1 /* 2 * linux/fs/nfs/blocklayout/blocklayout.c 3 * 4 * Module for the NFSv4.1 pNFS block layout driver. 5 * 6 * Copyright (c) 2006 The Regents of the University of Michigan. 7 * All rights reserved. 8 * 9 * Andy Adamson <andros@citi.umich.edu> 10 * Fred Isaman <iisaman@umich.edu> 11 * 12 * permission is granted to use, copy, create derivative works and 13 * redistribute this software and such derivative works for any purpose, 14 * so long as the name of the university of michigan is not used in 15 * any advertising or publicity pertaining to the use or distribution 16 * of this software without specific, written prior authorization. if 17 * the above copyright notice or any other identification of the 18 * university of michigan is included in any copy of any portion of 19 * this software, then the disclaimer below must also be included. 20 * 21 * this software is provided as is, without representation from the 22 * university of michigan as to its fitness for any purpose, and without 23 * warranty by the university of michigan of any kind, either express 24 * or implied, including without limitation the implied warranties of 25 * merchantability and fitness for a particular purpose. the regents 26 * of the university of michigan shall not be liable for any damages, 27 * including special, indirect, incidental, or consequential damages, 28 * with respect to any claim arising out or in connection with the use 29 * of the software, even if it has been or is hereafter advised of the 30 * possibility of such damages. 31 */ 32 33 #include <linux/module.h> 34 #include <linux/init.h> 35 #include <linux/mount.h> 36 #include <linux/namei.h> 37 #include <linux/bio.h> /* struct bio */ 38 #include <linux/prefetch.h> 39 #include <linux/pagevec.h> 40 41 #include "../pnfs.h" 42 #include "../nfs4session.h" 43 #include "../internal.h" 44 #include "blocklayout.h" 45 46 #define NFSDBG_FACILITY NFSDBG_PNFS_LD 47 48 MODULE_LICENSE("GPL"); 49 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>"); 50 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver"); 51 52 static bool is_hole(struct pnfs_block_extent *be) 53 { 54 switch (be->be_state) { 55 case PNFS_BLOCK_NONE_DATA: 56 return true; 57 case PNFS_BLOCK_INVALID_DATA: 58 return be->be_tag ? false : true; 59 default: 60 return false; 61 } 62 } 63 64 /* The data we are handed might be spread across several bios. We need 65 * to track when the last one is finished. 66 */ 67 struct parallel_io { 68 struct kref refcnt; 69 void (*pnfs_callback) (void *data); 70 void *data; 71 }; 72 73 static inline struct parallel_io *alloc_parallel(void *data) 74 { 75 struct parallel_io *rv; 76 77 rv = kmalloc(sizeof(*rv), GFP_NOFS); 78 if (rv) { 79 rv->data = data; 80 kref_init(&rv->refcnt); 81 } 82 return rv; 83 } 84 85 static inline void get_parallel(struct parallel_io *p) 86 { 87 kref_get(&p->refcnt); 88 } 89 90 static void destroy_parallel(struct kref *kref) 91 { 92 struct parallel_io *p = container_of(kref, struct parallel_io, refcnt); 93 94 dprintk("%s enter\n", __func__); 95 p->pnfs_callback(p->data); 96 kfree(p); 97 } 98 99 static inline void put_parallel(struct parallel_io *p) 100 { 101 kref_put(&p->refcnt, destroy_parallel); 102 } 103 104 static struct bio * 105 bl_submit_bio(struct bio *bio) 106 { 107 if (bio) { 108 get_parallel(bio->bi_private); 109 dprintk("%s submitting %s bio %u@%llu\n", __func__, 110 bio_op(bio) == READ ? "read" : "write", 111 bio->bi_iter.bi_size, 112 (unsigned long long)bio->bi_iter.bi_sector); 113 submit_bio(bio); 114 } 115 return NULL; 116 } 117 118 static struct bio *bl_alloc_init_bio(unsigned int npg, 119 struct block_device *bdev, sector_t disk_sector, 120 bio_end_io_t end_io, struct parallel_io *par) 121 { 122 