fs/iomap.c

   1 /*
   2  * Copyright (C) 2010 Red Hat, Inc.
   3  * Copyright (c) 2016 Christoph Hellwig.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14 #include <linux/module.h>
  15 #include <linux/compiler.h>
  16 #include <linux/fs.h>
  17 #include <linux/iomap.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/gfp.h>
  20 #include <linux/mm.h>
  21 #include <linux/swap.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/file.h>
  24 #include <linux/uio.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/buffer_head.h>
  27 #include <linux/dax.h>
  28 #include "internal.h"
  29
  30 /*
  31  * Execute a iomap write on a segment of the mapping that spans a
  32  * contiguous range of pages that have identical block mapping state.
  33  *
  34  * This avoids the need to map pages individually, do individual allocations
  35  * for each page and most importantly avoid the need for filesystem specific
  36  * locking per page. Instead, all the operations are amortised over the entire
  37  * range of pages. It is assumed that the filesystems will lock whatever
  38  * resources they require in the iomap_begin call, and release them in the
  39  * iomap_end call.
  40  */
  41 loff_t
  42 iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
  43                 struct iomap_ops *ops, void *data, iomap_actor_t actor)
  44 {
  45         struct iomap iomap = { 0 };
  46         loff_t written = 0, ret;
  47
  48         /*
  49          * Need to map a range from start position for length bytes. This can
  50          * span multiple pages - it is only guaranteed to return a range of a
  51          * single type of pages (e.g. all into a hole, all mapped or all
  52          * unwritten). Failure at this point has nothing to undo.
  53          *
  54          * If allocation is required for this range, reserve the space now so
  55          * that the allocation is guaranteed to succeed later on. Once we copy
  56          * the data into the page cache pages, then we cannot fail otherwise we
  57          * expose transient stale data. If the reserve fails, we can safely
  58          * back out at this point as there is nothing to undo.
  59          */
  60         ret = ops->iomap_begin(inode, pos, length, flags, &iomap);
  61         if (ret)
  62                 return ret;
  63         if (WARN_ON(iomap.offset > pos))
  64                 return -EIO;
  65
  66         /*
  67          * Cut down the length to the one actually provided by the filesystem,
  68          * as it might not be able to give us the whole size that we requested.
  69          */
  70         if (iomap.offset + iomap.length < pos + length)
  71                 length = iomap.offset + iomap.length - pos;
  72
  73         /*
  74          * Now that we have guaranteed that the space allocation will succeed.
  75          * we can do the copy-in page by page without having to worry about
  76          * failures exposing transient data.
  77          */
  78         written = actor(inode, pos, length, data, &iomap);
  79
  80         /*
  81          * Now the data has been copied, commit the range we've copied.  This
  82          * should not fail unless the filesystem has had a fatal error.
  83          */
  84         if (ops->iomap_end) {
  85                 ret = ops->iomap_end(inode, pos, length,
  86                                      written > 0 ? written : 0,
  87                                      flags, &iomap);
  88         }
  89
  90         return written ? written : ret;
  91 }
  92
  93 static void
  94 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
  95 {
  96         loff_t i_size = i_size_read(inode);
  97
  98         /*
  99          * Only truncate newly allocated pages beyoned EOF, even if the
 100          * write started inside the existing inode size.
 101          */
 102         if (pos + len > i_size)
 103                 truncate_pagecache_range(inode, max(pos, i_size), pos + len);
 104 }
 105
 106 static int
 107 iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
 108                 struct page **pagep, struct iomap *iomap)
 109 {
 110         pgoff_t index = pos >> PAGE_SHIFT;
 111         struct page *page;
 112         int status = 0;
 113
 114         BUG_ON(pos + len > iomap->offset + iomap->length);
 115
 116         if (fatal_signal_pending(current))
 117                 return -EINTR;
 118
 119         page = grab_cache_page_write_begin(inode->i_mapping, index, flags);
 120         if (!page)
 121                 return -ENOMEM;
 122
 123         status = __block_write_begin_int(page, pos, len, NULL, iomap);
 124         if (unlikely(status)) {
 125                 unlock_page(page);
 126                 put_page(page);
 127                 page = NULL;
 128
 129                 iomap_write_failed(inode, pos, len);
 130         }
 131
 132         *pagep = page;
 133         return status;
 134 }
 135
 136 static int
 137 iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
 138                 unsigned copied, struct page *page)
 139 {
 140         int ret;
 141
 142         ret = generic_write_end(NULL, inode->i_mapping, pos, len,
 143                         copied, page, NULL);
 144         if (ret < len)
 145                 iomap_write_failed(inode, pos, len);
 146         return ret;
 147 }
 148
 149 static loff_t
 150 iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 151                 struct iomap *iomap)
 152 {
 153         struct iov_iter *i = data;
 154         long status = 0;
 155         ssize_t written = 0;
 156         unsigned int flags = AOP_FLAG_NOFS;
 157
 158         /*
 159          * Copies from kernel address space cannot fail (NFSD is a big user).
