list) { 2410 errors[n].offset = pos->offset; 2411 errors[n].length = pos->length; 2412 nfs4_stateid_copy(&errors[n].stateid, &pos->stateid); 2413 errors[n].errors[0].dev_id = pos->deviceid; 2414 errors[n].errors[0].status = pos->status; 2415 errors[n].errors[0].opnum = pos->opnum; 2416 n++; 2417 if (!list_is_last(&pos->list, &head) && 2418 n < NFS42_LAYOUTERROR_MAX) 2419 continue; 2420 if (nfs42_proc_layouterror(lseg, errors, n) < 0) 2421 break; 2422 n = 0; 2423 } 2424 kfree(errors); 2425 } 2426 ff_layout_free_ds_ioerr(&head); 2427 } 2428 #else 2429 void 2430 ff_layout_send_layouterror(struct pnfs_layout_segment *lseg) 2431 { 2432 } 2433 #endif 2434 2435 static int 2436 ff_layout_ntop4(const struct sockaddr *sap, char *buf, const size_t buflen) 2437 { 2438 const struct sockaddr_in *sin = (struct sockaddr_in *)sap; 2439 2440 return snprintf(buf, buflen, "%pI4", &sin->sin_addr); 2441 } 2442 2443 static size_t 2444 ff_layout_ntop6_noscopeid(const struct sockaddr *sap, char *buf, 2445 const int buflen) 2446 { 2447 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sap; 2448 const struct in6_addr *addr = &sin6->sin6_addr; 2449 2450 /* 2451 * RFC 4291, Section 2.2.2 2452 * 2453 * Shorthanded ANY address 2454 */ 2455 if (ipv6_addr_any(addr)) 2456 return snprintf(buf, buflen, "::"); 2457 2458 /* 2459 * RFC 4291, Section 2.2.2 2460 * 2461 * Shorthanded loopback address 2462 */ 2463 if (ipv6_addr_loopback(addr)) 2464 return snprintf(buf, buflen, "::1"); 2465 2466 /* 2467 * RFC 4291, Section 2.2.3 2468 * 2469 * Special presentation address format for mapped v4 2470 * addresses. 2471 */ 2472 if (ipv6_addr_v4mapped(addr)) 2473 return snprintf(buf, buflen, "::ffff:%pI4", 2474 &addr->s6_addr32[3]); 2475 2476 /* 2477 * RFC 4291, Section 2.2.1 2478 */ 2479 return snprintf(buf, buflen, "%pI6c", addr); 2480 } 2481 2482 /* Derived from rpc_sockaddr2uaddr */ 2483 static void 2484 ff_layout_encode_netaddr(struct xdr_stream *xdr, struct nfs4_pnfs_ds_addr *da) 2485 { 2486 struct sockaddr *sap = (struct sockaddr *)&da->da_addr; 2487 char portbuf[RPCBIND_MAXUADDRPLEN]; 2488 char addrbuf[RPCBIND_MAXUADDRLEN]; 2489 unsigned short port; 2490 int len, netid_len; 2491 __be32 *p; 2492 2493 switch (sap->sa_family) { 2494 case AF_INET: 2495 if (ff_layout_ntop4(sap, addrbuf, sizeof(addrbuf)) == 0) 2496 return; 2497 port = ntohs(((struct sockaddr_in *)sap)->sin_port); 2498 break; 2499 case AF_INET6: 2500 if (ff_layout_ntop6_noscopeid(sap, addrbuf, sizeof(addrbuf)) == 0) 2501 return; 2502 port = ntohs(((struct sockaddr_in6 *)sap)->sin6_port); 2503 break; 2504 default: 2505 WARN_ON_ONCE(1); 2506 return; 2507 } 2508 2509 snprintf(portbuf, sizeof(portbuf), ".%u.%u", port >> 8, port & 0xff); 2510 len = strlcat(addrbuf, portbuf, sizeof(addrbuf)); 2511 2512 netid_len = strlen(da->da_netid); 2513 p = xdr_reserve_space(xdr, 4 + netid_len); 2514 xdr_encode_opaque(p, da->da_netid, netid_len); 2515 2516 p = xdr_reserve_space(xdr, 4 + len); 2517 xdr_encode_opaque(p, addrbuf, len); 2518 } 2519 2520 static void 2521 ff_layout_encode_nfstime(struct xdr_stream *xdr, 2522 ktime_t t) 2523 { 2524 struct timespec64 ts; 2525 __be32 *p; 2526 2527 p = xdr_reserve_space(xdr, 12); 2528 ts = ktime_to_timespec64(t); 2529 p = xdr_encode_hyper(p, ts.tv_sec); 2530 *p++ = cpu_to_be32(ts.tv_nsec); 2531 } 2532 2533 static void 2534 ff_layout_encode_io_latency(struct xdr_stream *xdr, 2535 struct nfs4_ff_io_stat *stat) 2536 { 2537 __be32 *p; 2538 2539 p = xdr_reserve_space(xdr, 5 * 8); 2540 p = xdr_encode_hyper(p, stat->ops_requested); 2541 p = xdr_encode_hyper(p, stat->bytes_requested); 2542 p = xdr_encode_hyper(p, stat->ops_completed); 2543 p = xdr_encode_hyper(p, stat->bytes_completed); 2544 p = xdr_encode_hyper(p, stat->bytes_not_delivered); 2545 ff_layout_encode_nfstime(xdr, stat->total_busy_time); 2546 ff_layout_encode_nfstime(xdr, stat->aggregate_completion_time); 2547 } 2548 2549 static void 2550 ff_layout_encode_ff_layoutupdate(struct xdr_stream *xdr, 2551 const struct nfs42_layoutstat_devinfo *devinfo, 2552 struct nfs4_ff_layout_mirror *mirror) 2553 { 2554 struct nfs4_pnfs_ds_addr *da; 2555 struct nfs4_pnfs_ds *ds = mirror->mirror_ds->ds; 2556 struct nfs_fh *fh = &mirror->fh_versions[0]; 2557 __be32 *p; 2558 2559 da = list_first_entry(&ds->ds_addrs, struct nfs4_pnfs_ds_addr, da_node); 2560 dprintk("%s: DS %s: encoding address %s\n", 2561 __func__, ds->ds_remotestr, da->da_remotestr); 2562 /* netaddr4 */ 2563 ff_layout_encode_netaddr(xdr, da); 2564 /* nfs_fh4 */ 2565 p = xdr_reserve_space(xdr, 4 + fh->size); 2566 xdr_encode_opaque(p, fh->data, fh->size); 2567 /* ff_io_latency4 read */ 2568 spin_lock(&mirror->lock); 2569 ff_layout_encode_io_latency(xdr, &mirror->read_stat.io_stat); 2570 /* ff_io_latency4 write */ 2571 ff_layout_encode_io_latency(xdr, &mirror->write_stat.io_stat); 2572 spin_unlock(&mirror->lock); 2573 /* nfstime4 */ 2574 ff_layout_encode_nfstime(xdr, ktime_sub(ktime_get(), mirror->start_time)); 2575 /* bool */ 2576 p = xdr_reserve_space(xdr, 4); 2577 *p = cpu_to_be32(false); 2578 } 2579 2580 static void 2581 ff_layout_encode_layoutstats(struct xdr_stream *xdr, const void *args, 2582 const struct nfs4_xdr_opaque_data *opaque) 2583 { 2584 struct nfs42_layoutstat_devinfo *devinfo = container_of(opaque, 2585 struct nfs42_layoutstat_devinfo, ld_private); 2586 __be32 *start; 2587 2588 /* layoutupdate length */ 2589 start = xdr_reserve_space(xdr, 4); 2590 ff_layout_encode_ff_layoutupdate(xdr, devinfo, opaque->data); 2591 2592 *start = cpu_to_be32((xdr->p - start - 1) * 4); 2593 } 2594 2595 static void 2596 ff_layout_free_layoutstats(struct nfs4_xdr_opaque_data *opaque) 2597 { 2598 struct nfs4_ff_layout_mirror *mirror = opaque->data; 2599 2600 ff_layout_put_mirror(mirror); 2601 } 2602 2603 static const struct nfs4_xdr_opaque_ops layoutstat_ops = { 2604 .encode = ff_layout_encode_layoutstats, 2605 .free = ff_layout_free_layoutstats, 2606 }; 2607 2608 static int 2609 ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo, 2610 struct nfs42_layoutstat_devinfo *devinfo, 2611 int dev_limit, enum nfs4_ff_op_type type) 2612 { 2613 struct nfs4_flexfile_layout *ff_layout = FF_LAYOUT_FROM_HDR(lo); 2614 struct nfs4_ff_layout_mirror *mirror; 2615 struct nfs4_deviceid_node *dev; 2616 int i = 0; 2617 2618 list_for_each_entry(mirror, &ff_layout->mirrors, mirrors) { 2619 if (i >= dev_limit) 2620 break; 2621 if (IS_ERR_OR_NULL(mirror->mirror_ds)) 2622 continue; 2623 if (!test_and_clear_bit(NFS4_FF_MIRROR_STAT_AVAIL, 2624 &mirror->flags) && 2625 type != NFS4_FF_OP_LAYOUTRETURN) 2626 continue; 2627 /* mirror refcount put in cleanup_layoutstats */ 2628 if (!refcount_inc_not_zero(&mirror->ref)) 2629 continue; 2630 dev = &mirror->mirror_ds->id_node; 2631 memcpy(&devinfo->dev_id, &dev->deviceid, NFS4_DEVICEID4_SIZE); 2632 devinfo->offset = 0; 2633 devinfo->length = NFS4_MAX_UINT64; 2634 spin_lock(&mirror->lock); 2635 devinfo->read_count = mirror->read_stat.io_stat.ops_completed; 2636 devinfo->read_bytes = mirror->read_stat.io_stat.bytes_completed; 2637 devinfo->write_count = mirror->write_stat.io_stat.ops_completed; 2638 devinfo->write_bytes = mirror->write_stat.io_stat.bytes_completed; 2639 spin_unlock(&mirror->lock); 2640 devinfo->layout_type = LAYOUT_FLEX_FILES; 2641 devinfo->ld_private.ops = &layoutstat_ops; 2642 devinfo->ld_private.data = mirror; 2643 2644 