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         page = grab_cache_page_write_begin(inode->i_mapping, index, flags);
 117         if (!page)
 118                 return -ENOMEM;
 119
 120         status = __block_write_begin_int(page, pos, len, NULL, iomap);
 121         if (unlikely(status)) {
 122                 unlock_page(page);
 123                 put_page(page);
 124                 page = NULL;
 125
 126                 iomap_write_failed(inode, pos, len);
 127         }
 128
 129         *pagep = page;
 130         return status;
 131 }
 132
 133 static int
 134 iomap_write_end(struct inode *inode, loff_t pos, unsigned len,
 135                 unsigned copied, struct page *page)
 136 {
 137         int ret;
 138
 139         ret = generic_write_end(NULL, inode->i_mapping, pos, len,
 140                         copied, page, NULL);
 141         if (ret < len)
 142                 iomap_write_failed(inode, pos, len);
 143         return ret;
 144 }
 145
 146 static loff_t
 147 iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 148                 struct iomap *iomap)
 149 {
 150         struct iov_iter *i = data;
 151         long status = 0;
 152         ssize_t written = 0;
 153         unsigned int flags = AOP_FLAG_NOFS;
 154
 155         /*
 156          * Copies from kernel address space cannot fail (NFSD is a big user).
 157          */
 158         if (!iter_is_iovec(i))
 159                 flags |= AOP_FLAG_UNINTERRUPTIBLE;
 160
 161         do {
 162                 struct page *page;
 163                 unsigned long offset;   /* Offset into pagecache page */
 164                 unsigned long bytes;    /* Bytes to write to page */
 165                 size_t copied;          /* Bytes copied from user */
 166
 167                 offset = (pos & (PAGE_SIZE - 1));
 168                 bytes = min_t(unsigned long, PAGE_SIZE - offset,
 169                                                 iov_iter_count(i));
 170 again:
 171                 if (bytes > length)
 172                         bytes = length;
 173
 174                 /*
 175                  * Bring in the user page that we will copy from _first_.
 176                  * Otherwise there's a nasty deadlock on copying from the
 177                  * same page as we're writing to, without it being marked
 178                  * up-to-date.
 179                  *
 180                  * Not only is this an optimisation, but it is also required
 181                  * to check that the address is actually valid, when atomic
 182                  * usercopies are used, below.
 183                  */
 184                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
 185                         status = -EFAULT;
 186                         break;
 187                 }
 188
 189                 status = iomap_write_begin(inode, pos, bytes, flags, &page,
 190                                 iomap);
 191                 if (unlikely(status))
 192                         break;
 193
 194                 if (mapping_writably_mapped(inode->i_mapping))
 195                         flush_dcache_page(page);
 196
 197                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
 198
 199                 flush_dcache_page(page);
 200
 201                 status = iomap_write_end(inode, pos, bytes, copied, page);
 202                 if (unlikely(status < 0))
 203                         break;
 204                 copied = status;
 205
 206                 cond_resched();
 207
 208                 iov_iter_advance(i, copied);
 209                 if (unlikely(copied == 0)) {
 210                         /*
 211                          * If we were unable to copy any data at all, we must
 212                          * fall back to a single segment length write.
 213                          *
 214                          * If we didn't fallback here, we could livelock
 215                          * because not all segments in the iov can be copied at
 216                          * once without a pagefault.
 217                          */
 218                         bytes = min_t(unsigned long, PAGE_SIZE - offset,
 219                                                 iov_iter_single_seg_count(i));
 220                         goto again;
 221                 }
 222                 pos += copied;
 223                 written += copied;
 224                 length -= copied;
 225
 226                 balance_dirty_pages_ratelimited(inode->i_mapping);
 227         } while (iov_iter_count(i) && length);
 228
 229         return written ? written : status;
 230 }
 231
 232 ssize_t
 233 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
 234                 struct iomap_ops *ops)
 235 {
 236         struct inode *inode = iocb->ki_filp->f_mapping->host;
 237         loff_t pos = iocb->ki_pos, ret = 0, written = 0;
 238
 239         while (iov_iter_count(iter)) {
 240                 ret = iomap_apply(inode, pos, iov_iter_count(iter),
 241                                 IOMAP_WRITE, ops, iter, iomap_write_actor);
 242                 if (ret <= 0)
 243                         break;
 244                 pos += ret;
 245                 written += ret;
 246         }
 247
 248         return written ? written : ret;
 249 }
 250 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
 251
 252 static struct page *
 253 __iomap_read_page(struct inode *inode, loff_t offset)
 254 {
 255         struct address_space *mapping = inode->i_mapping;
