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/buffer_head.h>  /* various write calls */
  39 #include <linux/prefetch.h>
  40
  41 #include "blocklayout.h"
  42
  43 #define NFSDBG_FACILITY NFSDBG_PNFS_LD
  44
  45 MODULE_LICENSE("GPL");
  46 MODULE_AUTHOR("Andy Adamson <andros@citi.umich.edu>");
  47 MODULE_DESCRIPTION("The NFSv4.1 pNFS Block layout driver");
  48
  49 struct dentry *bl_device_pipe;
  50 wait_queue_head_t bl_wq;
  51
  52 static void print_page(struct page *page)
  53 {
  54         dprintk("PRINTPAGE page %p\n", page);
  55         dprintk("       PagePrivate %d\n", PagePrivate(page));
  56         dprintk("       PageUptodate %d\n", PageUptodate(page));
  57         dprintk("       PageError %d\n", PageError(page));
  58         dprintk("       PageDirty %d\n", PageDirty(page));
  59         dprintk("       PageReferenced %d\n", PageReferenced(page));
  60         dprintk("       PageLocked %d\n", PageLocked(page));
  61         dprintk("       PageWriteback %d\n", PageWriteback(page));
  62         dprintk("       PageMappedToDisk %d\n", PageMappedToDisk(page));
  63         dprintk("\n");
  64 }
  65
  66 /* Given the be associated with isect, determine if page data needs to be
  67  * initialized.
  68  */
  69 static int is_hole(struct pnfs_block_extent *be, sector_t isect)
  70 {
  71         if (be->be_state == PNFS_BLOCK_NONE_DATA)
  72                 return 1;
  73         else if (be->be_state != PNFS_BLOCK_INVALID_DATA)
  74                 return 0;
  75         else
  76                 return !bl_is_sector_init(be->be_inval, isect);
  77 }
  78
  79 /* Given the be associated with isect, determine if page data can be
  80  * written to disk.
  81  */
  82 static int is_writable(struct pnfs_block_extent *be, sector_t isect)
  83 {
  84         return (be->be_state == PNFS_BLOCK_READWRITE_DATA ||
  85                 be->be_state == PNFS_BLOCK_INVALID_DATA);
  86 }
  87
  88 /* The data we are handed might be spread across several bios.  We need
  89  * to track when the last one is finished.
  90  */
  91 struct parallel_io {
  92         struct kref refcnt;
  93         struct rpc_call_ops call_ops;
  94         void (*pnfs_callback) (void *data);
  95         void *data;
  96 };
  97
  98 static inline struct parallel_io *alloc_parallel(void *data)
  99 {
 100         struct parallel_io *rv;
 101
 102         rv  = kmalloc(sizeof(*rv), GFP_NOFS);
 103         if (rv) {
 104                 rv->data = data;
 105                 kref_init(&rv->refcnt);
 106         }
 107         return rv;
 108 }
 109
 110 static inline void get_parallel(struct parallel_io *p)
 111 {
 112         kref_get(&p->refcnt);
 113 }
 114
 115 static void destroy_parallel(struct kref *kref)
 116 {
 117         struct parallel_io *p = container_of(kref, struct parallel_io, refcnt);
 118
 119         dprintk("%s enter\n", __func__);
 120         p->pnfs_callback(p->data);
 121         kfree(p);
 122 }
 123
 124 static inline void put_parallel(struct parallel_io *p)
 125 {
 126         kref_put(&p->refcnt, destroy_parallel);
 127 }
 128
 129 static struct bio *
 130 bl_submit_bio(int rw, struct bio *bio)
 131 {
 132         if (bio) {
 133                 get_parallel(bio->bi_private);
 134                 dprintk("%s submitting %s bio %u@%llu\n", __func__,
 135                         rw == READ ? "read" : "write",
 136                         bio->bi_size, (unsigned long long)bio->bi_sector);
 137                 submit_bio(rw, bio);
 138         }
 139         return NULL;
 140 }
 141
 142 static struct bio *bl_alloc_init_bio(int npg, sector_t isect,
 143                                      struct pnfs_block_extent *be,
 144                                      void (*end_io)(struct bio *, int err),
 145                                      struct parallel_io *par)
 146 {
 147         struct bio *bio;
 148
 149         bio = bio_alloc(GFP_NOIO, npg);
 150         if (!bio)
 151                 return NULL;
 152
 153         bio->bi_sector = isect - be->be_f_offset + be->be_v_offset;
 154         bio->bi_bdev = be->be_mdev;
 155         bio->bi_end_io = end_io;
 156         bio->bi_private = par;
 157         return bio;
 158 }
 159
 160 static struct bio *bl_add_page_to_bio(struct bio *bio, int npg, int rw,
 161                                       sector_t isect, struct page *page,
 162                                       struct pnfs_block_extent *be,
 163                                       void (*end_io)(struct bio *, int err),
 164                                       struct parallel_io *par)
 165 {
 166 retry:
 167         if (!bio) {
 168                 bio = bl_alloc_init_bio(npg, isect, be, end_io, par);
 169                 if (!bio)
 170                         return ERR_PTR(-ENOMEM);
 171         }
 172         if (bio_add_page(bio, page, PAGE_CACHE_SIZE, 0) < PAGE_CACHE_SIZE) {
 173                 bio = bl_submit_bio(rw, bio);
 174                 goto retry;
 175         }
 176         return bio;
 177 }
 178
 179 static void bl_set_lo_fail(struct pnfs_layout_segment *lseg)
 180 {
 181         if (lseg->pls_range.iomode == IOMODE_RW) {