struct bio *bio; 123 124 npg = bio_max_segs(npg); 125 bio = bio_alloc(GFP_NOIO, npg); 126 if (bio) { 127 bio->bi_iter.bi_sector = disk_sector; 128 bio_set_dev(bio, bdev); 129 bio->bi_end_io = end_io; 130 bio->bi_private = par; 131 } 132 return bio; 133 } 134 135 static bool offset_in_map(u64 offset, struct pnfs_block_dev_map *map) 136 { 137 return offset >= map->start && offset < map->start + map->len; 138 } 139 140 static struct bio * 141 do_add_page_to_bio(struct bio *bio, int npg, int rw, sector_t isect, 142 struct page *page, struct pnfs_block_dev_map *map, 143 struct pnfs_block_extent *be, bio_end_io_t end_io, 144 struct parallel_io *par, unsigned int offset, int *len) 145 { 146 struct pnfs_block_dev *dev = 147 container_of(be->be_device, struct pnfs_block_dev, node); 148 u64 disk_addr, end; 149 150 dprintk("%s: npg %d rw %d isect %llu offset %u len %d\n", __func__, 151 npg, rw, (unsigned long long)isect, offset, *len); 152 153 /* translate to device offset */ 154 isect += be->be_v_offset; 155 isect -= be->be_f_offset; 156 157 /* translate to physical disk offset */ 158 disk_addr = (u64)isect << SECTOR_SHIFT; 159 if (!offset_in_map(disk_addr, map)) { 160 if (!dev->map(dev, disk_addr, map) || !offset_in_map(disk_addr, map)) 161 return ERR_PTR(-EIO); 162 bio = bl_submit_bio(bio); 163 } 164 disk_addr += map->disk_offset; 165 disk_addr -= map->start; 166 167 /* limit length to what the device mapping allows */ 168 end = disk_addr + *len; 169 if (end >= map->start + map->len) 170 *len = map->start + map->len - disk_addr; 171 172 retry: 173 if (!bio) { 174 bio = bl_alloc_init_bio(npg, map->bdev, 175 disk_addr >> SECTOR_SHIFT, end_io, par); 176 if (!bio) 177 return ERR_PTR(-ENOMEM); 178 bio_set_op_attrs(bio, rw, 0); 179 } 180 if (bio_add_page(bio, page, *len, offset) < *len) { 181 bio = bl_submit_bio(bio); 182 goto retry; 183 } 184 return bio; 185 } 186 187 static void bl_mark_devices_unavailable(struct nfs_pgio_header *header, bool rw) 188 { 189 struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); 190 size_t bytes_left = header->args.count; 191 sector_t isect, extent_length = 0; 192 struct pnfs_block_extent be; 193 194 isect = header->args.offset >> SECTOR_SHIFT; 195 bytes_left += header->args.offset - (isect << SECTOR_SHIFT); 196 197 while (bytes_left > 0) { 198 if (!ext_tree_lookup(bl, isect, &be, rw)) 199 return; 200 extent_length = be.be_length - (isect - be.be_f_offset); 201 nfs4_mark_deviceid_unavailable(be.be_device); 202 isect += extent_length; 203 if (bytes_left > extent_length << SECTOR_SHIFT) 204 bytes_left -= extent_length << SECTOR_SHIFT; 205 else 206 bytes_left = 0; 207 } 208 } 209 210 static void bl_end_io_read(struct bio *bio) 211 { 212 struct parallel_io *par = bio->bi_private; 213 214 if (bio->bi_status) { 215 struct nfs_pgio_header *header = par->data; 216 217 if (!header->pnfs_error) 218 header->pnfs_error = -EIO; 219 pnfs_set_lo_fail(header->lseg); 220 bl_mark_devices_unavailable(header, false); 221 } 222 223 bio_put(bio); 224 put_parallel(par); 225 } 226 227 static void bl_read_cleanup(struct work_struct *work) 228 { 229 struct rpc_task *task; 230 struct nfs_pgio_header *hdr; 231 dprintk("%s enter\n", __func__); 232 task = container_of(work, struct rpc_task, u.tk_work); 233 hdr = container_of(task, struct nfs_pgio_header, task); 234 pnfs_ld_read_done(hdr); 235 } 236 237 static void 238 bl_end_par_io_read(void *data) 239 { 240 struct nfs_pgio_header *hdr = data; 241 242 hdr->task.tk_status = hdr->pnfs_error; 243 INIT_WORK(&hdr->task.u.tk_work, bl_read_cleanup); 244 schedule_work(&hdr->task.u.tk_work); 245 } 246 247 static enum pnfs_try_status 248 bl_read_pagelist(struct nfs_pgio_header *header) 249 { 250 struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); 