 160          */
 161         if (!iter_is_iovec(i))
 162                 flags |= AOP_FLAG_UNINTERRUPTIBLE;
 163
 164         do {
 165                 struct page *page;
 166                 unsigned long offset;   /* Offset into pagecache page */
 167                 unsigned long bytes;    /* Bytes to write to page */
 168                 size_t copied;          /* Bytes copied from user */
 169
 170                 offset = (pos & (PAGE_SIZE - 1));
 171                 bytes = min_t(unsigned long, PAGE_SIZE - offset,
 172                                                 iov_iter_count(i));
 173 again:
 174                 if (bytes > length)
 175                         bytes = length;
 176
 177                 /*
 178                  * Bring in the user page that we will copy from _first_.
 179                  * Otherwise there's a nasty deadlock on copying from the
 180                  * same page as we're writing to, without it being marked
 181                  * up-to-date.
 182                  *
 183                  * Not only is this an optimisation, but it is also required
 184                  * to check that the address is actually valid, when atomic
 185                  * usercopies are used, below.
 186                  */
 187                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
 188                         status = -EFAULT;
 189                         break;
 190                 }
 191
 192                 status = iomap_write_begin(inode, pos, bytes, flags, &page,
 193                                 iomap);
 194                 if (unlikely(status))
 195                         break;
 196
 197                 if (mapping_writably_mapped(inode->i_mapping))
 198                         flush_dcache_page(page);
 199
 200                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
 201
 202                 flush_dcache_page(page);
 203
 204                 status = iomap_write_end(inode, pos, bytes, copied, page);
 205                 if (unlikely(status < 0))
 206                         break;
 207                 copied = status;
 208
 209                 cond_resched();
 210
 211                 iov_iter_advance(i, copied);
 212                 if (unlikely(copied == 0)) {
 213                         /*
 214                          * If we were unable to copy any data at all, we must
 215                          * fall back to a single segment length write.
 216                          *
 217                          * If we didn't fallback here, we could livelock
 218                          * because not all segments in the iov can be copied at
 219                          * once without a pagefault.
 220                          */
 221                         bytes = min_t(unsigned long, PAGE_SIZE - offset,
 222                                                 iov_iter_single_seg_count(i));
 223                         goto again;
 224                 }
 225                 pos += copied;
 226                 written += copied;
 227                 length -= copied;
 228
 229                 balance_dirty_pages_ratelimited(inode->i_mapping);
 230         } while (iov_iter_count(i) && length);
 231
 232         return written ? written : status;
 233 }
 234
 235 ssize_t
 236 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
 237                 struct iomap_ops *ops)
 238 {
 239         struct inode *inode = iocb->ki_filp->f_mapping->host;
 240         loff_t pos = iocb->ki_pos, ret = 0, written = 0;
 241
 242         while (iov_iter_count(iter)) {
 243                 ret = iomap_apply(inode, pos, iov_iter_count(iter),
 244                                 IOMAP_WRITE, ops, iter, iomap_write_actor);
 245                 if (ret <= 0)
 246                         break;
 247                 pos += ret;
 248                 written += ret;
 249         }
 250
 251         return written ? written : ret;
 252 }
 253 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
 254
 255 static struct page *
 256 __iomap_read_page(struct inode *inode, loff_t offset)
 257 {
 258         struct address_space *mapping = inode->i_mapping;
 259         struct page *page;
 260
 261         page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL);
 262         if (IS_ERR(page))
 263                 return page;
 264         if (!PageUptodate(page)) {
 265                 put_page(page);
 266                 return ERR_PTR(-EIO);
 267         }
 268         return page;
 269 }
 270
 271 static loff_t
 272 iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 273                 struct iomap *iomap)
 274 {
 275         long status = 0;
 276         ssize_t written = 0;
 277
 278         do {
 279                 struct page *page, *rpage;
 280                 unsigned long offset;   /* Offset into pagecache page */
 281                 unsigned long bytes;    /* Bytes to write to page */
 282
 283                 offset = (pos & (PAGE_SIZE - 1));
 284                 bytes = min_t(unsigned long, PAGE_SIZE - offset, length);
 285
