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