 182                 dprintk("%s Setting layout IOMODE_RW fail bit\n", __func__);
 183                 set_bit(lo_fail_bit(IOMODE_RW), &lseg->pls_layout->plh_flags);
 184         } else {
 185                 dprintk("%s Setting layout IOMODE_READ fail bit\n", __func__);
 186                 set_bit(lo_fail_bit(IOMODE_READ), &lseg->pls_layout->plh_flags);
 187         }
 188 }
 189
 190 /* This is basically copied from mpage_end_io_read */
 191 static void bl_end_io_read(struct bio *bio, int err)
 192 {
 193         struct parallel_io *par = bio->bi_private;
 194         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 195         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 196         struct nfs_read_data *rdata = (struct nfs_read_data *)par->data;
 197
 198         do {
 199                 struct page *page = bvec->bv_page;
 200
 201                 if (--bvec >= bio->bi_io_vec)
 202                         prefetchw(&bvec->bv_page->flags);
 203                 if (uptodate)
 204                         SetPageUptodate(page);
 205         } while (bvec >= bio->bi_io_vec);
 206         if (!uptodate) {
 207                 if (!rdata->pnfs_error)
 208                         rdata->pnfs_error = -EIO;
 209                 bl_set_lo_fail(rdata->lseg);
 210         }
 211         bio_put(bio);
 212         put_parallel(par);
 213 }
 214
 215 static void bl_read_cleanup(struct work_struct *work)
 216 {
 217         struct rpc_task *task;
 218         struct nfs_read_data *rdata;
 219         dprintk("%s enter\n", __func__);
 220         task = container_of(work, struct rpc_task, u.tk_work);
 221         rdata = container_of(task, struct nfs_read_data, task);
 222         pnfs_ld_read_done(rdata);
 223 }
 224
 225 static void
 226 bl_end_par_io_read(void *data)
 227 {
 228         struct nfs_read_data *rdata = data;
 229
 230         INIT_WORK(&rdata->task.u.tk_work, bl_read_cleanup);
 231         schedule_work(&rdata->task.u.tk_work);
 232 }
 233
 234 /* We don't want normal .rpc_call_done callback used, so we replace it
 235  * with this stub.
 236  */
 237 static void bl_rpc_do_nothing(struct rpc_task *task, void *calldata)
 238 {
 239         return;
 240 }
 241
 242 static enum pnfs_try_status
 243 bl_read_pagelist(struct nfs_read_data *rdata)
 244 {
 245         int i, hole;
 246         struct bio *bio = NULL;
 247         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 248         sector_t isect, extent_length = 0;
 249         struct parallel_io *par;
 250         loff_t f_offset = rdata->args.offset;
 251         size_t count = rdata->args.count;
 252         struct page **pages = rdata->args.pages;
 253         int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
 254
 255         dprintk("%s enter nr_pages %u offset %lld count %Zd\n", __func__,
 256                rdata->npages, f_offset, count);
 257
 258         par = alloc_parallel(rdata);
 259         if (!par)
 260                 goto use_mds;
 261         par->call_ops = *rdata->mds_ops;
 262         par->call_ops.rpc_call_done = bl_rpc_do_nothing;
 263         par->pnfs_callback = bl_end_par_io_read;
 264         /* At this point, we can no longer jump to use_mds */
 265
 266         isect = (sector_t) (f_offset >> SECTOR_SHIFT);
 267         /* Code assumes extents are page-aligned */
 268         for (i = pg_index; i < rdata->npages; i++) {
 269                 if (!extent_length) {
 270                         /* We've used up the previous extent */
 271                         bl_put_extent(be);
 272                         bl_put_extent(cow_read);
 273                         bio = bl_submit_bio(READ, bio);
 274                         /* Get the next one */
 275                         be = bl_find_get_extent(BLK_LSEG2EXT(rdata->lseg),
 276                                              isect, &cow_read);
 277                         if (!be) {
 278                                 rdata->pnfs_error = -EIO;
 279                                 goto out;
 280                         }
 281                         extent_length = be->be_length -
 282                                 (isect - be->be_f_offset);
 283                         if (cow_read) {
 284                                 sector_t cow_length = cow_read->be_length -
 285                                         (isect - cow_read->be_f_offset);
 286                                 extent_length = min(extent_length, cow_length);
 287                         }
 288                 }
 289                 hole = is_hole(be, isect);
 290                 if (hole && !cow_read) {
 291                         bio = bl_submit_bio(READ, bio);
 292                         /* Fill hole w/ zeroes w/o accessing device */
 293                         dprintk("%s Zeroing page for hole\n", __func__);
 294                         zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
 295                         print_page(pages[i]);
 296                         SetPageUptodate(pages[i]);
 297                 } else {
 298                         struct pnfs_block_extent *be_read;
 299
 300                         be_read = (hole && cow_read) ? cow_read : be;