251 struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; 252 struct bio *bio = NULL; 253 struct pnfs_block_extent be; 254 sector_t isect, extent_length = 0; 255 struct parallel_io *par; 256 loff_t f_offset = header->args.offset; 257 size_t bytes_left = header->args.count; 258 unsigned int pg_offset = header->args.pgbase, pg_len; 259 struct page **pages = header->args.pages; 260 int pg_index = header->args.pgbase >> PAGE_SHIFT; 261 const bool is_dio = (header->dreq != NULL); 262 struct blk_plug plug; 263 int i; 264 265 dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__, 266 header->page_array.npages, f_offset, 267 (unsigned int)header->args.count); 268 269 par = alloc_parallel(header); 270 if (!par) 271 return PNFS_NOT_ATTEMPTED; 272 par->pnfs_callback = bl_end_par_io_read; 273 274 blk_start_plug(&plug); 275 276 isect = (sector_t) (f_offset >> SECTOR_SHIFT); 277 /* Code assumes extents are page-aligned */ 278 for (i = pg_index; i < header->page_array.npages; i++) { 279 if (extent_length <= 0) { 280 /* We've used up the previous extent */ 281 bio = bl_submit_bio(bio); 282 283 /* Get the next one */ 284 if (!ext_tree_lookup(bl, isect, &be, false)) { 285 header->pnfs_error = -EIO; 286 goto out; 287 } 288 extent_length = be.be_length - (isect - be.be_f_offset); 289 } 290 291 if (is_dio) { 292 if (pg_offset + bytes_left > PAGE_SIZE) 293 pg_len = PAGE_SIZE - pg_offset; 294 else 295 pg_len = bytes_left; 296 } else { 297 BUG_ON(pg_offset != 0); 298 pg_len = PAGE_SIZE; 299 } 300 301 if (is_hole(&be)) { 302 bio = bl_submit_bio(bio); 303 /* Fill hole w/ zeroes w/o accessing device */ 304 dprintk("%s Zeroing page for hole\n", __func__); 305 zero_user_segment(pages[i], pg_offset, pg_len); 306 307 /* invalidate map */ 308 map.start = NFS4_MAX_UINT64; 309 } else { 310 bio = do_add_page_to_bio(bio, 311 header->page_array.npages - i, 312 READ, 313 isect, pages[i], &map, &be, 314 bl_end_io_read, par, 315 pg_offset, &pg_len); 316 if (IS_ERR(bio)) { 317 header->pnfs_error = PTR_ERR(bio); 318 bio = NULL; 319 goto out; 320 } 321 } 322 isect += (pg_len >> SECTOR_SHIFT); 323 extent_length -= (pg_len >> SECTOR_SHIFT); 324 f_offset += pg_len; 325 bytes_left -= pg_len; 326 pg_offset = 0; 327 } 328 if ((isect << SECTOR_SHIFT) >= header->inode->i_size) { 329 header->res.eof = 1; 330 header->res.count = header->inode->i_size - header->args.offset; 331 } else { 332 header->res.count = (isect << SECTOR_SHIFT) - header->args.offset; 333 } 334 out: 335 bl_submit_bio(bio); 336 blk_finish_plug(&plug); 337 put_parallel(par); 338 return PNFS_ATTEMPTED; 339 } 340 341 static void bl_end_io_write(struct bio *bio) 342 { 343 struct parallel_io *par = bio->bi_private; 344 struct nfs_pgio_header *header = par->data; 345 346 if (bio->bi_status) { 347 if (!header->pnfs_error) 348 header->pnfs_error = -EIO; 349 pnfs_set_lo_fail(header->lseg); 350 bl_mark_devices_unavailable(header, true); 351 } 352 bio_put(bio); 353 put_parallel(par); 354 } 355 356 /* Function scheduled for call during bl_end_par_io_write, 357 * it marks sectors as written and extends the commitlist. 358 */ 359 static void bl_write_cleanup(struct work_struct *work) 360 { 361 struct rpc_task *task = container_of(work, struct rpc_task, u.tk_work); 362 struct nfs_pgio_header *hdr = 363 container_of(task, struct nfs_pgio_header, task); 364 365 dprintk("%s enter\n", __func__); 366 367 if (likely(!hdr->pnfs_error)) { 368 struct pnfs_block_layout *bl = BLK_LSEG2EXT(hdr->lseg); 369 u64 start = hdr->args.offset & (loff_t)PAGE_MASK; 370 u64 end = (hdr->args.offset + hdr->args.count + 371 PAGE_SIZE - 1) & (loff_t)PAGE_MASK; 372 u64 lwb = hdr->args.offset + hdr->args.count; 373 374 ext_tree_mark_written(bl, start >> SECTOR_SHIFT, 