 256         struct page *page;
 257
 258         page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL);
 259         if (IS_ERR(page))
 260                 return page;
 261         if (!PageUptodate(page)) {
 262                 put_page(page);
 263                 return ERR_PTR(-EIO);
 264         }
 265         return page;
 266 }
 267
 268 static loff_t
 269 iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 270                 struct iomap *iomap)
 271 {
 272         long status = 0;
 273         ssize_t written = 0;
 274
 275         do {
 276                 struct page *page, *rpage;
 277                 unsigned long offset;   /* Offset into pagecache page */
 278                 unsigned long bytes;    /* Bytes to write to page */
 279
 280                 offset = (pos & (PAGE_SIZE - 1));
 281                 bytes = min_t(unsigned long, PAGE_SIZE - offset, length);
 282
 283                 rpage = __iomap_read_page(inode, pos);
 284                 if (IS_ERR(rpage))
 285                         return PTR_ERR(rpage);
 286
 287                 status = iomap_write_begin(inode, pos, bytes,
 288                                 AOP_FLAG_NOFS | AOP_FLAG_UNINTERRUPTIBLE,
 289                                 &page, iomap);
 290                 put_page(rpage);
 291                 if (unlikely(status))
 292                         return status;
 293
 294                 WARN_ON_ONCE(!PageUptodate(page));
 295
 296                 status = iomap_write_end(inode, pos, bytes, bytes, page);
 297                 if (unlikely(status <= 0)) {
 298                         if (WARN_ON_ONCE(status == 0))
 299                                 return -EIO;
 300                         return status;
 301                 }
 302
 303                 cond_resched();
 304
 305                 pos += status;
 306                 written += status;
 307                 length -= status;
 308
 309                 balance_dirty_pages_ratelimited(inode->i_mapping);
 310         } while (length);
 311
 312         return written;
 313 }
 314
 315 int
 316 iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len,
 317                 struct iomap_ops *ops)
 318 {
 319         loff_t ret;
 320
 321         while (len) {
 322                 ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
 323                                 iomap_dirty_actor);
 324                 if (ret <= 0)
 325                         return ret;
 326                 pos += ret;
 327                 len -= ret;
 328         }
 329
 330         return 0;
 331 }
 332 EXPORT_SYMBOL_GPL(iomap_file_dirty);
 333
 334 static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset,
 335                 unsigned bytes, struct iomap *iomap)
 336 {
 337         struct page *page;
 338         int status;
 339
 340         status = iomap_write_begin(inode, pos, bytes,
 341                         AOP_FLAG_UNINTERRUPTIBLE | AOP_FLAG_NOFS, &page, iomap);
 342         if (status)
 343                 return status;
 344
 345         zero_user(page, offset, bytes);
 346         mark_page_accessed(page);
 347
 348         return iomap_write_end(inode, pos, bytes, bytes, page);
 349 }
 350
 351 static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes,
 352                 struct iomap *iomap)
 353 {
 354         sector_t sector = iomap->blkno +
 355                 (((pos & ~(PAGE_SIZE - 1)) - iomap->offset) >> 9);
 356
 357         return __dax_zero_page_range(iomap->bdev, sector, offset, bytes);
 358 }
 359
 360 static loff_t
 361 iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count,
 362                 void *data, struct iomap *iomap)
 363 {
 364         bool *did_zero = data;
 365         loff_t written = 0;
 366         int status;
 367
 368         /* already zeroed?  we're done. */
 369         if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN)
 370                 return count;
 371
 372         do {
 373                 unsigned offset, bytes;
 374
 375                 offset = pos & (PAGE_SIZE - 1); /* Within page */
 376                 bytes = min_t(unsigned, PAGE_SIZE - offset, count);
 377
 378                 if (IS_DAX(inode))
 379                         status = iomap_dax_zero(pos, offset, bytes, iomap);
 380                 else
 381                         status = iomap_zero(inode, pos, offset, bytes, iomap);
 382                 if (status < 0)
 383                         return status;
 384
 385                 pos += bytes;
 386                 count -= bytes;
 387                 written += bytes;
 388                 if (did_zero)
 389                         *did_zero = true;
 390         } while (count > 0);
 391
 392         return written;
 393 }
 394
 395 int
 396 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
 397                 struct iomap_ops *ops)
 398 {
 399         loff_t ret;
 400
 401         while (len > 0) {
 402                 ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
 403                                 ops, did_zero, iomap_zero_range_actor);
 404                 if (ret <= 0)
 405                         return ret;
 406
 407                 pos += ret;
 408                 len -= ret;
 409         }
 410
 411         return 0;
 412 }
 413 EXPORT_SYMBOL_GPL(iomap_zero_range);
 414
 415 int
 416 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
 417                 struct iomap_ops *ops)
 418 {
 419         unsigned blocksize = (1 << inode->i_blkbits);
 420         unsigned off = pos & (blocksize - 1);