 301                         bio = bl_add_page_to_bio(bio, rdata->npages - i, READ,
 302                                                  isect, pages[i], be_read,
 303                                                  bl_end_io_read, par);
 304                         if (IS_ERR(bio)) {
 305                                 rdata->pnfs_error = PTR_ERR(bio);
 306                                 goto out;
 307                         }
 308                 }
 309                 isect += PAGE_CACHE_SECTORS;
 310                 extent_length -= PAGE_CACHE_SECTORS;
 311         }
 312         if ((isect << SECTOR_SHIFT) >= rdata->inode->i_size) {
 313                 rdata->res.eof = 1;
 314                 rdata->res.count = rdata->inode->i_size - f_offset;
 315         } else {
 316                 rdata->res.count = (isect << SECTOR_SHIFT) - f_offset;
 317         }
 318 out:
 319         bl_put_extent(be);
 320         bl_put_extent(cow_read);
 321         bl_submit_bio(READ, bio);
 322         put_parallel(par);
 323         return PNFS_ATTEMPTED;
 324
 325  use_mds:
 326         dprintk("Giving up and using normal NFS\n");
 327         return PNFS_NOT_ATTEMPTED;
 328 }
 329
 330 static void mark_extents_written(struct pnfs_block_layout *bl,
 331                                  __u64 offset, __u32 count)
 332 {
 333         sector_t isect, end;
 334         struct pnfs_block_extent *be;
 335
 336         dprintk("%s(%llu, %u)\n", __func__, offset, count);
 337         if (count == 0)
 338                 return;
 339         isect = (offset & (long)(PAGE_CACHE_MASK)) >> SECTOR_SHIFT;
 340         end = (offset + count + PAGE_CACHE_SIZE - 1) & (long)(PAGE_CACHE_MASK);
 341         end >>= SECTOR_SHIFT;
 342         while (isect < end) {
 343                 sector_t len;
 344                 be = bl_find_get_extent(bl, isect, NULL);
 345                 BUG_ON(!be); /* FIXME */
 346                 len = min(end, be->be_f_offset + be->be_length) - isect;
 347                 if (be->be_state == PNFS_BLOCK_INVALID_DATA)
 348                         bl_mark_for_commit(be, isect, len); /* What if fails? */
 349                 isect += len;
 350                 bl_put_extent(be);
 351         }
 352 }
 353
 354 static void bl_end_io_write_zero(struct bio *bio, int err)
 355 {
 356         struct parallel_io *par = bio->bi_private;
 357         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 358         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
 359         struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
 360
 361         do {
 362                 struct page *page = bvec->bv_page;
 363
 364                 if (--bvec >= bio->bi_io_vec)
 365                         prefetchw(&bvec->bv_page->flags);
 366                 /* This is the zeroing page we added */
 367                 end_page_writeback(page);
 368                 page_cache_release(page);
 369         } while (bvec >= bio->bi_io_vec);
 370         if (!uptodate) {
 371                 if (!wdata->pnfs_error)
 372                         wdata->pnfs_error = -EIO;
 373                 bl_set_lo_fail(wdata->lseg);
 374         }
 375         bio_put(bio);
 376         put_parallel(par);
 377 }
 378
 379 /* This is basically copied from mpage_end_io_read */
 380 static void bl_end_io_write(struct bio *bio, int err)
 381 {
 382         struct parallel_io *par = bio->bi_private;
 383         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 384         struct nfs_write_data *wdata = (struct nfs_write_data *)par->data;
 385
 386         if (!uptodate) {
 387                 if (!wdata->pnfs_error)
 388                         wdata->pnfs_error = -EIO;
 389                 bl_set_lo_fail(wdata->lseg);
 390         }
 391         bio_put(bio);
 392         put_parallel(par);
 393 }
 394
 395 /* Function scheduled for call during bl_end_par_io_write,
 396  * it marks sectors as written and extends the commitlist.
 397  */
 398 static void bl_write_cleanup(struct work_struct *work)
 399 {
 400         struct rpc_task *task;
 401         struct nfs_write_data *wdata;
 402         dprintk("%s enter\n", __func__);
 403         task = container_of(work, struct rpc_task, u.tk_work);
 404         wdata = container_of(task, struct nfs_write_data, task);
 405         if (!wdata->pnfs_error) {
 406                 /* Marks for LAYOUTCOMMIT */
 407                 mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
 408                                      wdata->args.offset, wdata->args.count);
 409         }
 410         pnfs_ld_write_done(wdata);
 411 }
 412
 413 /* Called when last of bios associated with a bl_write_pagelist call finishes */
 414 static void bl_end_par_io_write(void *data)
 415 {
 416         struct nfs_write_data *wdata = data;
 417
 418         wdata->task.tk_status = 0;
 419         wdata->verf.committed = NFS_FILE_SYNC;
 420         INIT_WORK(&wdata->task.u.tk_work, bl_write_cleanup);
 421         schedule_work(&wdata->task.u.tk_work);
 422 }
 423
 424 /* FIXME STUB - mark intersection of layout and page as bad, so is not
 425  * used again.