375 (end - start) >> SECTOR_SHIFT, lwb); 376 } 377 378 pnfs_ld_write_done(hdr); 379 } 380 381 /* Called when last of bios associated with a bl_write_pagelist call finishes */ 382 static void bl_end_par_io_write(void *data) 383 { 384 struct nfs_pgio_header *hdr = data; 385 386 hdr->task.tk_status = hdr->pnfs_error; 387 hdr->verf.committed = NFS_FILE_SYNC; 388 INIT_WORK(&hdr->task.u.tk_work, bl_write_cleanup); 389 schedule_work(&hdr->task.u.tk_work); 390 } 391 392 static enum pnfs_try_status 393 bl_write_pagelist(struct nfs_pgio_header *header, int sync) 394 { 395 struct pnfs_block_layout *bl = BLK_LSEG2EXT(header->lseg); 396 struct pnfs_block_dev_map map = { .start = NFS4_MAX_UINT64 }; 397 struct bio *bio = NULL; 398 struct pnfs_block_extent be; 399 sector_t isect, extent_length = 0; 400 struct parallel_io *par = NULL; 401 loff_t offset = header->args.offset; 402 size_t count = header->args.count; 403 struct page **pages = header->args.pages; 404 int pg_index = header->args.pgbase >> PAGE_SHIFT; 405 unsigned int pg_len; 406 struct blk_plug plug; 407 int i; 408 409 dprintk("%s enter, %zu@%lld\n", __func__, count, offset); 410 411 /* At this point, header->page_aray is a (sequential) list of nfs_pages. 412 * We want to write each, and if there is an error set pnfs_error 413 * to have it redone using nfs. 414 */ 415 par = alloc_parallel(header); 416 if (!par) 417 return PNFS_NOT_ATTEMPTED; 418 par->pnfs_callback = bl_end_par_io_write; 419 420 blk_start_plug(&plug); 421 422 /* we always write out the whole page */ 423 offset = offset & (loff_t)PAGE_MASK; 424 isect = offset >> SECTOR_SHIFT; 425 426 for (i = pg_index; i < header->page_array.npages; i++) { 427 if (extent_length <= 0) { 428 /* We've used up the previous extent */ 429 bio = bl_submit_bio(bio); 430 /* Get the next one */ 431 if (!ext_tree_lookup(bl, isect, &be, true)) { 432 header->pnfs_error = -EINVAL; 433 goto out; 434 } 435 436 extent_length = be.be_length - (isect - be.be_f_offset); 437 } 438 439 pg_len = PAGE_SIZE; 440 bio = do_add_page_to_bio(bio, header->page_array.npages - i, 441 WRITE, isect, pages[i], &map, &be, 442 bl_end_io_write, par, 443 0, &pg_len); 444 if (IS_ERR(bio)) { 445 header->pnfs_error = PTR_ERR(bio); 446 bio = NULL; 447 goto out; 448 } 449 450 offset += pg_len; 451 count -= pg_len; 452 isect += (pg_len >> SECTOR_SHIFT); 453 extent_length -= (pg_len >> SECTOR_SHIFT); 454 } 455 456 header->res.count = header->args.count; 457 out: 458 bl_submit_bio(bio); 459 blk_finish_plug(&plug); 460 put_parallel(par); 461 return PNFS_ATTEMPTED; 462 } 463 464 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo) 465 { 466 struct pnfs_block_layout *bl = BLK_LO2EXT(lo); 467 int err; 468 469 dprintk("%s enter\n", __func__); 470 471 err = ext_tree_remove(bl, true, 0, LLONG_MAX); 472 WARN_ON(err); 473 474 kfree_rcu(bl, bl_layout.plh_rcu); 475 } 476 477 static struct pnfs_layout_hdr *__bl_alloc_layout_hdr(struct inode *inode, 478 gfp_t gfp_flags, bool is_scsi_layout) 479 { 480 struct pnfs_block_layout *bl; 481 482 dprintk("%s enter\n", __func__); 483 bl = kzalloc(sizeof(*bl), gfp_flags); 484 if (!bl) 485 return NULL; 486 487 bl->bl_ext_rw = RB_ROOT; 488 bl->bl_ext_ro = RB_ROOT; 489 spin_lock_init(&bl->bl_ext_lock); 490 491 bl->bl_scsi_layout = is_scsi_layout; 492 return &bl->bl_layout; 493 } 494 495 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode, 496 gfp_t gfp_flags) 497 { 498 return __bl_alloc_layout_hdr(inode, gfp_flags, false); 499 } 500 501 static struct pnfs_layout_hdr *sl_alloc_layout_hdr(struct inode *inode, 502 gfp_t gfp_flags) 503 { 504 return __bl_alloc_layout_hdr(inode, gfp_flags, true); 505 } 506 507 static void bl_free_lseg(struct pnfs_layout_segment *lseg) 508 { 509 dprintk("%s enter\n", __func__); 510 kfree(lseg); 511 } 512 513 /* Tracks info needed to ensure extents in layout obey constraints of spec */ 514 struct layout_verification { 515 u32 mode; /* R or RW */ 516 u64 start; /* Expected start of next non-COW extent */ 517 u64 inval; /* Start of INVAL coverage */ 518 u64 cowread; /* End of COW read coverage */ 519 }; 520 521 /* Verify the extent meets the layout requirements of the pnfs-block draft, 522 * section 2.3.1. 