1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2014-2016 Christoph Hellwig. 4 */ 5 #include <linux/sunrpc/svc.h> 6 #include <linux/blkdev.h> 7 #include <linux/fs_struct.h> 8 #include <linux/nfs4.h> 9 #include <linux/nfs_fs.h> 10 #include <linux/nfs_xdr.h> 11 #include <linux/pr.h> 12 13 #include "blocklayout.h" 14 #include "../nfs4trace.h" 15 16 #define NFSDBG_FACILITY NFSDBG_PNFS_LD 17 18 #ifdef CONFIG_PNFS_SCSI_LAYOUT 19 static void bl_unregister_scsi(struct pnfs_block_dev *dev) 20 { 21 struct block_device *bdev = file_bdev(dev->bdev_file); 22 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 23 int status; 24 25 status = ops->pr_register(bdev, dev->pr_key, 0, false); 26 if (status) 27 trace_bl_pr_key_unreg_err(bdev, dev->pr_key, status); 28 else 29 trace_bl_pr_key_unreg(bdev, dev->pr_key); 30 } 31 32 static bool bl_register_scsi(struct pnfs_block_dev *dev) 33 { 34 struct block_device *bdev = file_bdev(dev->bdev_file); 35 const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops; 36 int status; 37 38 if (test_and_set_bit(PNFS_BDEV_REGISTERED, &dev->flags)) 39 return true; 40 41 status = ops->pr_register(bdev, 0, dev->pr_key, true); 42 if (status) { 43 trace_bl_pr_key_reg_err(bdev, dev->pr_key, status); 44 return false; 45 } 46 trace_bl_pr_key_reg(bdev, dev->pr_key); 47 return true; 48 } 49 #else 50 static void bl_unregister_scsi(struct pnfs_block_dev *dev) 51 { 52 } 53 54 static bool bl_register_scsi(struct pnfs_block_dev *dev) 55 { 56 return false; 57 } 58 #endif /* CONFIG_PNFS_SCSI_LAYOUT */ 59 60 static void bl_unregister_dev(struct pnfs_block_dev *dev) 61 { 62 u32 i; 63 64 if (dev->nr_children) { 65 for (i = 0; i < dev->nr_children; i++) 66 bl_unregister_dev(&dev->children[i]); 67 return; 68 } 69 70 if (dev->type == PNFS_BLOCK_VOLUME_SCSI && 71 test_and_clear_bit(PNFS_BDEV_REGISTERED, &dev->flags)) 72 bl_unregister_scsi(dev); 73 } 74 75 bool bl_register_dev(struct pnfs_block_dev *dev) 76 { 77 u32 i; 78 79 if (dev->nr_children) { 80 for (i = 0; i < dev->nr_children; i++) { 81 if (!bl_register_dev(&dev->children[i])) { 82 while (i > 0) 83 bl_unregister_dev(&dev->children[--i]); 84 return false; 85 } 86 } 87 return true; 88 } 89 90 if (dev->type == PNFS_BLOCK_VOLUME_SCSI) 91 return bl_register_scsi(dev); 92 return true; 93 } 94 95 static void 96 bl_free_device(struct pnfs_block_dev *dev) 97 { 98 bl_unregister_dev(dev); 99 100 if (dev->children) { 101 int i; 102 103 for (i = 0; i < dev->nr_children; i++) 104 bl_free_device(&dev->children[i]); 105 kfree(dev->children); 106 dev->children = NULL; 107 dev->nr_children = 0; 108 } else if (dev->bdev_file) { 109 fput(dev->bdev_file); 110 dev->bdev_file = NULL; 111 } 112 } 113 114 void 115 bl_free_deviceid_node(struct nfs4_deviceid_node *d) 116 { 117 struct pnfs_block_dev *dev = 118 container_of(d, struct pnfs_block_dev, node); 119 120 bl_free_device(dev); 121 kfree_rcu(dev, node.rcu); 122 } 123 124 static int 125 nfs4_block_decode_volume(struct xdr_stream *xdr, struct pnfs_block_volume *b) 126 { 127 __be32 *p; 128 int i; 129 130 p = xdr_inline_decode(xdr, 4); 131 if (!p) 132 return -EIO; 133 b->type = be32_to_cpup(p++); 134 135 switch (b->type) { 136 case PNFS_BLOCK_VOLUME_SIMPLE: 137 p = xdr_inline_decode(xdr, 4); 138 if (!p) 139 return -EIO; 140 b->simple.nr_sigs = be32_to_cpup(p++); 141 if (!b->simple.nr_sigs || b->simple.nr_sigs > PNFS_BLOCK_MAX_UUIDS) { 142 dprintk("Bad signature count: %d\n", b->simple.nr_sigs); 143 return -EIO; 144 } 145 146 b->simple.len = 4 + 4; 147 for (i = 0; i < b->simple.nr_sigs; i++) { 148 p = xdr_inline_decode(xdr, 8 + 4); 149 if (!p) 150 return -EIO; 151 p = xdr_decode_hyper(p, &b->simple.sigs[i].offset); 152 b->simple.sigs[i].sig_len = be32_to_cpup(p++); 153 if (b->simple.sigs[i].sig_len > PNFS_BLOCK_UUID_LEN) { 154 pr_info("signature too long: %d\n", 155 b->simple.sigs[i].sig_len); 156 return -EIO; 157 } 158 159 p = xdr_inline_decode(xdr, b->simple.sigs[i].sig_len); 160 if (!p) 161 return -EIO; 162 memcpy(&b->simple.sigs[i].sig, p, 163 b->simple.sigs[i].sig_len); 164 165 b->simple.len += 8 + 4 + \ 166 (XDR_QUADLEN(b->simple.sigs[i].sig_len) << 2); 167 } 168 break; 169 case PNFS_BLOCK_VOLUME_SLICE: 170 p = xdr_inline_decode(xdr, 8 + 8 + 4); 171 if (!p) 172 return -EIO; 173 p = xdr_decode_hyper(p, &b->slice.start); 174 p = xdr_decode_hyper(p, &b->slice.len); 175 b->slice.volume = be32_to_cpup(p++); 176 break; 177 case PNFS_BLOCK_VOLUME_CONCAT: 178 p = xdr_inline_decode(xdr, 4); 179 if (!p) 180 return -EIO; 181 182 b->concat.volumes_count = be32_to_cpup(p++); 183 if (b->concat.volumes_count > PNFS_BLOCK_MAX_DEVICES) { 184 dprintk("Too many volumes: %d\n", b->concat.volumes_count); 185 return -EIO; 186 } 187 188 p = xdr_inline_decode(xdr, b->concat.volumes_count * 4); 189 if (!p) 190 return -EIO; 191 for (i = 0; i < b->concat.volumes_count; i++) 192 b->concat.volumes[i] = be32_to_cpup(p++); 193 break; 194 case PNFS_BLOCK_VOLUME_STRIPE: 195 p = xdr_inline_decode(xdr, 8 + 4); 196 if (!p) 197 return -EIO; 198 199 p = xdr_decode_hyper(p, &b->stripe.chunk_size); 200 b->stripe.volumes_count = be32_to_cpup(p++); 201 if (b->stripe.volumes_count > PNFS_BLOCK_MAX_DEVICES) { 202 dprintk("Too many volumes: %d\n", b->stripe.volumes_count); 203 return -EIO; 204 } 205 206 p = xdr_inline_decode(xdr, b->stripe.volumes_count * 4); 207 if (!p) 208 return -EIO; 209 for (i = 0; i < b->stripe.volumes_count; i++) 210 b->stripe.volumes[i] = be32_to_cpup(p++); 211 break; 212 case PNFS_BLOCK_VOLUME_SCSI: 213 p = xdr_inline_decode(xdr, 4 + 4 + 4); 214 if (!p) 215 return -EIO; 216 b->scsi.code_set = be32_to_cpup(p++); 217 b->scsi.designator_type = be32_to_cpup(p++); 218 b->scsi.designator_len = be32_to_cpup(p++); 219 p = xdr_inline_decode(xdr, b->scsi.designator_len); 220 if (!p) 221 return -EIO; 222 if (b->scsi.designator_len > 256) 223 return -EIO; 224 memcpy(&b->scsi.designator, p, b->scsi.designator_len); 225 p = xdr_inline_decode(xdr, 8); 226 if (!p) 227 return -EIO; 228 p = xdr_decode_hyper(p, &b->scsi.pr_key); 229 break; 230 default: 231 dprintk("unknown volume type!\n"); 232 return -EIO; 233 } 234 235 return 0; 236 } 237 238 static bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset, 239 struct pnfs_block_dev_map *map) 240 { 241 map->start = dev->start; 242 map->len = dev->len; 243 map->disk_offset = dev->disk_offset; 244 map->bdev = file_bdev(dev->bdev_file); 245 return true; 246 } 247 248 static bool bl_map_concat(struct pnfs_block_dev *dev, u64 offset, 249 struct pnfs_block_dev_map *map) 250 { 251 int i; 252 253 for (i = 0; i < dev->nr_children; i++) { 254 struct pnfs_block_dev *child = &dev->children[i]; 255 256 if (child->start > offset || 257 child->start + child->len <= offset) 258 continue; 259 260 child->map(child, offset - child->start, map); 261 return true; 262 } 263 264 dprintk("%s: ran off loop!