 426  */
 427 static void mark_bad_read(void)
 428 {
 429         return;
 430 }
 431
 432 /*
 433  * map_block:  map a requested I/0 block (isect) into an offset in the LVM
 434  * block_device
 435  */
 436 static void
 437 map_block(struct buffer_head *bh, sector_t isect, struct pnfs_block_extent *be)
 438 {
 439         dprintk("%s enter be=%p\n", __func__, be);
 440
 441         set_buffer_mapped(bh);
 442         bh->b_bdev = be->be_mdev;
 443         bh->b_blocknr = (isect - be->be_f_offset + be->be_v_offset) >>
 444             (be->be_mdev->bd_inode->i_blkbits - SECTOR_SHIFT);
 445
 446         dprintk("%s isect %llu, bh->b_blocknr %ld, using bsize %Zd\n",
 447                 __func__, (unsigned long long)isect, (long)bh->b_blocknr,
 448                 bh->b_size);
 449         return;
 450 }
 451
 452 /* Given an unmapped page, zero it or read in page for COW, page is locked
 453  * by caller.
 454  */
 455 static int
 456 init_page_for_write(struct page *page, struct pnfs_block_extent *cow_read)
 457 {
 458         struct buffer_head *bh = NULL;
 459         int ret = 0;
 460         sector_t isect;
 461
 462         dprintk("%s enter, %p\n", __func__, page);
 463         BUG_ON(PageUptodate(page));
 464         if (!cow_read) {
 465                 zero_user_segment(page, 0, PAGE_SIZE);
 466                 SetPageUptodate(page);
 467                 goto cleanup;
 468         }
 469
 470         bh = alloc_page_buffers(page, PAGE_CACHE_SIZE, 0);
 471         if (!bh) {
 472                 ret = -ENOMEM;
 473                 goto cleanup;
 474         }
 475
 476         isect = (sector_t) page->index << PAGE_CACHE_SECTOR_SHIFT;
 477         map_block(bh, isect, cow_read);
 478         if (!bh_uptodate_or_lock(bh))
 479                 ret = bh_submit_read(bh);
 480         if (ret)
 481                 goto cleanup;
 482         SetPageUptodate(page);
 483
 484 cleanup:
 485         bl_put_extent(cow_read);
 486         if (bh)
 487                 free_buffer_head(bh);
 488         if (ret) {
 489                 /* Need to mark layout with bad read...should now
 490                  * just use nfs4 for reads and writes.
 491                  */
 492                 mark_bad_read();
 493         }
 494         return ret;
 495 }
 496
 497 static enum pnfs_try_status
 498 bl_write_pagelist(struct nfs_write_data *wdata, int sync)
 499 {
 500         int i, ret, npg_zero, pg_index, last = 0;
 501         struct bio *bio = NULL;
 502         struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 503         sector_t isect, last_isect = 0, extent_length = 0;
 504         struct parallel_io *par;
 505         loff_t offset = wdata->args.offset;
 506         size_t count = wdata->args.count;
 507         struct page **pages = wdata->args.pages;
 508         struct page *page;
 509         pgoff_t index;
 510         u64 temp;
 511         int npg_per_block =
 512             NFS_SERVER(wdata->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
 513
 514         dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
 515         /* At this point, wdata->pages is a (sequential) list of nfs_pages.
 516          * We want to write each, and if there is an error set pnfs_error
 517          * to have it redone using nfs.