523 */ 524 static int verify_extent(struct pnfs_block_extent *be, 525 struct layout_verification *lv) 526 { 527 if (lv->mode == IOMODE_READ) { 528 if (be->be_state == PNFS_BLOCK_READWRITE_DATA || 529 be->be_state == PNFS_BLOCK_INVALID_DATA) 530 return -EIO; 531 if (be->be_f_offset != lv->start) 532 return -EIO; 533 lv->start += be->be_length; 534 return 0; 535 } 536 /* lv->mode == IOMODE_RW */ 537 if (be->be_state == PNFS_BLOCK_READWRITE_DATA) { 538 if (be->be_f_offset != lv->start) 539 return -EIO; 540 if (lv->cowread > lv->start) 541 return -EIO; 542 lv->start += be->be_length; 543 lv->inval = lv->start; 544 return 0; 545 } else if (be->be_state == PNFS_BLOCK_INVALID_DATA) { 546 if (be->be_f_offset != lv->start) 547 return -EIO; 548 lv->start += be->be_length; 549 return 0; 550 } else if (be->be_state == PNFS_BLOCK_READ_DATA) { 551 if (be->be_f_offset > lv->start) 552 return -EIO; 553 if (be->be_f_offset < lv->inval) 554 return -EIO; 555 if (be->be_f_offset < lv->cowread) 556 return -EIO; 557 /* It looks like you might want to min this with lv->start, 558 * but you really don't. 559 */ 560 lv->inval = lv->inval + be->be_length; 561 lv->cowread = be->be_f_offset + be->be_length; 562 return 0; 563 } else 564 return -EIO; 565 } 566 567 static int decode_sector_number(__be32 **rp, sector_t *sp) 568 { 569 uint64_t s; 570 571 *rp = xdr_decode_hyper(*rp, &s); 572 if (s & 0x1ff) { 573 printk(KERN_WARNING "NFS: %s: sector not aligned\n", __func__); 574 return -1; 575 } 576 *sp = s >> SECTOR_SHIFT; 577 return 0; 578 } 579 580 static struct nfs4_deviceid_node * 581 bl_find_get_deviceid(struct nfs_server *server, 582 const struct nfs4_deviceid *id, const struct cred *cred, 583 gfp_t gfp_mask) 584 { 585 struct nfs4_deviceid_node *node; 586 unsigned long start, end; 587 588 retry: 589 node = nfs4_find_get_deviceid(server, id, cred, gfp_mask); 590 if (!node) 591 return ERR_PTR(-ENODEV); 592 593 if (test_bit(NFS_DEVICEID_UNAVAILABLE, &node->flags) == 0) 594 return node; 595 596 end = jiffies; 597 start = end - PNFS_DEVICE_RETRY_TIMEOUT; 598 if (!time_in_range(node->timestamp_unavailable, start, end)) { 599 nfs4_delete_deviceid(node->ld, node->nfs_client, id); 600 goto retry; 601 } 602 return ERR_PTR(-ENODEV); 603 } 604 605 static int 606 bl_alloc_extent(struct xdr_stream *xdr, struct pnfs_layout_hdr *lo, 607 struct layout_verification *lv, struct list_head *extents, 608 gfp_t gfp_mask) 609 { 610 struct pnfs_block_extent *be; 611 struct nfs4_deviceid id; 612 int error; 613 __be32 *p; 614 615 p = xdr_inline_decode(xdr, 28 + NFS4_DEVICEID4_SIZE); 616 if (!p) 617 return -EIO; 618 619 be = kzalloc(sizeof(*be), GFP_NOFS); 620 if (!be) 621 return -ENOMEM; 622 623 memcpy(&id, p, NFS4_DEVICEID4_SIZE); 624 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE); 625 626 be->be_device = bl_find_get_deviceid(NFS_SERVER(lo->plh_inode), &id, 627 lo->plh_lc_cred, gfp_mask); 628 if (IS_ERR(be->be_device)) { 629 error = PTR_ERR(be->be_device); 630 goto out_free_be; 631 } 632 633 /* 634 * The next three values are read in as bytes, but stored in the 635 * extent structure in 512-byte granularity. 