Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Anton Altaparmakov | f25dfb5 | 2005-09-08 20:35:33 +0100 | [diff] [blame] | 2 | * file.c - NTFS kernel file operations. Part of the Linux-NTFS project. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3 | * |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 4 | * Copyright (c) 2001-2006 Anton Altaparmakov |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5 | * |
| 6 | * This program/include file is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License as published |
| 8 | * by the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program/include file is distributed in the hope that it will be |
| 12 | * useful, but WITHOUT ANY WARRANTY; without even the implied warranty |
| 13 | * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program (in the main directory of the Linux-NTFS |
| 18 | * distribution in the file COPYING); if not, write to the Free Software |
| 19 | * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 20 | */ |
| 21 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/buffer_head.h> |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 23 | #include <linux/pagemap.h> |
| 24 | #include <linux/pagevec.h> |
| 25 | #include <linux/sched.h> |
| 26 | #include <linux/swap.h> |
| 27 | #include <linux/uio.h> |
| 28 | #include <linux/writeback.h> |
Alexey Dobriyan | e8edc6e | 2007-05-21 01:22:52 +0400 | [diff] [blame] | 29 | #include <linux/sched.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 31 | #include <asm/page.h> |
| 32 | #include <asm/uaccess.h> |
| 33 | |
| 34 | #include "attrib.h" |
| 35 | #include "bitmap.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include "inode.h" |
| 37 | #include "debug.h" |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 38 | #include "lcnalloc.h" |
| 39 | #include "malloc.h" |
| 40 | #include "mft.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #include "ntfs.h" |
| 42 | |
| 43 | /** |
| 44 | * ntfs_file_open - called when an inode is about to be opened |
| 45 | * @vi: inode to be opened |
| 46 | * @filp: file structure describing the inode |
| 47 | * |
| 48 | * Limit file size to the page cache limit on architectures where unsigned long |
| 49 | * is 32-bits. This is the most we can do for now without overflowing the page |
| 50 | * cache page index. Doing it this way means we don't run into problems because |
| 51 | * of existing too large files. It would be better to allow the user to read |
| 52 | * the beginning of the file but I doubt very much anyone is going to hit this |
| 53 | * check on a 32-bit architecture, so there is no point in adding the extra |
| 54 | * complexity required to support this. |
| 55 | * |
| 56 | * On 64-bit architectures, the check is hopefully optimized away by the |
| 57 | * compiler. |
| 58 | * |
| 59 | * After the check passes, just call generic_file_open() to do its work. |
| 60 | */ |
| 61 | static int ntfs_file_open(struct inode *vi, struct file *filp) |
| 62 | { |
| 63 | if (sizeof(unsigned long) < 8) { |
Anton Altaparmakov | d4b9ba7 | 2004-11-17 15:45:08 +0000 | [diff] [blame] | 64 | if (i_size_read(vi) > MAX_LFS_FILESIZE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | return -EFBIG; |
| 66 | } |
| 67 | return generic_file_open(vi, filp); |
| 68 | } |
| 69 | |
| 70 | #ifdef NTFS_RW |
| 71 | |
| 72 | /** |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 73 | * ntfs_attr_extend_initialized - extend the initialized size of an attribute |
| 74 | * @ni: ntfs inode of the attribute to extend |
| 75 | * @new_init_size: requested new initialized size in bytes |
| 76 | * @cached_page: store any allocated but unused page here |
| 77 | * @lru_pvec: lru-buffering pagevec of the caller |
| 78 | * |
| 79 | * Extend the initialized size of an attribute described by the ntfs inode @ni |
| 80 | * to @new_init_size bytes. This involves zeroing any non-sparse space between |
| 81 | * the old initialized size and @new_init_size both in the page cache and on |
Anton Altaparmakov | dda65b9 | 2005-10-24 08:57:59 +0100 | [diff] [blame] | 82 | * disk (if relevant complete pages are already uptodate in the page cache then |
| 83 | * these are simply marked dirty). |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 84 | * |
| 85 | * As a side-effect, the file size (vfs inode->i_size) may be incremented as, |
| 86 | * in the resident attribute case, it is tied to the initialized size and, in |
| 87 | * the non-resident attribute case, it may not fall below the initialized size. |
| 88 | * |
| 89 | * Note that if the attribute is resident, we do not need to touch the page |
| 90 | * cache at all. This is because if the page cache page is not uptodate we |
| 91 | * bring it uptodate later, when doing the write to the mft record since we |
| 92 | * then already have the page mapped. And if the page is uptodate, the |
| 93 | * non-initialized region will already have been zeroed when the page was |
| 94 | * brought uptodate and the region may in fact already have been overwritten |
| 95 | * with new data via mmap() based writes, so we cannot just zero it. And since |
| 96 | * POSIX specifies that the behaviour of resizing a file whilst it is mmap()ped |
| 97 | * is unspecified, we choose not to do zeroing and thus we do not need to touch |
Anton Altaparmakov | dda65b9 | 2005-10-24 08:57:59 +0100 | [diff] [blame] | 98 | * the page at all. For a more detailed explanation see ntfs_truncate() in |
| 99 | * fs/ntfs/inode.c. |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 100 | * |
Anton Altaparmakov | dda65b9 | 2005-10-24 08:57:59 +0100 | [diff] [blame] | 101 | * @cached_page and @lru_pvec are just optimizations for dealing with multiple |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 102 | * pages. |
| 103 | * |
| 104 | * Return 0 on success and -errno on error. In the case that an error is |
| 105 | * encountered it is possible that the initialized size will already have been |
| 106 | * incremented some way towards @new_init_size but it is guaranteed that if |
| 107 | * this is the case, the necessary zeroing will also have happened and that all |
| 108 | * metadata is self-consistent. |
| 109 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 110 | * Locking: i_mutex on the vfs inode corrseponsind to the ntfs inode @ni must be |
Anton Altaparmakov | dda65b9 | 2005-10-24 08:57:59 +0100 | [diff] [blame] | 111 | * held by the caller. |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 112 | */ |
| 113 | static int ntfs_attr_extend_initialized(ntfs_inode *ni, const s64 new_init_size, |
| 114 | struct page **cached_page, struct pagevec *lru_pvec) |
| 115 | { |
| 116 | s64 old_init_size; |
| 117 | loff_t old_i_size; |
| 118 | pgoff_t index, end_index; |
| 119 | unsigned long flags; |
| 120 | struct inode *vi = VFS_I(ni); |
| 121 | ntfs_inode *base_ni; |
| 122 | MFT_RECORD *m = NULL; |
| 123 | ATTR_RECORD *a; |
| 124 | ntfs_attr_search_ctx *ctx = NULL; |
| 125 | struct address_space *mapping; |
| 126 | struct page *page = NULL; |
| 127 | u8 *kattr; |
| 128 | int err; |
| 129 | u32 attr_len; |
| 130 | |