 518          */
 519         par = alloc_parallel(wdata);
 520         if (!par)
 521                 return PNFS_NOT_ATTEMPTED;
 522         par->call_ops = *wdata->mds_ops;
 523         par->call_ops.rpc_call_done = bl_rpc_do_nothing;
 524         par->pnfs_callback = bl_end_par_io_write;
 525         /* At this point, have to be more careful with error handling */
 526
 527         isect = (sector_t) ((offset & (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 528         be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg), isect, &cow_read);
 529         if (!be || !is_writable(be, isect)) {
 530                 dprintk("%s no matching extents!\n", __func__);
 531                 wdata->pnfs_error = -EINVAL;
 532                 goto out;
 533         }
 534
 535         /* First page inside INVALID extent */
 536         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 537                 temp = offset >> PAGE_CACHE_SHIFT;
 538                 npg_zero = do_div(temp, npg_per_block);
 539                 isect = (sector_t) (((offset - npg_zero * PAGE_CACHE_SIZE) &
 540                                      (long)PAGE_CACHE_MASK) >> SECTOR_SHIFT);
 541                 extent_length = be->be_length - (isect - be->be_f_offset);
 542
 543 fill_invalid_ext:
 544                 dprintk("%s need to zero %d pages\n", __func__, npg_zero);
 545                 for (;npg_zero > 0; npg_zero--) {
 546                         /* page ref released in bl_end_io_write_zero */
 547                         index = isect >> PAGE_CACHE_SECTOR_SHIFT;
 548                         dprintk("%s zero %dth page: index %lu isect %llu\n",
 549                                 __func__, npg_zero, index,
 550                                 (unsigned long long)isect);
 551                         page =
 552                             find_or_create_page(wdata->inode->i_mapping, index,
 553                                                 GFP_NOFS);
 554                         if (!page) {
 555                                 dprintk("%s oom\n", __func__);
 556                                 wdata->pnfs_error = -ENOMEM;
 557                                 goto out;
 558                         }
 559
 560                         /* PageDirty: Other will write this out
 561                          * PageWriteback: Other is writing this out
 562                          * PageUptodate: It was read before
 563                          * sector_initialized: already written out
 564                          */
 565                         if (PageDirty(page) || PageWriteback(page) ||
 566                             bl_is_sector_init(be->be_inval, isect)) {
 567                                 print_page(page);
 568                                 unlock_page(page);
 569                                 page_cache_release(page);
 570                                 goto next_page;
 571                         }
 572                         if (!PageUptodate(page)) {
 573                                 /* New page, readin or zero it */
 574                                 init_page_for_write(page, cow_read);
 575                         }
 576                         set_page_writeback(page);
 577                         unlock_page(page);
 578
 579                         ret = bl_mark_sectors_init(be->be_inval, isect,
 580                                                        PAGE_CACHE_SECTORS,
 581                                                        NULL);
 582                         if (unlikely(ret)) {
 583                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 584                                         __func__, ret);
 585                                 end_page_writeback(page);
 586                                 page_cache_release(page);
 587                                 wdata->pnfs_error = ret;
 588                                 goto out;
 589                         }
 590                         bio = bl_add_page_to_bio(bio, npg_zero, WRITE,
 591                                                  isect, page, be,
 592                                                  bl_end_io_write_zero, par);
 593                         if (IS_ERR(bio)) {
 594                                 wdata->pnfs_error = PTR_ERR(bio);
 595                                 goto out;
 596                         }
 597                         /* FIXME: This should be done in bi_end_io */
 598                         mark_extents_written(BLK_LSEG2EXT(wdata->lseg),
 599                                              page->index << PAGE_CACHE_SHIFT,
 600                                              PAGE_CACHE_SIZE);
 601 next_page:
 602                         isect += PAGE_CACHE_SECTORS;
 603                         extent_length -= PAGE_CACHE_SECTORS;
 604                 }
 605                 if (last)
 606                         goto write_done;
 607         }
 608         bio = bl_submit_bio(WRITE, bio);
 609
 610         /* Middle pages */
 611         pg_index = wdata->args.pgbase >> PAGE_CACHE_SHIFT;
 612         for (i = pg_index; i < wdata->npages; i++) {
 613                 if (!extent_length) {
 614                         /* We've used up the previous extent */
 615                         bl_put_extent(be);
 616                         bio = bl_submit_bio(WRITE, bio);
 617                         /* Get the next one */
 618                         be = bl_find_get_extent(BLK_LSEG2EXT(wdata->lseg),
 619                                              isect, NULL);
 620                         if (!be || !is_writable(be, isect)) {