\n", __func__); 265 return false; 266 } 267 268 static bool bl_map_stripe(struct pnfs_block_dev *dev, u64 offset, 269 struct pnfs_block_dev_map *map) 270 { 271 struct pnfs_block_dev *child; 272 u64 chunk; 273 u32 chunk_idx; 274 u64 disk_chunk; 275 u64 disk_offset; 276 277 chunk = div_u64(offset, dev->chunk_size); 278 disk_chunk = div_u64_rem(chunk, dev->nr_children, &chunk_idx); 279 280 if (chunk_idx >= dev->nr_children) { 281 dprintk("%s: invalid chunk idx %d (%lld/%lld)\n", 282 __func__, chunk_idx, offset, dev->chunk_size); 283 /* error, should not happen */ 284 return false; 285 } 286 287 /* truncate offset to the beginning of the stripe */ 288 offset = chunk * dev->chunk_size; 289 290 /* disk offset of the stripe */ 291 disk_offset = disk_chunk * dev->chunk_size; 292 293 child = &dev->children[chunk_idx]; 294 child->map(child, disk_offset, map); 295 296 map->start += offset; 297 map->disk_offset += disk_offset; 298 map->len = dev->chunk_size; 299 return true; 300 } 301 302 static int 303 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d, 304 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask); 305 306 #ifdef CONFIG_PNFS_BLOCK_LAYOUT 307 static int 308 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d, 309 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 310 { 311 struct pnfs_block_volume *v = &volumes[idx]; 312 struct file *bdev_file; 313 dev_t dev; 314 315 dev = bl_resolve_deviceid(server, v, gfp_mask); 316 if (!dev) 317 return -EIO; 318 319 bdev_file = bdev_file_open_by_dev(dev, BLK_OPEN_READ | BLK_OPEN_WRITE, 320 NULL, NULL); 321 if (IS_ERR(bdev_file)) { 322 printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n", 323 MAJOR(dev), MINOR(dev), PTR_ERR(bdev_file)); 324 return PTR_ERR(bdev_file); 325 } 326 d->bdev_file = bdev_file; 327 d->len = bdev_nr_bytes(file_bdev(bdev_file)); 328 d->map = bl_map_simple; 329 330 printk(KERN_INFO "pNFS: using block device %s\n", 331 file_bdev(bdev_file)->bd_disk->disk_name); 332 return 0; 333 } 334 #else 335 static int 336 bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d, 337 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 338 { 339 dprintk("unsupported volume type: %d\n", PNFS_BLOCK_VOLUME_SIMPLE); 340 return -EIO; 341 } 342 #endif /* CONFIG_PNFS_BLOCK_LAYOUT */ 343 344 #ifdef CONFIG_PNFS_SCSI_LAYOUT 345 static bool 346 bl_validate_designator(struct pnfs_block_volume *v) 347 { 348 switch (v->scsi.designator_type) { 349 case PS_DESIGNATOR_EUI64: 350 if (v->scsi.code_set != PS_CODE_SET_BINARY) 351 return false; 352 353 if (v->scsi.designator_len != 8 && 354 v->scsi.designator_len != 10 && 355 v->scsi.designator_len != 16) 356 return false; 357 358 return true; 359 case PS_DESIGNATOR_NAA: 360 if (v->scsi.code_set != PS_CODE_SET_BINARY) 361 return false; 362 363 if (v->scsi.designator_len != 8 && 364 v->scsi.designator_len != 16) 365 return false; 366 367 return true; 368 case PS_DESIGNATOR_T10: 369 case PS_DESIGNATOR_NAME: 370 pr_err("pNFS: unsupported designator " 371 "(code set %d, type %d, len %d.\n", 372 v->scsi.code_set, 373 v->scsi.designator_type, 374 v->scsi.designator_len); 375 return false; 376 default: 377 pr_err("pNFS: invalid designator " 378 "(code set %d, type %d, len %d.