| 131 | read_lock_irqsave(&ni->size_lock, flags); |
| 132 | old_init_size = ni->initialized_size; |
| 133 | old_i_size = i_size_read(vi); |
| 134 | BUG_ON(new_init_size > ni->allocated_size); |
| 135 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 136 | ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " |
| 137 | "old_initialized_size 0x%llx, " |
| 138 | "new_initialized_size 0x%llx, i_size 0x%llx.", |
| 139 | vi->i_ino, (unsigned)le32_to_cpu(ni->type), |
| 140 | (unsigned long long)old_init_size, |
| 141 | (unsigned long long)new_init_size, old_i_size); |
| 142 | if (!NInoAttr(ni)) |
| 143 | base_ni = ni; |
| 144 | else |
| 145 | base_ni = ni->ext.base_ntfs_ino; |
| 146 | /* Use goto to reduce indentation and we need the label below anyway. */ |
| 147 | if (NInoNonResident(ni)) |
| 148 | goto do_non_resident_extend; |
| 149 | BUG_ON(old_init_size != old_i_size); |
| 150 | m = map_mft_record(base_ni); |
| 151 | if (IS_ERR(m)) { |
| 152 | err = PTR_ERR(m); |
| 153 | m = NULL; |
| 154 | goto err_out; |
| 155 | } |
| 156 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 157 | if (unlikely(!ctx)) { |
| 158 | err = -ENOMEM; |
| 159 | goto err_out; |
| 160 | } |
| 161 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 162 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 163 | if (unlikely(err)) { |
| 164 | if (err == -ENOENT) |
| 165 | err = -EIO; |
| 166 | goto err_out; |
| 167 | } |
| 168 | m = ctx->mrec; |
| 169 | a = ctx->attr; |
| 170 | BUG_ON(a->non_resident); |
| 171 | /* The total length of the attribute value. */ |
| 172 | attr_len = le32_to_cpu(a->data.resident.value_length); |
| 173 | BUG_ON(old_i_size != (loff_t)attr_len); |
| 174 | /* |
| 175 | * Do the zeroing in the mft record and update the attribute size in |
| 176 | * the mft record. |
| 177 | */ |
| 178 | kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); |
| 179 | memset(kattr + attr_len, 0, new_init_size - attr_len); |
| 180 | a->data.resident.value_length = cpu_to_le32((u32)new_init_size); |
| 181 | /* Finally, update the sizes in the vfs and ntfs inodes. */ |
| 182 | write_lock_irqsave(&ni->size_lock, flags); |
| 183 | i_size_write(vi, new_init_size); |
| 184 | ni->initialized_size = new_init_size; |
| 185 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 186 | goto done; |
| 187 | do_non_resident_extend: |
| 188 | /* |
| 189 | * If the new initialized size @new_init_size exceeds the current file |
| 190 | * size (vfs inode->i_size), we need to extend the file size to the |
| 191 | * new initialized size. |
| 192 | */ |
| 193 | if (new_init_size > old_i_size) { |
| 194 | m = map_mft_record(base_ni); |
| 195 | if (IS_ERR(m)) { |
| 196 | err = PTR_ERR(m); |
| 197 | m = NULL; |
| 198 | goto err_out; |
| 199 | } |
| 200 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 201 | if (unlikely(!ctx)) { |
| 202 | err = -ENOMEM; |
| 203 | goto err_out; |
| 204 | } |
| 205 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 206 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 207 | if (unlikely(err)) { |
| 208 | if (err == -ENOENT) |
| 209 | err = -EIO; |
| 210 | goto err_out; |
| 211 | } |
| 212 | m = ctx->mrec; |
| 213 | a = ctx->attr; |
| 214 | BUG_ON(!a->non_resident); |
| 215 | BUG_ON(old_i_size != (loff_t) |
| 216 | sle64_to_cpu(a->data.non_resident.data_size)); |
| 217 | a->data.non_resident.data_size = cpu_to_sle64(new_init_size); |
| 218 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 219 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 220 | /* Update the file size in the vfs inode. */ |
| 221 | i_size_write(vi, new_init_size); |
| 222 | ntfs_attr_put_search_ctx(ctx); |
| 223 | ctx = NULL; |
| 224 | unmap_mft_record(base_ni); |
| 225 | m = NULL; |
| 226 | } |
| 227 | mapping = vi->i_mapping; |
| 228 | index = old_init_size >> PAGE_CACHE_SHIFT; |
| 229 | end_index = (new_init_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 230 | do { |
| 231 | /* |
| 232 | * Read the page. If the page is not present, this will zero |
| 233 | * the uninitialized regions for us. |
| 234 | */ |
Pekka Enberg | 090d2b1 | 2006-06-23 02:05:08 -0700 | [diff] [blame] | 235 | page = read_mapping_page(mapping, index, NULL); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 236 | if (IS_ERR(page)) { |
| 237 | err = PTR_ERR(page); |
| 238 | goto init_err_out; |
| 239 | } |
Nick Piggin | 6fe6900 | 2007-05-06 14:49:04 -0700 | [diff] [blame] | 240 | if (unlikely(PageError(page))) { |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 241 | page_cache_release(page); |
| 242 | err = -EIO; |
| 243 | goto init_err_out; |
| 244 | } |
| 245 | /* |
| 246 | * Update the initialized size in the ntfs inode. This is |
| 247 | * enough to make ntfs_writepage() work. |
| 248 | */ |
| 249 | write_lock_irqsave(&ni->size_lock, flags); |
Anton Altaparmakov | 3c6af7f | 2005-11-24 13:41:33 +0000 | [diff] [blame] | 250 | ni->initialized_size = (s64)(index + 1) << PAGE_CACHE_SHIFT; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 251 | if (ni->initialized_size > new_init_size) |
| 252 | ni->initialized_size = new_init_size; |
| 253 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 254 | /* Set the page dirty so it gets written out. */ |
| 255 | set_page_dirty(page); |
| 256 | page_cache_release(page); |
| 257 | /* |
| 258 | * Play nice with the vm and the rest of the system. This is |
| 259 | * very much needed as we can potentially be modifying the |
| 260 | * initialised size from a very small value to a really huge |
| 261 | * value, e.g. |
| 262 | * f = open(somefile, O_TRUNC); |
| 263 | * truncate(f, 10GiB); |
| 264 | * seek(f, 10GiB); |
| 265 | * write(f, 1); |
| 266 | * And this would mean we would be marking dirty hundreds of |
| 267 | * thousands of pages or as in the above example more than |
| 268 | * two and a half million pages! |
| 269 | * |
| 270 | * TODO: For sparse pages could optimize this workload by using |
| 271 | * the FsMisc / MiscFs page bit as a "PageIsSparse" bit. This |
| 272 | * would be set in readpage for sparse pages and here we would |
| 273 | * not need to mark dirty any pages which have this bit set. |
| 274 | * The only caveat is that we have to clear the bit everywhere |
| 275 | * where we allocate any clusters that lie in the page or that |
| 276 | * contain the page. |
| 277 | * |
| 278 | * TODO: An even greater optimization would be for us to only |
| 279 | * call readpage() on pages which are not in sparse regions as |
| 280 | * determined from the runlist. This would greatly reduce the |
| 281 | * number of pages we read and make dirty in the case of sparse |
| 282 | * files. |
| 283 | */ |
| 284 | balance_dirty_pages_ratelimited(mapping); |
| 285 | cond_resched(); |
| 286 | } while (++index < end_index); |
| 287 | read_lock_irqsave(&ni->size_lock, flags); |
| 288 | BUG_ON(ni->initialized_size != new_init_size); |
| 289 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 290 | /* Now bring in sync the initialized_size in the mft record. */ |
| 291 | m = map_mft_record(base_ni); |
| 292 | if (IS_ERR(m)) { |
| 293 | err = PTR_ERR(m); |
| 294 | m = NULL; |