 286                 rpage = __iomap_read_page(inode, pos);
 287                 if (IS_ERR(rpage))
 288                         return PTR_ERR(rpage);
 289
 290                 status = iomap_write_begin(inode, pos, bytes,
 291                                 AOP_FLAG_NOFS | AOP_FLAG_UNINTERRUPTIBLE,
 292                                 &page, iomap);
 293                 put_page(rpage);
 294                 if (unlikely(status))
 295                         return status;
 296
 297                 WARN_ON_ONCE(!PageUptodate(page));
 298
 299                 status = iomap_write_end(inode, pos, bytes, bytes, page);
 300                 if (unlikely(status <= 0)) {
 301                         if (WARN_ON_ONCE(status == 0))
 302                                 return -EIO;
 303                         return status;
 304                 }
 305
 306                 cond_resched();
 307
 308                 pos += status;
 309                 written += status;
 310                 length -= status;
 311
 312                 balance_dirty_pages_ratelimited(inode->i_mapping);
 313         } while (length);
 314
 315         return written;
 316 }
 317
 318 int
 319 iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len,
 320                 struct iomap_ops *ops)
 321 {
 322         loff_t ret;
 323
 324         while (len) {
 325                 ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
 326                                 iomap_dirty_actor);
 327                 if (ret <= 0)
 328                         return ret;
 329                 pos += ret;
 330                 len -= ret;
 331         }
 332
 333         return 0;
 334 }
 335 EXPORT_SYMBOL_GPL(iomap_file_dirty);
 336
 337 static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
 338                 unsigned bytes, struct iomap *iomap)
 339 {
 340         struct page *page;
 341         int status;
 342
 343         status = iomap_write_begin(inode, pos, bytes,
 344                         AOP_FLAG_UNINTERRUPTIBLE | AOP_FLAG_NOFS, &page, iomap);
 345         if (status)
 346                 return status;
 347
 348         zero_user(page, offset, bytes);
 349         mark_page_accessed(page);
 350
 351         return iomap_write_end(inode, pos, bytes, bytes, page);
 352 }
 353
 354 static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes,
 355                 struct iomap *iomap)
 356 {
 357         sector_t sector = iomap->blkno +
 358                 (((pos & ~(PAGE_SIZE - 1)) - iomap->offset) >> 9);
 359
 360         return __dax_zero_page_range(iomap->bdev, sector, offset, bytes);
 361 }
 362
 363 static loff_t
 364 iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
 365                 void *data, struct iomap *iomap)
 366 {
 367         bool *did_zero = data;
 368         loff_t written = 0;
 369         int status;
 370
 371         /* already zeroed?  we're done. */
 372         if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN)
 373                 return count;
 374
 375         do {
 376                 unsigned offset, bytes;
 377
 378                 offset = pos & (PAGE_SIZE - 1); /* Within page */
 379                 bytes = min_t(unsigned, PAGE_SIZE - offset, count);
 380
 381                 if (IS_DAX(inode))
 382                         status = iomap_dax_zero(pos, offset, bytes, iomap);
 383                 else
 384                         status = iomap_zero(inode, pos, offset, bytes, iomap);
 385                 if (status < 0)
 386                         return status;
 387
 388                 pos += bytes;
 389                 count -= bytes;
 390                 written += bytes;
 391                 if (did_zero)
 392                         *did_zero = true;
 393         } while (count > 0);
 394
 395         return written;
 396 }
 397
 398 int
 399 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
 400                 struct iomap_ops *ops)
 401 {
 402         loff_t ret;
 403
 404         while (len > 0) {
 405                 ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
 406                                 ops, did_zero, iomap_zero_range_actor);
 407                 if (ret <= 0)
 408                         return ret;
 409
 410                 pos += ret;
 411                 len -= ret;
 412         }
 413
 414         return 0;
 415 }
 416 EXPORT_SYMBOL_GPL(iomap_zero_range);
 417
 418 int
 419 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
 420                 struct iomap_ops *ops)
 421 {
 422         unsigned blocksize = (1 << inode->i_blkbits);
 423         unsigned off = pos & (blocksize - 1);
 424
 425         /* Block boundary? Nothing to do */
 426         if (!off)
 427                 return 0;
 428         return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
 429 }
 430 EXPORT_SYMBOL_GPL(iomap_truncate_page);
 431
 432 static loff_t
 433 iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
 434                 void *data, struct iomap *iomap)
 435 {
 436         struct page *page = data;
 437         int ret;
 438