 421
 422         /* Block boundary? Nothing to do */
 423         if (!off)
 424                 return 0;
 425         return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
 426 }
 427 EXPORT_SYMBOL_GPL(iomap_truncate_page);
 428
 429 static loff_t
 430 iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
 431                 void *data, struct iomap *iomap)
 432 {
 433         struct page *page = data;
 434         int ret;
 435
 436         ret = __block_write_begin_int(page, pos, length, NULL, iomap);
 437         if (ret)
 438                 return ret;
 439
 440         block_commit_write(page, 0, length);
 441         return length;
 442 }
 443
 444 int iomap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
 445                 struct iomap_ops *ops)
 446 {
 447         struct page *page = vmf->page;
 448         struct inode *inode = file_inode(vma->vm_file);
 449         unsigned long length;
 450         loff_t offset, size;
 451         ssize_t ret;
 452
 453         lock_page(page);
 454         size = i_size_read(inode);
 455         if ((page->mapping != inode->i_mapping) ||
 456             (page_offset(page) > size)) {
 457                 /* We overload EFAULT to mean page got truncated */
 458                 ret = -EFAULT;
 459                 goto out_unlock;
 460         }
 461
 462         /* page is wholly or partially inside EOF */
 463         if (((page->index + 1) << PAGE_SHIFT) > size)
 464                 length = size & ~PAGE_MASK;
 465         else
 466                 length = PAGE_SIZE;
 467
 468         offset = page_offset(page);
 469         while (length > 0) {
 470                 ret = iomap_apply(inode, offset, length, IOMAP_WRITE,
 471                                 ops, page, iomap_page_mkwrite_actor);
 472                 if (unlikely(ret <= 0))
 473                         goto out_unlock;
 474                 offset += ret;
 475                 length -= ret;
 476         }
 477
 478         set_page_dirty(page);
 479         wait_for_stable_page(page);
 480         return 0;
 481 out_unlock:
 482         unlock_page(page);
 483         return ret;
 484 }
 485 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
 486
 487 struct fiemap_ctx {
 488         struct fiemap_extent_info *fi;
 489         struct iomap prev;
 490 };
 491
 492 static int iomap_to_fiemap(struct fiemap_extent_info *fi,
 493                 struct iomap *iomap, u32 flags)
 494 {
 495         switch (iomap->type) {
 496         case IOMAP_HOLE:
 497                 /* skip holes */
 498                 return 0;
 499         case IOMAP_DELALLOC:
 500                 flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN;
 501                 break;
 502         case IOMAP_UNWRITTEN:
 503                 flags |= FIEMAP_EXTENT_UNWRITTEN;
 504                 break;
 505         case IOMAP_MAPPED:
 506                 break;
 507         }
 508
 509         if (iomap->flags & IOMAP_F_MERGED)
 510                 flags |= FIEMAP_EXTENT_MERGED;
 511         if (iomap->flags & IOMAP_F_SHARED)
 512                 flags |= FIEMAP_EXTENT_SHARED;
 513
 514         return fiemap_fill_next_extent(fi, iomap->offset,
 515                         iomap->blkno != IOMAP_NULL_BLOCK ? iomap->blkno << 9: 0,
 516                         iomap->length, flags);
 517
 518 }
 519
 520 static loff_t
 521 iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 522                 struct iomap *iomap)
 523 {
 524         struct fiemap_ctx *ctx = data;
 525         loff_t ret = length;
 526
 527         if (iomap->type == IOMAP_HOLE)
 528                 return length;
 529
 530         ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0);
 531         ctx->prev = *iomap;
 532         switch (ret) {
 533         case 0:         /* success */
 534                 return length;
 535         case 1:         /* extent array full */
 536                 return 0;
 537         default:
 538                 return ret;
 539         }
 540 }
 541
 542 int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi,
 543                 loff_t start, loff_t len, struct iomap_ops *ops)
 544 {
 545         struct fiemap_ctx ctx;
 546         loff_t ret;
 547
 548         memset(&ctx, 0, sizeof(ctx));
 549         ctx.fi = fi;
 550         ctx.prev.type = IOMAP_HOLE;
 551
 552         ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC);
 553         if (ret)
 554                 return ret;
 555
 556         if (fi->fi_flags & FIEMAP_FLAG_SYNC) {
 557                 ret = filemap_write_and_wait(inode->i_mapping);
 558                 if (ret)
 559                         return ret;
 560         }
 561
 562         while (len > 0) {
 563                 ret = iomap_apply(inode, start, len, IOMAP_REPORT, ops, &ctx,
 564                                 iomap_fiemap_actor);
 565                 /* inode with no (attribute) mapping will give ENOENT */
 566                 if (ret == -ENOENT)
 567                         break;
 568                 if (ret < 0)
 569                         return ret;
 570                 if (ret == 0)
 571                         break;
 572
 573                 start += ret;
 574                 len -= ret;
 575         }
 576
 577         if (ctx.prev.type != IOMAP_HOLE) {
 578                 ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST);
 579                 if (ret < 0)
 580                         return ret;
 581         }
 582
 583         return 0;
 584 }
 585 EXPORT_SYMBOL_GPL(iomap_fiemap);