| 295 | goto init_err_out; |
| 296 | } |
| 297 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 298 | if (unlikely(!ctx)) { |
| 299 | err = -ENOMEM; |
| 300 | goto init_err_out; |
| 301 | } |
| 302 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 303 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 304 | if (unlikely(err)) { |
| 305 | if (err == -ENOENT) |
| 306 | err = -EIO; |
| 307 | goto init_err_out; |
| 308 | } |
| 309 | m = ctx->mrec; |
| 310 | a = ctx->attr; |
| 311 | BUG_ON(!a->non_resident); |
| 312 | a->data.non_resident.initialized_size = cpu_to_sle64(new_init_size); |
| 313 | done: |
| 314 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 315 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 316 | if (ctx) |
| 317 | ntfs_attr_put_search_ctx(ctx); |
| 318 | if (m) |
| 319 | unmap_mft_record(base_ni); |
| 320 | ntfs_debug("Done, initialized_size 0x%llx, i_size 0x%llx.", |
| 321 | (unsigned long long)new_init_size, i_size_read(vi)); |
| 322 | return 0; |
| 323 | init_err_out: |
| 324 | write_lock_irqsave(&ni->size_lock, flags); |
| 325 | ni->initialized_size = old_init_size; |
| 326 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 327 | err_out: |
| 328 | if (ctx) |
| 329 | ntfs_attr_put_search_ctx(ctx); |
| 330 | if (m) |
| 331 | unmap_mft_record(base_ni); |
| 332 | ntfs_debug("Failed. Returning error code %i.", err); |
| 333 | return err; |
| 334 | } |
| 335 | |
| 336 | /** |
| 337 | * ntfs_fault_in_pages_readable - |
| 338 | * |
| 339 | * Fault a number of userspace pages into pagetables. |
| 340 | * |
| 341 | * Unlike include/linux/pagemap.h::fault_in_pages_readable(), this one copes |
| 342 | * with more than two userspace pages as well as handling the single page case |
| 343 | * elegantly. |
| 344 | * |
| 345 | * If you find this difficult to understand, then think of the while loop being |
| 346 | * the following code, except that we do without the integer variable ret: |
| 347 | * |
| 348 | * do { |
| 349 | * ret = __get_user(c, uaddr); |
| 350 | * uaddr += PAGE_SIZE; |
| 351 | * } while (!ret && uaddr < end); |
| 352 | * |
| 353 | * Note, the final __get_user() may well run out-of-bounds of the user buffer, |
| 354 | * but _not_ out-of-bounds of the page the user buffer belongs to, and since |
| 355 | * this is only a read and not a write, and since it is still in the same page, |
| 356 | * it should not matter and this makes the code much simpler. |
| 357 | */ |
| 358 | static inline void ntfs_fault_in_pages_readable(const char __user *uaddr, |
| 359 | int bytes) |
| 360 | { |
| 361 | const char __user *end; |
| 362 | volatile char c; |
| 363 | |
| 364 | /* Set @end to the first byte outside the last page we care about. */ |
| 365 | end = (const char __user*)PAGE_ALIGN((ptrdiff_t __user)uaddr + bytes); |
| 366 | |
| 367 | while (!__get_user(c, uaddr) && (uaddr += PAGE_SIZE, uaddr < end)) |
| 368 | ; |
| 369 | } |
| 370 | |
| 371 | /** |
| 372 | * ntfs_fault_in_pages_readable_iovec - |
| 373 | * |
| 374 | * Same as ntfs_fault_in_pages_readable() but operates on an array of iovecs. |
| 375 | */ |
| 376 | static inline void ntfs_fault_in_pages_readable_iovec(const struct iovec *iov, |
| 377 | size_t iov_ofs, int bytes) |
| 378 | { |
| 379 | do { |
| 380 | const char __user *buf; |
| 381 | unsigned len; |
| 382 | |
| 383 | buf = iov->iov_base + iov_ofs; |
| 384 | len = iov->iov_len - iov_ofs; |
| 385 | if (len > bytes) |
| 386 | len = bytes; |
| 387 | ntfs_fault_in_pages_readable(buf, len); |
| 388 | bytes -= len; |
| 389 | iov++; |
| 390 | iov_ofs = 0; |
| 391 | } while (bytes); |
| 392 | } |
| 393 | |
| 394 | /** |
| 395 | * __ntfs_grab_cache_pages - obtain a number of locked pages |
| 396 | * @mapping: address space mapping from which to obtain page cache pages |
| 397 | * @index: starting index in @mapping at which to begin obtaining pages |
| 398 | * @nr_pages: number of page cache pages to obtain |
| 399 | * @pages: array of pages in which to return the obtained page cache pages |
| 400 | * @cached_page: allocated but as yet unused page |
| 401 | * @lru_pvec: lru-buffering pagevec of caller |
| 402 | * |
| 403 | * Obtain @nr_pages locked page cache pages from the mapping @maping and |
| 404 | * starting at index @index. |
| 405 | * |
| 406 | * If a page is newly created, increment its refcount and add it to the |
| 407 | * caller's lru-buffering pagevec @lru_pvec. |
| 408 | * |
| 409 | * This is the same as mm/filemap.c::__grab_cache_page(), except that @nr_pages |
| 410 | * are obtained at once instead of just one page and that 0 is returned on |
| 411 | * success and -errno on error. |
| 412 | * |
| 413 | * Note, the page locks are obtained in ascending page index order. |
| 414 | */ |
| 415 | static inline int __ntfs_grab_cache_pages(struct address_space *mapping, |
| 416 | pgoff_t index, const unsigned nr_pages, struct page **pages, |
| 417 | struct page **cached_page, struct pagevec *lru_pvec) |
| 418 | { |
| 419 | int err, nr; |
| 420 | |
| 421 | BUG_ON(!nr_pages); |
| 422 | err = nr = 0; |
| 423 | do { |
| 424 | pages[nr] = find_lock_page(mapping, index); |
| 425 | if (!pages[nr]) { |
| 426 | if (!*cached_page) { |
| 427 | *cached_page = page_cache_alloc(mapping); |
| 428 | if (unlikely(!*cached_page)) { |
| 429 | err = -ENOMEM; |
| 430 | goto err_out; |
| 431 | } |
| 432 | } |
| 433 | err = add_to_page_cache(*cached_page, mapping, index, |
| 434 | GFP_KERNEL); |
| 435 | if (unlikely(err)) { |
| 436 | if (err == -EEXIST) |
| 437 | continue; |
| 438 | goto err_out; |
| 439 | } |
| 440 | pages[nr] = *cached_page; |
| 441 | page_cache_get(*cached_page); |
| 442 | if (unlikely(!pagevec_add(lru_pvec, *cached_page))) |
| 443 | __pagevec_lru_add(lru_pvec); |
| 444 | *cached_page = NULL; |
| 445 | } |
| 446 | index++; |
| 447 | nr++; |
| 448 | } while (nr < nr_pages); |
| 449 | out: |
| 450 | return err; |
| 451 | err_out: |
| 452 | while (nr > 0) { |
| 453 | unlock_page(pages[--nr]); |
| 454 | page_cache_release(pages[nr]); |
| 455 | } |
| 456 | goto out; |
| 457 | } |
| 458 | |
| 459 | static inline int ntfs_submit_bh_for_read(struct buffer_head *bh) |
| 460 | { |
| 461 | lock_buffer(bh); |
| 462 | get_bh(bh); |
| 463 | bh->b_end_io = end_buffer_read_sync; |
| 464 | return submit_bh(READ, bh); |
| 465 | } |
| 466 | |
| 467 | /** |
| 468 | * ntfs_prepare_pages_for_non_resident_write - prepare pages for receiving data |
| 469 | * @pages: array of destination pages |
| 470 | * @nr_pages: number of pages in @pages |
| 471 | * @pos: byte position in file at which the write begins |
| 472 | * @bytes: number of bytes to be written |
| 473 | * |
| 474 | * This is called for non-resident attributes from ntfs_file_buffered_write() |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 475 | * with i_mutex held on the inode (@pages[0]->mapping->host). There are |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 476 | * @nr_pages pages in @pages which are locked but not kmap()ped. The source |
| 477 | * data has not yet been copied into the @pages. |
| 478 | * |
| 479 | * Need to fill any holes with actual clusters, allocate buffers if necessary, |