 439         ret = __block_write_begin_int(page, pos, length, NULL, iomap);
 440         if (ret)
 441                 return ret;
 442
 443         block_commit_write(page, 0, length);
 444         return length;
 445 }
 446
 447 int iomap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 448                 struct iomap_ops *ops)
 449 {
 450         struct page *page = vmf->page;
 451         struct inode *inode = file_inode(vma->vm_file);
 452         unsigned long length;
 453         loff_t offset, size;
 454         ssize_t ret;
 455
 456         lock_page(page);
 457         size = i_size_read(inode);
 458         if ((page->mapping != inode->i_mapping) ||
 459             (page_offset(page) > size)) {
 460                 /* We overload EFAULT to mean page got truncated */
 461                 ret = -EFAULT;
 462                 goto out_unlock;
 463         }
 464
 465         /* page is wholly or partially inside EOF */
 466         if (((page->index + 1) << PAGE_SHIFT) > size)
 467                 length = size & ~PAGE_MASK;
 468         else
 469                 length = PAGE_SIZE;
 470
 471         offset = page_offset(page);
 472         while (length > 0) {
 473                 ret = iomap_apply(inode, offset, length, IOMAP_WRITE,
 474                                 ops, page, iomap_page_mkwrite_actor);
 475                 if (unlikely(ret <= 0))
 476                         goto out_unlock;
 477                 offset += ret;
 478                 length -= ret;
 479         }
 480
 481         set_page_dirty(page);
 482         wait_for_stable_page(page);
 483         return 0;
 484 out_unlock:
 485         unlock_page(page);
 486         return ret;
 487 }
 488 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
 489
 490 struct fiemap_ctx {
 491         struct fiemap_extent_info *fi;
 492         struct iomap prev;
 493 };
 494
 495 static int iomap_to_fiemap(struct fiemap_extent_info *fi,
 496                 struct iomap *iomap, u32 flags)
 497 {
 498         switch (iomap->type) {
 499         case IOMAP_HOLE:
 500                 /* skip holes */
 501                 return 0;
 502         case IOMAP_DELALLOC:
 503                 flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN;
 504                 break;
 505         case IOMAP_UNWRITTEN:
 506                 flags |= FIEMAP_EXTENT_UNWRITTEN;
 507                 break;
 508         case IOMAP_MAPPED:
 509                 break;
 510         }
 511
 512         if (iomap->flags & IOMAP_F_MERGED)
 513                 flags |= FIEMAP_EXTENT_MERGED;
 514         if (iomap->flags & IOMAP_F_SHARED)
 515                 flags |= FIEMAP_EXTENT_SHARED;
 516
 517         return fiemap_fill_next_extent(fi, iomap->offset,
 518                         iomap->blkno != IOMAP_NULL_BLOCK ? iomap->blkno << 9: 0,
 519                         iomap->length, flags);
 520
 521 }
 522
 523 static loff_t
 524 iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 525                 struct iomap *iomap)
 526 {
 527         struct fiemap_ctx *ctx = data;
 528         loff_t ret = length;
 529
 530         if (iomap->type == IOMAP_HOLE)
 531                 return length;
 532
 533         ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0);
 534         ctx->prev = *iomap;
 535         switch (ret) {
 536         case 0:         /* success */
 537                 return length;
 538         case 1:         /* extent array full */
 539                 return 0;
 540         default:
 541                 return ret;
 542         }
 543 }
 544
 545 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi,
 546                 loff_t start, loff_t len, struct iomap_ops *ops)
 547 {
 548         struct fiemap_ctx ctx;
 549         loff_t ret;
 550
 551         memset(&ctx, 0, sizeof(ctx));
 552         ctx.fi = fi;
 553         ctx.prev.type = IOMAP_HOLE;
 554
 555         ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC);
 556         if (ret)
 557                 return ret;
 558
 559         if (fi->fi_flags & FIEMAP_FLAG_SYNC) {
 560                 ret = filemap_write_and_wait(inode->i_mapping);
 561                 if (ret)
 562                         return ret;
 563         }
 564
 565         while (len > 0) {
 566                 ret = iomap_apply(inode, start, len, IOMAP_REPORT, ops, &ctx,
 567                                 iomap_fiemap_actor);
 568                 /* inode with no (attribute) mapping will give ENOENT */
 569                 if (ret == -ENOENT)
 570                         break;
 571                 if (ret < 0)
 572                         return ret;
 573                 if (ret == 0)
 574                         break;
 575
 576                 start += ret;
 577                 len -= ret;
 578         }
 579
 580         if (ctx.prev.type != IOMAP_HOLE) {
 581                 ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST);
 582                 if (ret < 0)
 583                         return ret;
 584         }
 585
 586         return 0;
 587 }
 588 EXPORT_SYMBOL_GPL(iomap_fiemap);