fs/nfs/blocklayout/blocklayout.c

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