| 480 | * ensure all the buffers are mapped, and bring uptodate any buffers that are |
| 481 | * only partially being written to. |
| 482 | * |
| 483 | * If @nr_pages is greater than one, we are guaranteed that the cluster size is |
| 484 | * greater than PAGE_CACHE_SIZE, that all pages in @pages are entirely inside |
| 485 | * the same cluster and that they are the entirety of that cluster, and that |
| 486 | * the cluster is sparse, i.e. we need to allocate a cluster to fill the hole. |
| 487 | * |
| 488 | * i_size is not to be modified yet. |
| 489 | * |
| 490 | * Return 0 on success or -errno on error. |
| 491 | */ |
| 492 | static int ntfs_prepare_pages_for_non_resident_write(struct page **pages, |
| 493 | unsigned nr_pages, s64 pos, size_t bytes) |
| 494 | { |
| 495 | VCN vcn, highest_vcn = 0, cpos, cend, bh_cpos, bh_cend; |
| 496 | LCN lcn; |
| 497 | s64 bh_pos, vcn_len, end, initialized_size; |
| 498 | sector_t lcn_block; |
| 499 | struct page *page; |
| 500 | struct inode *vi; |
| 501 | ntfs_inode *ni, *base_ni = NULL; |
| 502 | ntfs_volume *vol; |
| 503 | runlist_element *rl, *rl2; |
| 504 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; |
| 505 | ntfs_attr_search_ctx *ctx = NULL; |
| 506 | MFT_RECORD *m = NULL; |
| 507 | ATTR_RECORD *a = NULL; |
| 508 | unsigned long flags; |
| 509 | u32 attr_rec_len = 0; |
| 510 | unsigned blocksize, u; |
| 511 | int err, mp_size; |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 512 | bool rl_write_locked, was_hole, is_retry; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 513 | unsigned char blocksize_bits; |
| 514 | struct { |
| 515 | u8 runlist_merged:1; |
| 516 | u8 mft_attr_mapped:1; |
| 517 | u8 mp_rebuilt:1; |
| 518 | u8 attr_switched:1; |
| 519 | } status = { 0, 0, 0, 0 }; |
| 520 | |
| 521 | BUG_ON(!nr_pages); |
| 522 | BUG_ON(!pages); |
| 523 | BUG_ON(!*pages); |
| 524 | vi = pages[0]->mapping->host; |
| 525 | ni = NTFS_I(vi); |
| 526 | vol = ni->vol; |
| 527 | ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page " |
Anton Altaparmakov | d04bd1f | 2005-10-24 08:41:24 +0100 | [diff] [blame] | 528 | "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.", |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 529 | vi->i_ino, ni->type, pages[0]->index, nr_pages, |
| 530 | (long long)pos, bytes); |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 531 | blocksize = vol->sb->s_blocksize; |
| 532 | blocksize_bits = vol->sb->s_blocksize_bits; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 533 | u = 0; |
| 534 | do { |
| 535 | struct page *page = pages[u]; |
| 536 | /* |
| 537 | * create_empty_buffers() will create uptodate/dirty buffers if |
| 538 | * the page is uptodate/dirty. |
| 539 | */ |
| 540 | if (!page_has_buffers(page)) { |
| 541 | create_empty_buffers(page, blocksize, 0); |
| 542 | if (unlikely(!page_has_buffers(page))) |
| 543 | return -ENOMEM; |
| 544 | } |
| 545 | } while (++u < nr_pages); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 546 | rl_write_locked = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 547 | rl = NULL; |
| 548 | err = 0; |
| 549 | vcn = lcn = -1; |
| 550 | vcn_len = 0; |
| 551 | lcn_block = -1; |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 552 | was_hole = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 553 | cpos = pos >> vol->cluster_size_bits; |
| 554 | end = pos + bytes; |
| 555 | cend = (end + vol->cluster_size - 1) >> vol->cluster_size_bits; |
| 556 | /* |
| 557 | * Loop over each page and for each page over each buffer. Use goto to |
| 558 | * reduce indentation. |
| 559 | */ |
| 560 | u = 0; |
| 561 | do_next_page: |
| 562 | page = pages[u]; |
| 563 | bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; |
| 564 | bh = head = page_buffers(page); |
| 565 | do { |
| 566 | VCN cdelta; |
| 567 | s64 bh_end; |
| 568 | unsigned bh_cofs; |
| 569 | |
| 570 | /* Clear buffer_new on all buffers to reinitialise state. */ |
| 571 | if (buffer_new(bh)) |
| 572 | clear_buffer_new(bh); |
| 573 | bh_end = bh_pos + blocksize; |
| 574 | bh_cpos = bh_pos >> vol->cluster_size_bits; |
| 575 | bh_cofs = bh_pos & vol->cluster_size_mask; |
| 576 | if (buffer_mapped(bh)) { |
| 577 | /* |
| 578 | * The buffer is already mapped. If it is uptodate, |
| 579 | * ignore it. |
| 580 | */ |
| 581 | if (buffer_uptodate(bh)) |
| 582 | continue; |
| 583 | /* |
| 584 | * The buffer is not uptodate. If the page is uptodate |
| 585 | * set the buffer uptodate and otherwise ignore it. |
| 586 | */ |
| 587 | if (PageUptodate(page)) { |
| 588 | set_buffer_uptodate(bh); |
| 589 | continue; |
| 590 | } |
| 591 | /* |
| 592 | * Neither the page nor the buffer are uptodate. If |
| 593 | * the buffer is only partially being written to, we |
| 594 | * need to read it in before the write, i.e. now. |
| 595 | */ |
| 596 | if ((bh_pos < pos && bh_end > pos) || |
| 597 | (bh_pos < end && bh_end > end)) { |
| 598 | /* |
| 599 | * If the buffer is fully or partially within |
| 600 | * the initialized size, do an actual read. |
| 601 | * Otherwise, simply zero the buffer. |
| 602 | */ |
| 603 | read_lock_irqsave(&ni->size_lock, flags); |
| 604 | initialized_size = ni->initialized_size; |
| 605 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 606 | if (bh_pos < initialized_size) { |
| 607 | ntfs_submit_bh_for_read(bh); |
| 608 | *wait_bh++ = bh; |
| 609 | } else { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 610 | zero_user_page(page, bh_offset(bh), |
| 611 | blocksize, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 612 | set_buffer_uptodate(bh); |
| 613 | } |
| 614 | } |
| 615 | continue; |
| 616 | } |
| 617 | /* Unmapped buffer. Need to map it. */ |
| 618 | bh->b_bdev = vol->sb->s_bdev; |
| 619 | /* |
| 620 | * If the current buffer is in the same clusters as the map |
| 621 | * cache, there is no need to check the runlist again. The |
| 622 | * map cache is made up of @vcn, which is the first cached file |
| 623 | * cluster, @vcn_len which is the number of cached file |
| 624 | * clusters, @lcn is the device cluster corresponding to @vcn, |
| 625 | * and @lcn_block is the block number corresponding to @lcn. |
| 626 | */ |
| 627 | cdelta = bh_cpos - vcn; |
| 628 | if (likely(!cdelta || (cdelta > 0 && cdelta < vcn_len))) { |
| 629 | map_buffer_cached: |
| 630 | BUG_ON(lcn < 0); |
| 631 | bh->b_blocknr = lcn_block + |
| 632 | (cdelta << (vol->cluster_size_bits - |
| 633 | blocksize_bits)) + |
| 634 | (bh_cofs >> blocksize_bits); |
| 635 | set_buffer_mapped(bh); |
| 636 | /* |
| 637 | * If the page is uptodate so is the buffer. If the |
| 638 | * buffer is fully outside the write, we ignore it if |
| 639 | * it was already allocated and we mark it dirty so it |
| 640 | * gets written out if we allocated it. On the other |
| 641 | * hand, if we allocated the buffer but we are not |
| 642 | * marking it dirty we set buffer_new so we can do |
| 643 | * error recovery. |
| 644 | */ |
| 645 | if (PageUptodate(page)) { |
| 646 | if (!buffer_uptodate(bh)) |
| 647 | set_buffer_uptodate(bh); |
| 648 | if (unlikely(was_hole)) { |
| 649 | /* We allocated the buffer. */ |