636 */ 637 error = -EIO; 638 if (decode_sector_number(&p, &be->be_f_offset) < 0) 639 goto out_put_deviceid; 640 if (decode_sector_number(&p, &be->be_length) < 0) 641 goto out_put_deviceid; 642 if (decode_sector_number(&p, &be->be_v_offset) < 0) 643 goto out_put_deviceid; 644 be->be_state = be32_to_cpup(p++); 645 646 error = verify_extent(be, lv); 647 if (error) { 648 dprintk("%s: extent verification failed\n", __func__); 649 goto out_put_deviceid; 650 } 651 652 list_add_tail(&be->be_list, extents); 653 return 0; 654 655 out_put_deviceid: 656 nfs4_put_deviceid_node(be->be_device); 657 out_free_be: 658 kfree(be); 659 return error; 660 } 661 662 static struct pnfs_layout_segment * 663 bl_alloc_lseg(struct pnfs_layout_hdr *lo, struct nfs4_layoutget_res *lgr, 664 gfp_t gfp_mask) 665 { 666 struct layout_verification lv = { 667 .mode = lgr->range.iomode, 668 .start = lgr->range.offset >> SECTOR_SHIFT, 669 .inval = lgr->range.offset >> SECTOR_SHIFT, 670 .cowread = lgr->range.offset >> SECTOR_SHIFT, 671 }; 672 struct pnfs_block_layout *bl = BLK_LO2EXT(lo); 673 struct pnfs_layout_segment *lseg; 674 struct xdr_buf buf; 675 struct xdr_stream xdr; 676 struct page *scratch; 677 int status, i; 678 uint32_t count; 679 __be32 *p; 680 LIST_HEAD(extents); 681 682 dprintk("---> %s\n", __func__); 683 684 lseg = kzalloc(sizeof(*lseg), gfp_mask); 685 if (!lseg) 686 return ERR_PTR(-ENOMEM); 687 688 status = -ENOMEM; 689 scratch = alloc_page(gfp_mask); 690 if (!scratch) 691 goto out; 692 693 xdr_init_decode_pages(&xdr, &buf, 694 lgr->layoutp->pages, lgr->layoutp->len); 695 xdr_set_scratch_page(&xdr, scratch); 696 697 status = -EIO; 698 p = xdr_inline_decode(&xdr, 4); 699 if (unlikely(!p)) 700 goto out_free_scratch; 701 702 count = be32_to_cpup(p++); 703 dprintk("%s: number of extents %d\n", __func__, count); 704 705 /* 706 * Decode individual extents, putting them in temporary staging area 707 * until whole layout is decoded to make error recovery easier. 708 */ 709 for (i = 0; i < count; i++) { 710 status = bl_alloc_extent(&xdr, lo, &lv, &extents, gfp_mask); 711 if (status) 712 goto process_extents; 713 } 714 715 if (lgr->range.offset + lgr->range.length != 716 lv.start << SECTOR_SHIFT) { 717 dprintk("%s Final length mismatch\n", __func__); 718 status = -EIO; 719 goto process_extents; 720 } 721 722 if (lv.start < lv.cowread) { 723 dprintk("%s Final uncovered COW extent\n", __func__); 724 status = -EIO; 725 } 726 727 process_extents: 728 while (!list_empty(&extents)) { 729 struct pnfs_block_extent *be = 730 list_first_entry(&extents, struct pnfs_block_extent, 731 be_list); 732 list_del(&be->be_list); 733 734 if (!status) 735 status = ext_tree_insert(bl, be); 736 737 if (status) { 738 nfs4_put_deviceid_node(be->be_device); 739 kfree(be); 740 } 741 } 742 743 out_free_scratch: 744 __free_page(scratch); 745 out: 746 dprintk("%s returns %d\n", __func__, status); 747 switch (status) { 748 case -ENODEV: 749 /* Our extent block devices are unavailable */ 750 set_bit(NFS_LSEG_UNAVAILABLE, &lseg->pls_flags); 751 fallthrough; 752 case 0: 753 return lseg; 754 default: 755 kfree(lseg); 756 return ERR_PTR(status); 757 } 758 } 759 760 static void 761 bl_return_range(struct pnfs_layout_hdr *lo, 762 struct pnfs_layout_range *range) 763 { 764 struct pnfs_block_layout *bl = BLK_LO2EXT(lo); 765 sector_t offset = range->offset >> SECTOR_SHIFT, end; 766 767 if (range->offset % 8) { 768 dprintk("%s: offset %lld not block size aligned\n", 769 __func__, range->offset); 770 return; 771 } 772 773 if (range->length != NFS4_MAX_UINT64) { 774 if (range->length % 8) { 775 dprintk("%s: length %lld not block size aligned\n", 776 __func__, range->length); 777 return; 778 } 779 780 end = offset + (range->length >> SECTOR_SHIFT); 781 } else { 782 end = round_down(NFS4_MAX_UINT64, PAGE_SIZE); 783 } 784 785 ext_tree_remove(bl, range->iomode & IOMODE_RW, offset, end); 786 } 787 788 static int 789 