\n", 379 v->scsi.code_set, 380 v->scsi.designator_type, 381 v->scsi.designator_len); 382 return false; 383 } 384 } 385 386 static struct file * 387 bl_open_path(struct pnfs_block_volume *v, const char *prefix) 388 { 389 struct file *bdev_file; 390 const char *devname __free(kfree) = NULL; 391 392 devname = kasprintf(GFP_KERNEL, "/dev/disk/by-id/%s%*phN", 393 prefix, v->scsi.designator_len, v->scsi.designator); 394 if (!devname) 395 return ERR_PTR(-ENOMEM); 396 397 if (tsk_is_kthread(current)) { 398 scoped_with_init_fs() 399 bdev_file = bdev_file_open_by_path(devname, 400 BLK_OPEN_READ | BLK_OPEN_WRITE, 401 NULL, NULL); 402 } else { 403 bdev_file = bdev_file_open_by_path(devname, 404 BLK_OPEN_READ | BLK_OPEN_WRITE, NULL, NULL); 405 } 406 if (IS_ERR(bdev_file)) { 407 dprintk("failed to open device %s (%ld)\n", 408 devname, PTR_ERR(bdev_file)); 409 } else { 410 pr_info("pNFS: using block device %s\n", 411 file_bdev(bdev_file)->bd_disk->disk_name); 412 } 413 414 return bdev_file; 415 } 416 417 static int 418 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d, 419 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 420 { 421 struct pnfs_block_volume *v = &volumes[idx]; 422 struct block_device *bdev; 423 const struct pr_ops *ops; 424 struct file *bdev_file; 425 int error; 426 427 if (!bl_validate_designator(v)) 428 return -EINVAL; 429 430 /* 431 * Try to open the RH/Fedora specific dm-mpath udev path first, as the 432 * wwn- links will only point to the first discovered SCSI device there. 433 * On other distributions like Debian, the default SCSI by-id path will 434 * point to the dm-multipath device if one exists. 435 */ 436 bdev_file = bl_open_path(v, "dm-uuid-mpath-0x"); 437 if (IS_ERR(bdev_file)) 438 bdev_file = bl_open_path(v, "wwn-0x"); 439 if (IS_ERR(bdev_file)) 440 bdev_file = bl_open_path(v, "nvme-eui."); 441 if (IS_ERR(bdev_file)) { 442 pr_warn("pNFS: no device found for volume %*phN\n", 443 v->scsi.designator_len, v->scsi.designator); 444 return PTR_ERR(bdev_file); 445 } 446 d->bdev_file = bdev_file; 447 bdev = file_bdev(bdev_file); 448 449 d->len = bdev_nr_bytes(bdev); 450 d->map = bl_map_simple; 451 d->pr_key = v->scsi.pr_key; 452 453 if (d->len == 0) { 454 error = -ENODEV; 455 goto out_blkdev_put; 456 } 457 458 ops = bdev->bd_disk->fops->pr_ops; 459 if (!ops) { 460 pr_err("pNFS: block device %s does not support reservations.", 461 bdev->bd_disk->disk_name); 462 error = -EINVAL; 463 goto out_blkdev_put; 464 } 465 466 return 0; 467 468 out_blkdev_put: 469 fput(d->bdev_file); 470 d->bdev_file = NULL; 471 return error; 472 } 473 #else 474 static int 475 bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d, 476 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 477 { 478 dprintk("unsupported volume type: %d\n", PNFS_BLOCK_VOLUME_SCSI); 479 return -EIO; 480 } 481 #endif /* CONFIG_PNFS_SCSI_LAYOUT */ 482 483 static int 484 bl_parse_slice(struct nfs_server *server, struct pnfs_block_dev *d, 485 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 486 { 487 struct pnfs_block_volume *v = &volumes[idx]; 488 int ret; 489 490 ret = bl_parse_deviceid(server, d, volumes, v->slice.volume, gfp_mask); 491 if (ret) 492 return ret; 493 494 d->disk_offset = v->slice.start; 495 d->len = v->slice.len; 496 return 0; 497 } 498 499 static int 500 bl_parse_concat(struct nfs_server *server, struct pnfs_block_dev *d, 501 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 502 { 503 struct pnfs_block_volume *v = &volumes[idx]; 504 u64 len = 0; 505 int ret, i; 506 507 d->children = kzalloc_objs(struct pnfs_block_dev, 508 v->concat.volumes_count, gfp_mask); 509 if (!d->children) 510 return -ENOMEM; 511 512 for (i = 0; i < v->concat.volumes_count; i++) { 513 ret = bl_parse_deviceid(server, &d->children[i], 514 volumes, v->concat.volumes[i], gfp_mask); 515 if (ret) { 516 bl_free_device(&d->children[i]); 517 bl_free_device(d); 518 return ret; 519 } 520 521 d->nr_children++; 522 d->children[i].start += len; 523 len += d->children[i].len; 524 } 525 526 d->len = len; 527 d->map = bl_map_concat; 528 return 0; 529 } 530 531 static int 532 bl_parse_stripe(struct nfs_server *server, struct pnfs_block_dev *d, 533 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 534 { 535 struct pnfs_block_volume *v = &volumes[idx]; 536 u64 len = 0; 537 int ret, i; 538 539 d->children = kzalloc_objs(struct pnfs_block_dev, 540 v->stripe.volumes_count, gfp_mask); 541 if (!d->children) 542 return -ENOMEM; 543 544 for (i = 0; i < v->stripe.volumes_count; i++) { 545 ret = bl_parse_deviceid(server, &d->children[i], 546 volumes, v->stripe.volumes[i], gfp_mask); 547 if (ret) { 548 bl_free_device(&d->children[i]); 549 bl_free_device(d); 550 return ret; 551 } 552 553 d->nr_children++; 554 len += d->children[i].len; 555 } 556 557 d->len = len; 558 d->chunk_size = v->stripe.chunk_size; 559 d->map = bl_map_stripe; 560 return 0; 561 } 562 563 static int 564 bl_parse_deviceid(struct nfs_server *server, struct pnfs_block_dev *d, 565 struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask) 566 { 567 d->type = volumes[idx].type; 568 569 switch (d->type) { 570 case PNFS_BLOCK_VOLUME_SIMPLE: 571 return bl_parse_simple(server, d, volumes, idx, gfp_mask); 572 case PNFS_BLOCK_VOLUME_SLICE: 573 return bl_parse_slice(server, d, volumes, idx, gfp_mask); 574 case PNFS_BLOCK_VOLUME_CONCAT: 575 return bl_parse_concat(server, d, volumes, idx, gfp_mask); 576 case PNFS_BLOCK_VOLUME_STRIPE: 577 return bl_parse_stripe(server, d, volumes, idx, gfp_mask); 578 case PNFS_BLOCK_VOLUME_SCSI: 579 return bl_parse_scsi(server, d, volumes, idx, gfp_mask); 580 default: 581 dprintk("unsupported volume type: %d\n", d->type); 582 return -EIO; 583 } 584 } 585 586 struct nfs4_deviceid_node * 587 bl_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev, 588 gfp_t gfp_mask) 589 { 590 struct nfs4_deviceid_node *node = NULL; 591 struct pnfs_block_volume *volumes; 592 struct pnfs_block_dev *top; 593 struct xdr_stream xdr; 594 struct xdr_buf buf; 595 struct folio *scratch; 596 int nr_volumes, ret, i; 597 __be32 *p; 598 599 scratch = folio_alloc(gfp_mask, 0); 600 if (!scratch) 601 goto out; 602 603 xdr_init_decode_pages(&xdr, &buf, pdev->pages, pdev->pglen); 604 xdr_set_scratch_folio(&xdr, scratch); 605 606 p = xdr_inline_decode(&xdr, sizeof(__be32)); 607 if (!p) 608 goto out_free_scratch; 609 nr_volumes = be32_to_cpup(p++); 610 611 volumes = kzalloc_objs(struct pnfs_block_volume, nr_volumes, gfp_mask); 612 if (!volumes) 613 goto out_free_scratch; 614 615 for (i = 0; i < nr_volumes; i++) { 616 ret = nfs4_block_decode_volume(&xdr, &volumes[i]); 617 if (ret < 0) 618 goto out_free_volumes; 619 } 620 621 top = kzalloc_obj(*top, gfp_mask); 622 if (!top) 623 goto out_free_volumes; 624 625 ret = bl_parse_deviceid(server, top, volumes, nr_volumes - 1, gfp_mask); 626 627 node = &top->node; 628 nfs4_init_deviceid_node(node, server, &pdev->dev_id); 629 if (ret) 630 nfs4_mark_deviceid_unavailable(node); 631 632 out_free_volumes: 633 kfree(volumes); 634 out_free_scratch: 635 folio_put(scratch); 636 out: 637 return node; 638 } 639