| 650 | unmap_underlying_metadata(bh->b_bdev, |
| 651 | bh->b_blocknr); |
| 652 | if (bh_end <= pos || bh_pos >= end) |
| 653 | mark_buffer_dirty(bh); |
| 654 | else |
| 655 | set_buffer_new(bh); |
| 656 | } |
| 657 | continue; |
| 658 | } |
| 659 | /* Page is _not_ uptodate. */ |
| 660 | if (likely(!was_hole)) { |
| 661 | /* |
| 662 | * Buffer was already allocated. If it is not |
| 663 | * uptodate and is only partially being written |
| 664 | * to, we need to read it in before the write, |
| 665 | * i.e. now. |
| 666 | */ |
Anton Altaparmakov | 3aebf25 | 2005-11-01 15:49:31 +0000 | [diff] [blame] | 667 | if (!buffer_uptodate(bh) && bh_pos < end && |
| 668 | bh_end > pos && |
| 669 | (bh_pos < pos || |
| 670 | bh_end > end)) { |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 671 | /* |
| 672 | * If the buffer is fully or partially |
| 673 | * within the initialized size, do an |
| 674 | * actual read. Otherwise, simply zero |
| 675 | * the buffer. |
| 676 | */ |
| 677 | read_lock_irqsave(&ni->size_lock, |
| 678 | flags); |
| 679 | initialized_size = ni->initialized_size; |
| 680 | read_unlock_irqrestore(&ni->size_lock, |
| 681 | flags); |
| 682 | if (bh_pos < initialized_size) { |
| 683 | ntfs_submit_bh_for_read(bh); |
| 684 | *wait_bh++ = bh; |
| 685 | } else { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 686 | zero_user_page(page, |
| 687 | bh_offset(bh), |
| 688 | blocksize, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 689 | set_buffer_uptodate(bh); |
| 690 | } |
| 691 | } |
| 692 | continue; |
| 693 | } |
| 694 | /* We allocated the buffer. */ |
| 695 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); |
| 696 | /* |
| 697 | * If the buffer is fully outside the write, zero it, |
| 698 | * set it uptodate, and mark it dirty so it gets |
| 699 | * written out. If it is partially being written to, |
| 700 | * zero region surrounding the write but leave it to |
| 701 | * commit write to do anything else. Finally, if the |
| 702 | * buffer is fully being overwritten, do nothing. |
| 703 | */ |
| 704 | if (bh_end <= pos || bh_pos >= end) { |
| 705 | if (!buffer_uptodate(bh)) { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 706 | zero_user_page(page, bh_offset(bh), |
| 707 | blocksize, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 708 | set_buffer_uptodate(bh); |
| 709 | } |
| 710 | mark_buffer_dirty(bh); |
| 711 | continue; |
| 712 | } |
| 713 | set_buffer_new(bh); |
| 714 | if (!buffer_uptodate(bh) && |
| 715 | (bh_pos < pos || bh_end > end)) { |
| 716 | u8 *kaddr; |
| 717 | unsigned pofs; |
| 718 | |
| 719 | kaddr = kmap_atomic(page, KM_USER0); |
| 720 | if (bh_pos < pos) { |
| 721 | pofs = bh_pos & ~PAGE_CACHE_MASK; |
| 722 | memset(kaddr + pofs, 0, pos - bh_pos); |
| 723 | } |
| 724 | if (bh_end > end) { |
| 725 | pofs = end & ~PAGE_CACHE_MASK; |
| 726 | memset(kaddr + pofs, 0, bh_end - end); |
| 727 | } |
| 728 | kunmap_atomic(kaddr, KM_USER0); |
| 729 | flush_dcache_page(page); |
| 730 | } |
| 731 | continue; |
| 732 | } |
| 733 | /* |
| 734 | * Slow path: this is the first buffer in the cluster. If it |
| 735 | * is outside allocated size and is not uptodate, zero it and |
| 736 | * set it uptodate. |
| 737 | */ |
| 738 | read_lock_irqsave(&ni->size_lock, flags); |
| 739 | initialized_size = ni->allocated_size; |
| 740 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 741 | if (bh_pos > initialized_size) { |
| 742 | if (PageUptodate(page)) { |
| 743 | if (!buffer_uptodate(bh)) |
| 744 | set_buffer_uptodate(bh); |
| 745 | } else if (!buffer_uptodate(bh)) { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 746 | zero_user_page(page, bh_offset(bh), blocksize, |
| 747 | KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 748 | set_buffer_uptodate(bh); |
| 749 | } |
| 750 | continue; |
| 751 | } |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 752 | is_retry = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 753 | if (!rl) { |
| 754 | down_read(&ni->runlist.lock); |
| 755 | retry_remap: |
| 756 | rl = ni->runlist.rl; |
| 757 | } |
| 758 | if (likely(rl != NULL)) { |
| 759 | /* Seek to element containing target cluster. */ |
| 760 | while (rl->length && rl[1].vcn <= bh_cpos) |
| 761 | rl++; |
| 762 | lcn = ntfs_rl_vcn_to_lcn(rl, bh_cpos); |
| 763 | if (likely(lcn >= 0)) { |
| 764 | /* |
| 765 | * Successful remap, setup the map cache and |
| 766 | * use that to deal with the buffer. |
| 767 | */ |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 768 | was_hole = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 769 | vcn = bh_cpos; |
| 770 | vcn_len = rl[1].vcn - vcn; |
| 771 | lcn_block = lcn << (vol->cluster_size_bits - |
| 772 | blocksize_bits); |
Anton Altaparmakov | d5aeaef | 2005-10-19 12:23:10 +0100 | [diff] [blame] | 773 | cdelta = 0; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 774 | /* |
Anton Altaparmakov | 3aebf25 | 2005-11-01 15:49:31 +0000 | [diff] [blame] | 775 | * If the number of remaining clusters touched |
| 776 | * by the write is smaller or equal to the |
| 777 | * number of cached clusters, unlock the |
| 778 | * runlist as the map cache will be used from |
| 779 | * now on. |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 780 | */ |
| 781 | if (likely(vcn + vcn_len >= cend)) { |
| 782 | if (rl_write_locked) { |
| 783 | up_write(&ni->runlist.lock); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 784 | rl_write_locked = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 785 | } else |
| 786 | up_read(&ni->runlist.lock); |
| 787 | rl = NULL; |
| 788 | } |
| 789 | goto map_buffer_cached; |
| 790 | } |
| 791 | } else |
| 792 | lcn = LCN_RL_NOT_MAPPED; |
| 793 | /* |
| 794 | * If it is not a hole and not out of bounds, the runlist is |
| 795 | * probably unmapped so try to map it now. |
| 796 | */ |
| 797 | if (unlikely(lcn != LCN_HOLE && lcn != LCN_ENOENT)) { |
| 798 | if (likely(!is_retry && lcn == LCN_RL_NOT_MAPPED)) { |
| 799 | /* Attempt to map runlist. */ |
| 800 | if (!rl_write_locked) { |
| 801 | /* |
| 802 | * We need the runlist locked for |
| 803 | * writing, so if it is locked for |
| 804 | * reading relock it now and retry in |
| 805 | * case it changed whilst we dropped |
| 806 | * the lock. |
| 807 | */ |
| 808 | up_read(&ni->runlist.lock); |
| 809 | down_write(&ni->runlist.lock); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 810 | rl_write_locked = true; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 811 | goto retry_remap; |
| 812 | } |
| 813 | err = ntfs_map_runlist_nolock(ni, bh_cpos, |
| 814 | NULL); |
| 815 | if (likely(!err)) { |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 816 | is_retry = true; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 817 | goto retry_remap; |
| 818 | } |
| 819 | /* |
| 820 | * If @vcn is out of bounds, pretend @lcn is |
| 821 | * LCN_ENOENT. As long as the buffer is out |
| 822 | * of bounds this will work fine. |
| 823 | */ |
| 824 | if (err == -ENOENT) { |
| 825 | lcn = LCN_ENOENT; |
| 826 | err = 0; |
| 827 | goto rl_not_mapped_enoent; |
| 828 | } |
| 829 | } else |
| 830 | err = -EIO; |
