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