 621                                 wdata->pnfs_error = -EINVAL;
 622                                 goto out;
 623                         }
 624                         extent_length = be->be_length -
 625                             (isect - be->be_f_offset);
 626                 }
 627                 if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 628                         ret = bl_mark_sectors_init(be->be_inval, isect,
 629                                                        PAGE_CACHE_SECTORS,
 630                                                        NULL);
 631                         if (unlikely(ret)) {
 632                                 dprintk("%s bl_mark_sectors_init fail %d\n",
 633                                         __func__, ret);
 634                                 wdata->pnfs_error = ret;
 635                                 goto out;
 636                         }
 637                 }
 638                 bio = bl_add_page_to_bio(bio, wdata->npages - i, WRITE,
 639                                          isect, pages[i], be,
 640                                          bl_end_io_write, par);
 641                 if (IS_ERR(bio)) {
 642                         wdata->pnfs_error = PTR_ERR(bio);
 643                         goto out;
 644                 }
 645                 isect += PAGE_CACHE_SECTORS;
 646                 last_isect = isect;
 647                 extent_length -= PAGE_CACHE_SECTORS;
 648         }
 649
 650         /* Last page inside INVALID extent */
 651         if (be->be_state == PNFS_BLOCK_INVALID_DATA) {
 652                 bio = bl_submit_bio(WRITE, bio);
 653                 temp = last_isect >> PAGE_CACHE_SECTOR_SHIFT;
 654                 npg_zero = npg_per_block - do_div(temp, npg_per_block);
 655                 if (npg_zero < npg_per_block) {
 656                         last = 1;
 657                         goto fill_invalid_ext;
 658                 }
 659         }
 660
 661 write_done:
 662         wdata->res.count = (last_isect << SECTOR_SHIFT) - (offset);
 663         if (count < wdata->res.count) {
 664                 wdata->res.count = count;
 665         }
 666 out:
 667         bl_put_extent(be);
 668         bl_submit_bio(WRITE, bio);
 669         put_parallel(par);
 670         return PNFS_ATTEMPTED;
 671 }
 672
 673 /* FIXME - range ignored */
 674 static void
 675 release_extents(struct pnfs_block_layout *bl, struct pnfs_layout_range *range)
 676 {
 677         int i;
 678         struct pnfs_block_extent *be;
 679
 680         spin_lock(&bl->bl_ext_lock);
 681         for (i = 0; i < EXTENT_LISTS; i++) {
 682                 while (!list_empty(&bl->bl_extents[i])) {
 683                         be = list_first_entry(&bl->bl_extents[i],
 684                                               struct pnfs_block_extent,
 685                                               be_node);
 686                         list_del(&be->be_node);
 687                         bl_put_extent(be);
 688                 }
 689         }
 690         spin_unlock(&bl->bl_ext_lock);
 691 }
 692
 693 static void
 694 release_inval_marks(struct pnfs_inval_markings *marks)
 695 {
 696         struct pnfs_inval_tracking *pos, *temp;
 697
 698         list_for_each_entry_safe(pos, temp, &marks->im_tree.mtt_stub, it_link) {
 699                 list_del(&pos->it_link);
 700                 kfree(pos);
 701         }
 702         return;
 703 }
 704
 705 static void bl_free_layout_hdr(struct pnfs_layout_hdr *lo)
 706 {
 707         struct pnfs_block_layout *bl = BLK_LO2EXT(lo);
 708
 709         dprintk("%s enter\n", __func__);
 710         release_extents(bl, NULL);
 711         release_inval_marks(&bl->bl_inval);
 712         kfree(bl);
 713 }
 714
 715 static struct pnfs_layout_hdr *bl_alloc_layout_hdr(struct inode *inode,
 716                                                    gfp_t gfp_flags)
 717 {
 718         struct pnfs_block_layout *bl;
 719
 720         dprintk("%s enter\n", __func__);
 721         bl = kzalloc(sizeof(*bl), gfp_flags);
 722         if (!bl)
 723                 return NULL;
 724         spin_lock_init(&bl->bl_ext_lock);
 725         INIT_LIST_HEAD(&bl->bl_extents[0]);
 726         INIT_LIST_HEAD(&bl->bl_extents[1]);
 727         INIT_LIST_HEAD(&bl->bl_commit);
 728         INIT_LIST_HEAD(&bl->bl_committing);
 729         bl->bl_count = 0;
 730         bl->bl_blocksize = NFS_SERVER(inode)->pnfs_blksize >> SECTOR_SHIFT;
 731         BL_INIT_INVAL_MARKS(&bl->bl_inval, bl->bl_blocksize);
 732         return &bl->bl_layout;
 733 }
 734
 735 static void bl_free_lseg(struct pnfs_layout_segment *lseg)
 736 {
 737         dprintk("%s enter\n", __func__);
 738         kfree(lseg);
 739 }
 740
 741 /* We pretty much ignore lseg, and store all data layout wide, so we
 742  * can correctly merge.
 743  */
 744 static struct pnfs_layout_segment *bl_alloc_lseg(struct pnfs_layout_hdr *lo,
 745                                                  struct nfs4_layoutget_res *lgr,
 746                                                  gfp_t gfp_flags)
 747 {
 748         struct pnfs_layout_segment *lseg;
 749         int status;
 750
 751         dprintk("%s enter\n", __func__);
 752         lseg = kzalloc(sizeof(*lseg), gfp_flags);
 753         if (!lseg)
 754                 return ERR_PTR(-ENOMEM);
 755         status = nfs4_blk_process_layoutget(lo, lgr, gfp_flags);
 756         if (status) {
 757                 /* We don't want to call the full-blown bl_free_lseg,
 758                  * since on error extents were not touched.