| 831 | /* Failed to map the buffer, even after retrying. */ |
| 832 | bh->b_blocknr = -1; |
| 833 | ntfs_error(vol->sb, "Failed to write to inode 0x%lx, " |
| 834 | "attribute type 0x%x, vcn 0x%llx, " |
| 835 | "vcn offset 0x%x, because its " |
| 836 | "location on disk could not be " |
| 837 | "determined%s (error code %i).", |
| 838 | ni->mft_no, ni->type, |
| 839 | (unsigned long long)bh_cpos, |
| 840 | (unsigned)bh_pos & |
| 841 | vol->cluster_size_mask, |
| 842 | is_retry ? " even after retrying" : "", |
| 843 | err); |
| 844 | break; |
| 845 | } |
| 846 | rl_not_mapped_enoent: |
| 847 | /* |
| 848 | * The buffer is in a hole or out of bounds. We need to fill |
| 849 | * the hole, unless the buffer is in a cluster which is not |
| 850 | * touched by the write, in which case we just leave the buffer |
| 851 | * unmapped. This can only happen when the cluster size is |
| 852 | * less than the page cache size. |
| 853 | */ |
| 854 | if (unlikely(vol->cluster_size < PAGE_CACHE_SIZE)) { |
| 855 | bh_cend = (bh_end + vol->cluster_size - 1) >> |
| 856 | vol->cluster_size_bits; |
| 857 | if ((bh_cend <= cpos || bh_cpos >= cend)) { |
| 858 | bh->b_blocknr = -1; |
| 859 | /* |
| 860 | * If the buffer is uptodate we skip it. If it |
| 861 | * is not but the page is uptodate, we can set |
| 862 | * the buffer uptodate. If the page is not |
| 863 | * uptodate, we can clear the buffer and set it |
| 864 | * uptodate. Whether this is worthwhile is |
| 865 | * debatable and this could be removed. |
| 866 | */ |
| 867 | if (PageUptodate(page)) { |
| 868 | if (!buffer_uptodate(bh)) |
| 869 | set_buffer_uptodate(bh); |
| 870 | } else if (!buffer_uptodate(bh)) { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 871 | zero_user_page(page, bh_offset(bh), |
| 872 | blocksize, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 873 | set_buffer_uptodate(bh); |
| 874 | } |
| 875 | continue; |
| 876 | } |
| 877 | } |
| 878 | /* |
| 879 | * Out of bounds buffer is invalid if it was not really out of |
| 880 | * bounds. |
| 881 | */ |
| 882 | BUG_ON(lcn != LCN_HOLE); |
| 883 | /* |
| 884 | * We need the runlist locked for writing, so if it is locked |
| 885 | * for reading relock it now and retry in case it changed |
| 886 | * whilst we dropped the lock. |
| 887 | */ |
| 888 | BUG_ON(!rl); |
| 889 | if (!rl_write_locked) { |
| 890 | up_read(&ni->runlist.lock); |
| 891 | down_write(&ni->runlist.lock); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 892 | rl_write_locked = true; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 893 | goto retry_remap; |
| 894 | } |
| 895 | /* Find the previous last allocated cluster. */ |
| 896 | BUG_ON(rl->lcn != LCN_HOLE); |
| 897 | lcn = -1; |
| 898 | rl2 = rl; |
| 899 | while (--rl2 >= ni->runlist.rl) { |
| 900 | if (rl2->lcn >= 0) { |
| 901 | lcn = rl2->lcn + rl2->length; |
| 902 | break; |
| 903 | } |
| 904 | } |
| 905 | rl2 = ntfs_cluster_alloc(vol, bh_cpos, 1, lcn, DATA_ZONE, |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 906 | false); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 907 | if (IS_ERR(rl2)) { |
| 908 | err = PTR_ERR(rl2); |
| 909 | ntfs_debug("Failed to allocate cluster, error code %i.", |
| 910 | err); |
| 911 | break; |
| 912 | } |
| 913 | lcn = rl2->lcn; |
| 914 | rl = ntfs_runlists_merge(ni->runlist.rl, rl2); |
| 915 | if (IS_ERR(rl)) { |
| 916 | err = PTR_ERR(rl); |
| 917 | if (err != -ENOMEM) |
| 918 | err = -EIO; |
| 919 | if (ntfs_cluster_free_from_rl(vol, rl2)) { |
| 920 | ntfs_error(vol->sb, "Failed to release " |
| 921 | "allocated cluster in error " |
| 922 | "code path. Run chkdsk to " |
| 923 | "recover the lost cluster."); |
| 924 | NVolSetErrors(vol); |
| 925 | } |
| 926 | ntfs_free(rl2); |
| 927 | break; |
| 928 | } |
| 929 | ni->runlist.rl = rl; |
| 930 | status.runlist_merged = 1; |
Anton Altaparmakov | bb8047d | 2006-03-07 11:53:46 +0000 | [diff] [blame] | 931 | ntfs_debug("Allocated cluster, lcn 0x%llx.", |
| 932 | (unsigned long long)lcn); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 933 | /* Map and lock the mft record and get the attribute record. */ |
| 934 | if (!NInoAttr(ni)) |
| 935 | base_ni = ni; |
| 936 | else |
| 937 | base_ni = ni->ext.base_ntfs_ino; |
| 938 | m = map_mft_record(base_ni); |
| 939 | if (IS_ERR(m)) { |
| 940 | err = PTR_ERR(m); |
| 941 | break; |
| 942 | } |
| 943 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 944 | if (unlikely(!ctx)) { |
| 945 | err = -ENOMEM; |
| 946 | unmap_mft_record(base_ni); |
| 947 | break; |
| 948 | } |
| 949 | status.mft_attr_mapped = 1; |
| 950 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 951 | CASE_SENSITIVE, bh_cpos, NULL, 0, ctx); |
| 952 | if (unlikely(err)) { |
| 953 | if (err == -ENOENT) |
| 954 | err = -EIO; |
| 955 | break; |
| 956 | } |
| 957 | m = ctx->mrec; |
| 958 | a = ctx->attr; |
| 959 | /* |
| 960 | * Find the runlist element with which the attribute extent |
| 961 | * starts. Note, we cannot use the _attr_ version because we |
| 962 | * have mapped the mft record. That is ok because we know the |
| 963 | * runlist fragment must be mapped already to have ever gotten |
| 964 | * here, so we can just use the _rl_ version. |
| 965 | */ |
| 966 | vcn = sle64_to_cpu(a->data.non_resident.lowest_vcn); |
| 967 | rl2 = ntfs_rl_find_vcn_nolock(rl, vcn); |
| 968 | BUG_ON(!rl2); |
| 969 | BUG_ON(!rl2->length); |
| 970 | BUG_ON(rl2->lcn < LCN_HOLE); |
| 971 | highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn); |
| 972 | /* |
| 973 | * If @highest_vcn is zero, calculate the real highest_vcn |
| 974 | * (which can really be zero). |
| 975 | */ |
| 976 | if (!highest_vcn) |
| 977 | highest_vcn = (sle64_to_cpu( |
| 978 | a->data.non_resident.allocated_size) >> |
| 979 | vol->cluster_size_bits) - 1; |
| 980 | /* |
| 981 | * Determine the size of the mapping pairs array for the new |
| 982 | * extent, i.e. the old extent with the hole filled. |
| 983 | */ |
| 984 | mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, vcn, |
| 985 | highest_vcn); |
| 986 | if (unlikely(mp_size <= 0)) { |
| 987 | if (!(err = mp_size)) |
| 988 | err = -EIO; |
| 989 | ntfs_debug("Failed to get size for mapping pairs " |
| 990 | "array, error code %i.", err); |
| 991 | break; |
| 992 | } |
| 993 | /* |
| 994 | * Resize the attribute record to fit the new mapping pairs |
| 995 | * array. |
| 996 | */ |
| 997 | attr_rec_len = le32_to_cpu(a->length); |
| 998 | err = ntfs_attr_record_resize(m, a, mp_size + le16_to_cpu( |
| 999 | a->data.non_resident.mapping_pairs_offset)); |
| 1000 | if (unlikely(err)) { |
| 1001 | BUG_ON(err != -ENOSPC); |
| 1002 | // TODO: Deal with this by using the current attribute |
| 1003 | // and fill it with as much of the mapping pairs |
| 1004 | // array as possible. Then loop over each attribute |
| 1005 | // extent rewriting the mapping pairs arrays as we go |
| 1006 | // along and if when we reach the end we have not |
| 1007 | // enough space, try to resize the last attribute |
| 1008 | // extent and if even that fails, add a new attribute |
| 1009 | // extent. |
| 1010 | // We could also try to resize at each step in the hope |