bl_prepare_layoutcommit(struct nfs4_layoutcommit_args *arg) 790 { 791 return ext_tree_prepare_commit(arg); 792 } 793 794 static void 795 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata) 796 { 797 ext_tree_mark_committed(&lcdata->args, lcdata->res.status); 798 } 799 800 static int 801 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh) 802 { 803 dprintk("%s enter\n", __func__); 804 805 if (server->pnfs_blksize == 0) { 806 dprintk("%s Server did not return blksize\n", __func__); 807 return -EINVAL; 808 } 809 if (server->pnfs_blksize > PAGE_SIZE) { 810 printk(KERN_ERR "%s: pNFS blksize %d not supported.\n", 811 __func__, server->pnfs_blksize); 812 return -EINVAL; 813 } 814 815 return 0; 816 } 817 818 static bool 819 is_aligned_req(struct nfs_pageio_descriptor *pgio, 820 struct nfs_page *req, unsigned int alignment, bool is_write) 821 { 822 /* 823 * Always accept buffered writes, higher layers take care of the 824 * right alignment. 825 */ 826 if (pgio->pg_dreq == NULL) 827 return true; 828 829 if (!IS_ALIGNED(req->wb_offset, alignment)) 830 return false; 831 832 if (IS_ALIGNED(req->wb_bytes, alignment)) 833 return true; 834 835 if (is_write && 836 (req_offset(req) + req->wb_bytes == i_size_read(pgio->pg_inode))) { 837 /* 838 * If the write goes up to the inode size, just write 839 * the full page. Data past the inode size is 840 * guaranteed to be zeroed by the higher level client 841 * code, and this behaviour is mandated by RFC 5663 842 * section 2.3.2. 843 */ 844 return true; 845 } 846 847 return false; 848 } 849 850 static void 851 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) 852 { 853 if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) { 854 nfs_pageio_reset_read_mds(pgio); 855 return; 856 } 857 858 pnfs_generic_pg_init_read(pgio, req); 859 860 if (pgio->pg_lseg && 861 test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) { 862 pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg); 863 pnfs_set_lo_fail(pgio->pg_lseg); 864 nfs_pageio_reset_read_mds(pgio); 865 } 866 } 867 868 /* 869 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number 870 * of bytes (maximum @req->wb_bytes) that can be coalesced. 871 */ 872 static size_t 873 bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, 874 struct nfs_page *req) 875 { 876 if (!is_aligned_req(pgio, req, SECTOR_SIZE, false)) 877 return 0; 878 return pnfs_generic_pg_test(pgio, prev, req); 879 } 880 881 /* 882 * Return the number of contiguous bytes for a given inode 883 * starting at page frame idx. 884 */ 885 static u64 pnfs_num_cont_bytes(struct inode *inode, pgoff_t idx) 886 { 887 struct address_space *mapping = inode->i_mapping; 888 pgoff_t end; 889 890 /* Optimize common case that writes from 0 to end of file */ 891 end = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); 892 if (end != inode->i_mapping->nrpages) { 893 rcu_read_lock(); 894 end = page_cache_next_miss(mapping, idx + 1, ULONG_MAX); 895 rcu_read_unlock(); 896 } 897 898 if (!end) 899 return i_size_read(inode) - (idx << PAGE_SHIFT); 900 else 901 return (end - idx) << PAGE_SHIFT; 902 } 903 904 static void 905 bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) 906 { 907 u64 wb_size; 908 909 if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) { 910 nfs_pageio_reset_write_mds(pgio); 911 return; 912 } 913 914 if (pgio->pg_dreq == NULL) 915 wb_size = pnfs_num_cont_bytes(pgio->pg_inode, 916 req->wb_index); 917 else 918 wb_size = nfs_dreq_bytes_left(pgio->pg_dreq); 919 920 pnfs_generic_pg_init_write(pgio, req, wb_size); 921 922 if (pgio->pg_lseg && 923 test_bit(NFS_LSEG_UNAVAILABLE, &pgio->pg_lseg->pls_flags)) { 924 925 pnfs_error_mark_layout_for_return(pgio->pg_inode, pgio->pg_lseg); 926 pnfs_set_lo_fail(pgio->pg_lseg); 927 nfs_pageio_reset_write_mds(pgio); 928 } 929 } 930 931 /* 932 * Return 0 if @req cannot be coalesced into @pgio, otherwise return the number 933 * of bytes (maximum @req->wb_bytes) that can be coalesced. 