 759                  */
 760                 kfree(lseg);
 761                 return ERR_PTR(status);
 762         }
 763         return lseg;
 764 }
 765
 766 static void
 767 bl_encode_layoutcommit(struct pnfs_layout_hdr *lo, struct xdr_stream *xdr,
 768                        const struct nfs4_layoutcommit_args *arg)
 769 {
 770         dprintk("%s enter\n", __func__);
 771         encode_pnfs_block_layoutupdate(BLK_LO2EXT(lo), xdr, arg);
 772 }
 773
 774 static void
 775 bl_cleanup_layoutcommit(struct nfs4_layoutcommit_data *lcdata)
 776 {
 777         struct pnfs_layout_hdr *lo = NFS_I(lcdata->args.inode)->layout;
 778
 779         dprintk("%s enter\n", __func__);
 780         clean_pnfs_block_layoutupdate(BLK_LO2EXT(lo), &lcdata->args, lcdata->res.status);
 781 }
 782
 783 static void free_blk_mountid(struct block_mount_id *mid)
 784 {
 785         if (mid) {
 786                 struct pnfs_block_dev *dev;
 787                 spin_lock(&mid->bm_lock);
 788                 while (!list_empty(&mid->bm_devlist)) {
 789                         dev = list_first_entry(&mid->bm_devlist,
 790                                                struct pnfs_block_dev,
 791                                                bm_node);
 792                         list_del(&dev->bm_node);
 793                         bl_free_block_dev(dev);
 794                 }
 795                 spin_unlock(&mid->bm_lock);
 796                 kfree(mid);
 797         }
 798 }
 799
 800 /* This is mostly copied from the filelayout's get_device_info function.
 801  * It seems much of this should be at the generic pnfs level.
 802  */
 803 static struct pnfs_block_dev *
 804 nfs4_blk_get_deviceinfo(struct nfs_server *server, const struct nfs_fh *fh,
 805                         struct nfs4_deviceid *d_id)
 806 {
 807         struct pnfs_device *dev;
 808         struct pnfs_block_dev *rv = NULL;
 809         u32 max_resp_sz;
 810         int max_pages;
 811         struct page **pages = NULL;
 812         int i, rc;
 813
 814         /*
 815          * Use the session max response size as the basis for setting
 816          * GETDEVICEINFO's maxcount
 817          */
 818         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
 819         max_pages = max_resp_sz >> PAGE_SHIFT;
 820         dprintk("%s max_resp_sz %u max_pages %d\n",
 821                 __func__, max_resp_sz, max_pages);
 822
 823         dev = kmalloc(sizeof(*dev), GFP_NOFS);
 824         if (!dev) {
 825                 dprintk("%s kmalloc failed\n", __func__);
 826                 return NULL;
 827         }
 828
 829         pages = kzalloc(max_pages * sizeof(struct page *), GFP_NOFS);
 830         if (pages == NULL) {
 831                 kfree(dev);
 832                 return NULL;
 833         }
 834         for (i = 0; i < max_pages; i++) {
 835                 pages[i] = alloc_page(GFP_NOFS);
 836                 if (!pages[i])
 837                         goto out_free;
 838         }
 839
 840         memcpy(&dev->dev_id, d_id, sizeof(*d_id));
 841         dev->layout_type = LAYOUT_BLOCK_VOLUME;
 842         dev->pages = pages;
 843         dev->pgbase = 0;
 844         dev->pglen = PAGE_SIZE * max_pages;
 845         dev->mincount = 0;
 846
 847         dprintk("%s: dev_id: %s\n", __func__, dev->dev_id.data);
 848         rc = nfs4_proc_getdeviceinfo(server, dev);
 849         dprintk("%s getdevice info returns %d\n", __func__, rc);
 850         if (rc)
 851                 goto out_free;
 852
 853         rv = nfs4_blk_decode_device(server, dev);
 854  out_free:
 855         for (i = 0; i < max_pages; i++)
 856                 __free_page(pages[i]);
 857         kfree(pages);
 858         kfree(dev);
 859         return rv;
 860 }
 861
 862 static int
 863 bl_set_layoutdriver(struct nfs_server *server, const struct nfs_fh *fh)
 864 {
 865         struct block_mount_id *b_mt_id = NULL;
 866         struct pnfs_devicelist *dlist = NULL;
 867         struct pnfs_block_dev *bdev;
 868         LIST_HEAD(block_disklist);
 869         int status = 0, i;
 870
 871         dprintk("%s enter\n", __func__);
 872
 873         if (server->pnfs_blksize == 0) {
 874                 dprintk("%s Server did not return blksize\n", __func__);
 875                 return -EINVAL;
 876         }
 877         b_mt_id = kzalloc(sizeof(struct block_mount_id), GFP_NOFS);
 878         if (!b_mt_id) {
 879                 status = -ENOMEM;
 880                 goto out_error;
 881         }
 882         /* Initialize nfs4 block layout mount id */
 883         spin_lock_init(&b_mt_id->bm_lock);
 884         INIT_LIST_HEAD(&b_mt_id->bm_devlist);
 885
 886         dlist = kmalloc(sizeof(struct pnfs_devicelist), GFP_NOFS);
 887         if (!dlist) {
 888                 status = -ENOMEM;
 889                 goto out_error;
 890         }
 891         dlist->eof = 0;