| 1011 | // that we will not need to rewrite every single extent. |
| 1012 | // Note, we may need to decompress some extents to fill |
| 1013 | // the runlist as we are walking the extents... |
| 1014 | ntfs_error(vol->sb, "Not enough space in the mft " |
| 1015 | "record for the extended attribute " |
| 1016 | "record. This case is not " |
| 1017 | "implemented yet."); |
| 1018 | err = -EOPNOTSUPP; |
| 1019 | break ; |
| 1020 | } |
| 1021 | status.mp_rebuilt = 1; |
| 1022 | /* |
| 1023 | * Generate the mapping pairs array directly into the attribute |
| 1024 | * record. |
| 1025 | */ |
| 1026 | err = ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( |
| 1027 | a->data.non_resident.mapping_pairs_offset), |
| 1028 | mp_size, rl2, vcn, highest_vcn, NULL); |
| 1029 | if (unlikely(err)) { |
| 1030 | ntfs_error(vol->sb, "Cannot fill hole in inode 0x%lx, " |
| 1031 | "attribute type 0x%x, because building " |
| 1032 | "the mapping pairs failed with error " |
| 1033 | "code %i.", vi->i_ino, |
| 1034 | (unsigned)le32_to_cpu(ni->type), err); |
| 1035 | err = -EIO; |
| 1036 | break; |
| 1037 | } |
| 1038 | /* Update the highest_vcn but only if it was not set. */ |
| 1039 | if (unlikely(!a->data.non_resident.highest_vcn)) |
| 1040 | a->data.non_resident.highest_vcn = |
| 1041 | cpu_to_sle64(highest_vcn); |
| 1042 | /* |
| 1043 | * If the attribute is sparse/compressed, update the compressed |
| 1044 | * size in the ntfs_inode structure and the attribute record. |
| 1045 | */ |
| 1046 | if (likely(NInoSparse(ni) || NInoCompressed(ni))) { |
| 1047 | /* |
| 1048 | * If we are not in the first attribute extent, switch |
| 1049 | * to it, but first ensure the changes will make it to |
| 1050 | * disk later. |
| 1051 | */ |
| 1052 | if (a->data.non_resident.lowest_vcn) { |
| 1053 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1054 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1055 | ntfs_attr_reinit_search_ctx(ctx); |
| 1056 | err = ntfs_attr_lookup(ni->type, ni->name, |
| 1057 | ni->name_len, CASE_SENSITIVE, |
| 1058 | 0, NULL, 0, ctx); |
| 1059 | if (unlikely(err)) { |
| 1060 | status.attr_switched = 1; |
| 1061 | break; |
| 1062 | } |
| 1063 | /* @m is not used any more so do not set it. */ |
| 1064 | a = ctx->attr; |
| 1065 | } |
| 1066 | write_lock_irqsave(&ni->size_lock, flags); |
| 1067 | ni->itype.compressed.size += vol->cluster_size; |
| 1068 | a->data.non_resident.compressed_size = |
| 1069 | cpu_to_sle64(ni->itype.compressed.size); |
| 1070 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 1071 | } |
| 1072 | /* Ensure the changes make it to disk. */ |
| 1073 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1074 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1075 | ntfs_attr_put_search_ctx(ctx); |
| 1076 | unmap_mft_record(base_ni); |
| 1077 | /* Successfully filled the hole. */ |
| 1078 | status.runlist_merged = 0; |
| 1079 | status.mft_attr_mapped = 0; |
| 1080 | status.mp_rebuilt = 0; |
| 1081 | /* Setup the map cache and use that to deal with the buffer. */ |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1082 | was_hole = true; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1083 | vcn = bh_cpos; |
| 1084 | vcn_len = 1; |
| 1085 | lcn_block = lcn << (vol->cluster_size_bits - blocksize_bits); |
| 1086 | cdelta = 0; |
| 1087 | /* |
| 1088 | * If the number of remaining clusters in the @pages is smaller |
| 1089 | * or equal to the number of cached clusters, unlock the |
| 1090 | * runlist as the map cache will be used from now on. |
| 1091 | */ |
| 1092 | if (likely(vcn + vcn_len >= cend)) { |
| 1093 | up_write(&ni->runlist.lock); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1094 | rl_write_locked = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1095 | rl = NULL; |
| 1096 | } |
| 1097 | goto map_buffer_cached; |
| 1098 | } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); |
| 1099 | /* If there are no errors, do the next page. */ |
| 1100 | if (likely(!err && ++u < nr_pages)) |
| 1101 | goto do_next_page; |
| 1102 | /* If there are no errors, release the runlist lock if we took it. */ |
| 1103 | if (likely(!err)) { |
| 1104 | if (unlikely(rl_write_locked)) { |
| 1105 | up_write(&ni->runlist.lock); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1106 | rl_write_locked = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1107 | } else if (unlikely(rl)) |
| 1108 | up_read(&ni->runlist.lock); |
| 1109 | rl = NULL; |
| 1110 | } |
| 1111 | /* If we issued read requests, let them complete. */ |
| 1112 | read_lock_irqsave(&ni->size_lock, flags); |
| 1113 | initialized_size = ni->initialized_size; |
| 1114 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1115 | while (wait_bh > wait) { |
| 1116 | bh = *--wait_bh; |
| 1117 | wait_on_buffer(bh); |
| 1118 | if (likely(buffer_uptodate(bh))) { |
| 1119 | page = bh->b_page; |
| 1120 | bh_pos = ((s64)page->index << PAGE_CACHE_SHIFT) + |
| 1121 | bh_offset(bh); |
| 1122 | /* |
| 1123 | * If the buffer overflows the initialized size, need |
| 1124 | * to zero the overflowing region. |
| 1125 | */ |
| 1126 | if (unlikely(bh_pos + blocksize > initialized_size)) { |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1127 | int ofs = 0; |
| 1128 | |
| 1129 | if (likely(bh_pos < initialized_size)) |
| 1130 | ofs = initialized_size - bh_pos; |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 1131 | zero_user_page(page, bh_offset(bh) + ofs, |
| 1132 | blocksize - ofs, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1133 | } |
| 1134 | } else /* if (unlikely(!buffer_uptodate(bh))) */ |
| 1135 | err = -EIO; |
| 1136 | } |
| 1137 | if (likely(!err)) { |
| 1138 | /* Clear buffer_new on all buffers. */ |
| 1139 | u = 0; |
| 1140 | do { |
| 1141 | bh = head = page_buffers(pages[u]); |
| 1142 | do { |
| 1143 | if (buffer_new(bh)) |
| 1144 | clear_buffer_new(bh); |
| 1145 | } while ((bh = bh->b_this_page) != head); |
| 1146 | } while (++u < nr_pages); |
| 1147 | ntfs_debug("Done."); |
| 1148 | return err; |
| 1149 | } |
| 1150 | if (status.attr_switched) { |
| 1151 | /* Get back to the attribute extent we modified. */ |
| 1152 | ntfs_attr_reinit_search_ctx(ctx); |
| 1153 | if (ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 1154 | CASE_SENSITIVE, bh_cpos, NULL, 0, ctx)) { |
| 1155 | ntfs_error(vol->sb, "Failed to find required " |
| 1156 | "attribute extent of attribute in " |
| 1157 | "error code path. Run chkdsk to " |
| 1158 | "recover."); |
| 1159 | write_lock_irqsave(&ni->size_lock, flags); |
| 1160 | ni->itype.compressed.size += vol->cluster_size; |
| 1161 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 1162 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1163 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1164 | /* |
| 1165 | * The only thing that is now wrong is the compressed |
| 1166 | * size of the base attribute extent which chkdsk |
| 1167 | * should be able to fix. |
| 1168 | */ |
| 1169 | NVolSetErrors(vol); |
| 1170 | } else { |
| 1171 | m = ctx->mrec; |
| 1172 | a = ctx->attr; |
| 1173 | status.attr_switched = 0; |
| 1174 | } |
| 1175 | } |
| 1176 | /* |