934 */ 935 static size_t 936 bl_pg_test_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev, 937 struct nfs_page *req) 938 { 939 if (!is_aligned_req(pgio, req, PAGE_SIZE, true)) 940 return 0; 941 return pnfs_generic_pg_test(pgio, prev, req); 942 } 943 944 static const struct nfs_pageio_ops bl_pg_read_ops = { 945 .pg_init = bl_pg_init_read, 946 .pg_test = bl_pg_test_read, 947 .pg_doio = pnfs_generic_pg_readpages, 948 .pg_cleanup = pnfs_generic_pg_cleanup, 949 }; 950 951 static const struct nfs_pageio_ops bl_pg_write_ops = { 952 .pg_init = bl_pg_init_write, 953 .pg_test = bl_pg_test_write, 954 .pg_doio = pnfs_generic_pg_writepages, 955 .pg_cleanup = pnfs_generic_pg_cleanup, 956 }; 957 958 static struct pnfs_layoutdriver_type blocklayout_type = { 959 .id = LAYOUT_BLOCK_VOLUME, 960 .name = "LAYOUT_BLOCK_VOLUME", 961 .owner = THIS_MODULE, 962 .flags = PNFS_LAYOUTRET_ON_SETATTR | 963 PNFS_LAYOUTRET_ON_ERROR | 964 PNFS_READ_WHOLE_PAGE, 965 .read_pagelist = bl_read_pagelist, 966 .write_pagelist = bl_write_pagelist, 967 .alloc_layout_hdr = bl_alloc_layout_hdr, 968 .free_layout_hdr = bl_free_layout_hdr, 969 .alloc_lseg = bl_alloc_lseg, 970 .free_lseg = bl_free_lseg, 971 .return_range = bl_return_range, 972 .prepare_layoutcommit = bl_prepare_layoutcommit, 973 .cleanup_layoutcommit = bl_cleanup_layoutcommit, 974 .set_layoutdriver = bl_set_layoutdriver, 975 .alloc_deviceid_node = bl_alloc_deviceid_node, 976 .free_deviceid_node = bl_free_deviceid_node, 977 .pg_read_ops = &bl_pg_read_ops, 978 .pg_write_ops = &bl_pg_write_ops, 979 .sync = pnfs_generic_sync, 980 }; 981 982 static struct pnfs_layoutdriver_type scsilayout_type = { 983 .id = LAYOUT_SCSI, 984 .name = "LAYOUT_SCSI", 985 .owner = THIS_MODULE, 986 .flags = PNFS_LAYOUTRET_ON_SETATTR | 987 PNFS_LAYOUTRET_ON_ERROR | 988 PNFS_READ_WHOLE_PAGE, 989 .read_pagelist = bl_read_pagelist, 990 .write_pagelist = bl_write_pagelist, 991 .alloc_layout_hdr = sl_alloc_layout_hdr, 992 .free_layout_hdr = bl_free_layout_hdr, 993 .alloc_lseg = bl_alloc_lseg, 994 .free_lseg = bl_free_lseg, 995 .return_range = bl_return_range, 996 .prepare_layoutcommit = bl_prepare_layoutcommit, 997 .cleanup_layoutcommit = bl_cleanup_layoutcommit, 998 .set_layoutdriver = bl_set_layoutdriver, 999 .alloc_deviceid_node = bl_alloc_deviceid_node, 1000 .free_deviceid_node = bl_free_deviceid_node, 1001 .pg_read_ops = &bl_pg_read_ops, 1002 .pg_write_ops = &bl_pg_write_ops, 1003 .sync = pnfs_generic_sync, 1004 }; 1005 1006 1007 static int __init nfs4blocklayout_init(void) 1008 { 1009 int ret; 1010 1011 dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__); 1012 1013 ret = bl_init_pipefs(); 1014 if (ret) 1015 goto out; 1016 1017 ret = pnfs_register_layoutdriver(&blocklayout_type); 1018 if (ret) 1019 goto out_cleanup_pipe; 1020 1021 ret = pnfs_register_layoutdriver(&scsilayout_type); 1022 if (ret) 1023 goto out_unregister_block; 1024 return 0; 1025 1026 out_unregister_block: 1027 pnfs_unregister_layoutdriver(&blocklayout_type); 1028 out_cleanup_pipe: 1029 bl_cleanup_pipefs(); 1030 out: 1031 return ret; 1032 } 1033 1034 static void __exit nfs4blocklayout_exit(void) 1035 { 1036 dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n", 1037 __func__); 1038 1039 pnfs_unregister_layoutdriver(&scsilayout_type); 1040 pnfs_unregister_layoutdriver(&blocklayout_type); 1041 bl_cleanup_pipefs(); 1042 } 1043 1044 MODULE_ALIAS("nfs-layouttype4-3"); 1045 MODULE_ALIAS("nfs-layouttype4-5"); 1046 1047 module_init(nfs4blocklayout_init); 1048 module_exit(nfs4blocklayout_exit); 1049