 892         while (!dlist->eof) {
 893                 status = nfs4_proc_getdevicelist(server, fh, dlist);
 894                 if (status)
 895                         goto out_error;
 896                 dprintk("%s GETDEVICELIST numdevs=%i, eof=%i\n",
 897                         __func__, dlist->num_devs, dlist->eof);
 898                 for (i = 0; i < dlist->num_devs; i++) {
 899                         bdev = nfs4_blk_get_deviceinfo(server, fh,
 900                                                        &dlist->dev_id[i]);
 901                         if (!bdev) {
 902                                 status = -ENODEV;
 903                                 goto out_error;
 904                         }
 905                         spin_lock(&b_mt_id->bm_lock);
 906                         list_add(&bdev->bm_node, &b_mt_id->bm_devlist);
 907                         spin_unlock(&b_mt_id->bm_lock);
 908                 }
 909         }
 910         dprintk("%s SUCCESS\n", __func__);
 911         server->pnfs_ld_data = b_mt_id;
 912
 913  out_return:
 914         kfree(dlist);
 915         return status;
 916
 917  out_error:
 918         free_blk_mountid(b_mt_id);
 919         goto out_return;
 920 }
 921
 922 static int
 923 bl_clear_layoutdriver(struct nfs_server *server)
 924 {
 925         struct block_mount_id *b_mt_id = server->pnfs_ld_data;
 926
 927         dprintk("%s enter\n", __func__);
 928         free_blk_mountid(b_mt_id);
 929         dprintk("%s RETURNS\n", __func__);
 930         return 0;
 931 }
 932
 933 static const struct nfs_pageio_ops bl_pg_read_ops = {
 934         .pg_init = pnfs_generic_pg_init_read,
 935         .pg_test = pnfs_generic_pg_test,
 936         .pg_doio = pnfs_generic_pg_readpages,
 937 };
 938
 939 static const struct nfs_pageio_ops bl_pg_write_ops = {
 940         .pg_init = pnfs_generic_pg_init_write,
 941         .pg_test = pnfs_generic_pg_test,
 942         .pg_doio = pnfs_generic_pg_writepages,
 943 };
 944
 945 static struct pnfs_layoutdriver_type blocklayout_type = {
 946         .id                             = LAYOUT_BLOCK_VOLUME,
 947         .name                           = "LAYOUT_BLOCK_VOLUME",
 948         .read_pagelist                  = bl_read_pagelist,
 949         .write_pagelist                 = bl_write_pagelist,
 950         .alloc_layout_hdr               = bl_alloc_layout_hdr,
 951         .free_layout_hdr                = bl_free_layout_hdr,
 952         .alloc_lseg                     = bl_alloc_lseg,
 953         .free_lseg                      = bl_free_lseg,
 954         .encode_layoutcommit            = bl_encode_layoutcommit,
 955         .cleanup_layoutcommit           = bl_cleanup_layoutcommit,
 956         .set_layoutdriver               = bl_set_layoutdriver,
 957         .clear_layoutdriver             = bl_clear_layoutdriver,
 958         .pg_read_ops                    = &bl_pg_read_ops,
 959         .pg_write_ops                   = &bl_pg_write_ops,
 960 };
 961
 962 static const struct rpc_pipe_ops bl_upcall_ops = {
 963         .upcall         = bl_pipe_upcall,
 964         .downcall       = bl_pipe_downcall,
 965         .destroy_msg    = bl_pipe_destroy_msg,
 966 };
 967
 968 static int __init nfs4blocklayout_init(void)
 969 {
 970         struct vfsmount *mnt;
 971         struct path path;
 972         int ret;
 973
 974         dprintk("%s: NFSv4 Block Layout Driver Registering...\n", __func__);
 975
 976         ret = pnfs_register_layoutdriver(&blocklayout_type);
 977         if (ret)
 978                 goto out;
 979
 980         init_waitqueue_head(&bl_wq);
 981
 982         mnt = rpc_get_mount();
 983         if (IS_ERR(mnt)) {
 984                 ret = PTR_ERR(mnt);
 985                 goto out_remove;
 986         }
 987
 988         ret = vfs_path_lookup(mnt->mnt_root,
 989                               mnt,
 990                               NFS_PIPE_DIRNAME, 0, &path);
 991         if (ret)
 992                 goto out_remove;
 993
 994         bl_device_pipe = rpc_mkpipe(path.dentry, "blocklayout", NULL,
 995                                     &bl_upcall_ops, 0);
 996         if (IS_ERR(bl_device_pipe)) {
 997                 ret = PTR_ERR(bl_device_pipe);
 998                 goto out_remove;
 999         }
1000 out:
1001         return ret;
1002
1003 out_remove:
1004         pnfs_unregister_layoutdriver(&blocklayout_type);
1005         return ret;
1006 }
1007
1008 static void __exit nfs4blocklayout_exit(void)
1009 {
1010         dprintk("%s: NFSv4 Block Layout Driver Unregistering...\n",
1011                __func__);
1012
1013         pnfs_unregister_layoutdriver(&blocklayout_type);
1014         rpc_unlink(bl_device_pipe);
1015 }
1016
1017 MODULE_ALIAS("nfs-layouttype4-3");
1018
1019 module_init(nfs4blocklayout_init);
1020 module_exit(nfs4blocklayout_exit);