devinfo++; 2645 i++; 2646 } 2647 return i; 2648 } 2649 2650 static int ff_layout_prepare_layoutstats(struct nfs42_layoutstat_args *args) 2651 { 2652 struct pnfs_layout_hdr *lo; 2653 struct nfs4_flexfile_layout *ff_layout; 2654 const int dev_count = PNFS_LAYOUTSTATS_MAXDEV; 2655 2656 /* For now, send at most PNFS_LAYOUTSTATS_MAXDEV statistics */ 2657 args->devinfo = kmalloc_array(dev_count, sizeof(*args->devinfo), 2658 nfs_io_gfp_mask()); 2659 if (!args->devinfo) 2660 return -ENOMEM; 2661 2662 spin_lock(&args->inode->i_lock); 2663 lo = NFS_I(args->inode)->layout; 2664 if (lo && pnfs_layout_is_valid(lo)) { 2665 ff_layout = FF_LAYOUT_FROM_HDR(lo); 2666 args->num_dev = ff_layout_mirror_prepare_stats( 2667 &ff_layout->generic_hdr, &args->devinfo[0], dev_count, 2668 NFS4_FF_OP_LAYOUTSTATS); 2669 } else 2670 args->num_dev = 0; 2671 spin_unlock(&args->inode->i_lock); 2672 if (!args->num_dev) { 2673 kfree(args->devinfo); 2674 args->devinfo = NULL; 2675 return -ENOENT; 2676 } 2677 2678 return 0; 2679 } 2680 2681 static int 2682 ff_layout_set_layoutdriver(struct nfs_server *server, 2683 const struct nfs_fh *dummy) 2684 { 2685 #if IS_ENABLED(CONFIG_NFS_V4_2) 2686 server->caps |= NFS_CAP_LAYOUTSTATS | NFS_CAP_REBOOT_LAYOUTRETURN; 2687 #endif 2688 return 0; 2689 } 2690 2691 static const struct pnfs_commit_ops ff_layout_commit_ops = { 2692 .setup_ds_info = ff_layout_setup_ds_info, 2693 .release_ds_info = ff_layout_release_ds_info, 2694 .mark_request_commit = pnfs_layout_mark_request_commit, 2695 .clear_request_commit = pnfs_generic_clear_request_commit, 2696 .scan_commit_lists = pnfs_generic_scan_commit_lists, 2697 .recover_commit_reqs = pnfs_generic_recover_commit_reqs, 2698 .commit_pagelist = ff_layout_commit_pagelist, 2699 }; 2700 2701 static struct pnfs_layoutdriver_type flexfilelayout_type = { 2702 .id = LAYOUT_FLEX_FILES, 2703 .name = "LAYOUT_FLEX_FILES", 2704 .owner = THIS_MODULE, 2705 .flags = PNFS_LAYOUTGET_ON_OPEN, 2706 .max_layoutget_response = 4096, /* 1 page or so... */ 2707 .set_layoutdriver = ff_layout_set_layoutdriver, 2708 .alloc_layout_hdr = ff_layout_alloc_layout_hdr, 2709 .free_layout_hdr = ff_layout_free_layout_hdr, 2710 .alloc_lseg = ff_layout_alloc_lseg, 2711 .free_lseg = ff_layout_free_lseg, 2712 .add_lseg = ff_layout_add_lseg, 2713 .pg_read_ops = &ff_layout_pg_read_ops, 2714 .pg_write_ops = &ff_layout_pg_write_ops, 2715 .get_ds_info = ff_layout_get_ds_info, 2716 .free_deviceid_node = ff_layout_free_deviceid_node, 2717 .read_pagelist = ff_layout_read_pagelist, 2718 .write_pagelist = ff_layout_write_pagelist, 2719 .alloc_deviceid_node = ff_layout_alloc_deviceid_node, 2720 .prepare_layoutreturn = ff_layout_prepare_layoutreturn, 2721 .sync = pnfs_nfs_generic_sync, 2722 .prepare_layoutstats = ff_layout_prepare_layoutstats, 2723 .cancel_io = ff_layout_cancel_io, 2724 }; 2725 2726 static int __init nfs4flexfilelayout_init(void) 2727 { 2728 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n", 2729 __func__); 2730 return pnfs_register_layoutdriver(&flexfilelayout_type); 2731 } 2732 2733 static void __exit nfs4flexfilelayout_exit(void) 2734 { 2735 printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n", 2736 __func__); 2737 pnfs_unregister_layoutdriver(&flexfilelayout_type); 2738 } 2739 2740 MODULE_ALIAS("nfs-layouttype4-4"); 2741 2742 MODULE_LICENSE("GPL"); 2743 MODULE_DESCRIPTION("The NFSv4 flexfile layout driver"); 2744 2745 module_init(nfs4flexfilelayout_init); 2746 module_exit(nfs4flexfilelayout_exit); 2747 2748 module_param(io_maxretrans, ushort, 0644); 2749 MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client " 2750 "retries an I/O request before returning an error. "); 2751