| 1177 | * If the runlist has been modified, need to restore it by punching a |
| 1178 | * hole into it and we then need to deallocate the on-disk cluster as |
| 1179 | * well. Note, we only modify the runlist if we are able to generate a |
| 1180 | * new mapping pairs array, i.e. only when the mapped attribute extent |
| 1181 | * is not switched. |
| 1182 | */ |
| 1183 | if (status.runlist_merged && !status.attr_switched) { |
| 1184 | BUG_ON(!rl_write_locked); |
| 1185 | /* Make the file cluster we allocated sparse in the runlist. */ |
| 1186 | if (ntfs_rl_punch_nolock(vol, &ni->runlist, bh_cpos, 1)) { |
| 1187 | ntfs_error(vol->sb, "Failed to punch hole into " |
| 1188 | "attribute runlist in error code " |
| 1189 | "path. Run chkdsk to recover the " |
| 1190 | "lost cluster."); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1191 | NVolSetErrors(vol); |
| 1192 | } else /* if (success) */ { |
| 1193 | status.runlist_merged = 0; |
| 1194 | /* |
| 1195 | * Deallocate the on-disk cluster we allocated but only |
| 1196 | * if we succeeded in punching its vcn out of the |
| 1197 | * runlist. |
| 1198 | */ |
| 1199 | down_write(&vol->lcnbmp_lock); |
| 1200 | if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { |
| 1201 | ntfs_error(vol->sb, "Failed to release " |
| 1202 | "allocated cluster in error " |
| 1203 | "code path. Run chkdsk to " |
| 1204 | "recover the lost cluster."); |
| 1205 | NVolSetErrors(vol); |
| 1206 | } |
| 1207 | up_write(&vol->lcnbmp_lock); |
| 1208 | } |
| 1209 | } |
| 1210 | /* |
| 1211 | * Resize the attribute record to its old size and rebuild the mapping |
| 1212 | * pairs array. Note, we only can do this if the runlist has been |
| 1213 | * restored to its old state which also implies that the mapped |
| 1214 | * attribute extent is not switched. |
| 1215 | */ |
| 1216 | if (status.mp_rebuilt && !status.runlist_merged) { |
| 1217 | if (ntfs_attr_record_resize(m, a, attr_rec_len)) { |
| 1218 | ntfs_error(vol->sb, "Failed to restore attribute " |
| 1219 | "record in error code path. Run " |
| 1220 | "chkdsk to recover."); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1221 | NVolSetErrors(vol); |
| 1222 | } else /* if (success) */ { |
| 1223 | if (ntfs_mapping_pairs_build(vol, (u8*)a + |
| 1224 | le16_to_cpu(a->data.non_resident. |
| 1225 | mapping_pairs_offset), attr_rec_len - |
| 1226 | le16_to_cpu(a->data.non_resident. |
| 1227 | mapping_pairs_offset), ni->runlist.rl, |
| 1228 | vcn, highest_vcn, NULL)) { |
| 1229 | ntfs_error(vol->sb, "Failed to restore " |
| 1230 | "mapping pairs array in error " |
| 1231 | "code path. Run chkdsk to " |
| 1232 | "recover."); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1233 | NVolSetErrors(vol); |
| 1234 | } |
| 1235 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1236 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1237 | } |
| 1238 | } |
| 1239 | /* Release the mft record and the attribute. */ |
| 1240 | if (status.mft_attr_mapped) { |
| 1241 | ntfs_attr_put_search_ctx(ctx); |
| 1242 | unmap_mft_record(base_ni); |
| 1243 | } |
| 1244 | /* Release the runlist lock. */ |
| 1245 | if (rl_write_locked) |
| 1246 | up_write(&ni->runlist.lock); |
| 1247 | else if (rl) |
| 1248 | up_read(&ni->runlist.lock); |
| 1249 | /* |
| 1250 | * Zero out any newly allocated blocks to avoid exposing stale data. |
| 1251 | * If BH_New is set, we know that the block was newly allocated above |
| 1252 | * and that it has not been fully zeroed and marked dirty yet. |
| 1253 | */ |
| 1254 | nr_pages = u; |
| 1255 | u = 0; |
| 1256 | end = bh_cpos << vol->cluster_size_bits; |
| 1257 | do { |
| 1258 | page = pages[u]; |
| 1259 | bh = head = page_buffers(page); |
| 1260 | do { |
| 1261 | if (u == nr_pages && |
| 1262 | ((s64)page->index << PAGE_CACHE_SHIFT) + |
| 1263 | bh_offset(bh) >= end) |
| 1264 | break; |
| 1265 | if (!buffer_new(bh)) |
| 1266 | continue; |
| 1267 | clear_buffer_new(bh); |
| 1268 | if (!buffer_uptodate(bh)) { |
| 1269 | if (PageUptodate(page)) |
| 1270 | set_buffer_uptodate(bh); |
| 1271 | else { |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 1272 | zero_user_page(page, bh_offset(bh), |
| 1273 | blocksize, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1274 | set_buffer_uptodate(bh); |
| 1275 | } |
| 1276 | } |
| 1277 | mark_buffer_dirty(bh); |
| 1278 | } while ((bh = bh->b_this_page) != head); |
| 1279 | } while (++u <= nr_pages); |
| 1280 | ntfs_error(vol->sb, "Failed. Returning error code %i.", err); |
| 1281 | return err; |
| 1282 | } |
| 1283 | |
| 1284 | /* |
| 1285 | * Copy as much as we can into the pages and return the number of bytes which |
| 1286 | * were sucessfully copied. If a fault is encountered then clear the pages |
| 1287 | * out to (ofs + bytes) and return the number of bytes which were copied. |
| 1288 | */ |
| 1289 | static inline size_t ntfs_copy_from_user(struct page **pages, |
| 1290 | unsigned nr_pages, unsigned ofs, const char __user *buf, |
| 1291 | size_t bytes) |
| 1292 | { |
| 1293 | struct page **last_page = pages + nr_pages; |
| 1294 | char *kaddr; |
| 1295 | size_t total = 0; |
| 1296 | unsigned len; |
| 1297 | int left; |
| 1298 | |
| 1299 | do { |
| 1300 | len = PAGE_CACHE_SIZE - ofs; |
| 1301 | if (len > bytes) |
| 1302 | len = bytes; |
| 1303 | kaddr = kmap_atomic(*pages, KM_USER0); |
| 1304 | left = __copy_from_user_inatomic(kaddr + ofs, buf, len); |
| 1305 | kunmap_atomic(kaddr, KM_USER0); |
| 1306 | if (unlikely(left)) { |
| 1307 | /* Do it the slow way. */ |
| 1308 | kaddr = kmap(*pages); |
| 1309 | left = __copy_from_user(kaddr + ofs, buf, len); |
| 1310 | kunmap(*pages); |
| 1311 | if (unlikely(left)) |
| 1312 | goto err_out; |
| 1313 | } |
| 1314 | total += len; |
| 1315 | bytes -= len; |
| 1316 | if (!bytes) |
| 1317 | break; |
| 1318 | buf += len; |
| 1319 | ofs = 0; |
| 1320 | } while (++pages < last_page); |
| 1321 | out: |
| 1322 | return total; |
| 1323 | err_out: |
| 1324 | total += len - left; |
| 1325 | /* Zero the rest of the target like __copy_from_user(). */ |
| 1326 | while (++pages < last_page) { |
| 1327 | bytes -= len; |
| 1328 | if (!bytes) |
| 1329 | break; |
| 1330 | len = PAGE_CACHE_SIZE; |
| 1331 | if (len > bytes) |
| 1332 | len = bytes; |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 1333 | zero_user_page(*pages, 0, len, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1334 | } |
| 1335 | goto out; |
| 1336 | } |
| 1337 | |
NeilBrown | 01408c4 | 2006-06-25 05:47:58 -0700 | [diff] [blame] | 1338 | static size_t __ntfs_copy_from_user_iovec_inatomic(char *vaddr, |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1339 | const struct iovec *iov, size_t iov_ofs, size_t bytes) |
| 1340 | { |
| 1341 | size_t total = 0; |
| 1342 | |
| 1343 | while (1) { |
| 1344 | const char __user *buf = iov->iov_base + iov_ofs; |
| 1345 | unsigned len; |
| 1346 | size_t left; |
| 1347 | |
| 1348 | len = iov->iov_len - iov_ofs; |
| 1349 | if (len > bytes) |
| 1350 | len = bytes; |
| 1351 | left = __copy_from_user_inatomic(vaddr, buf, len); |
| 1352 | total += len; |
| 1353 | bytes -= len; |
| 1354 | vaddr += len; |
| 1355 | if (unlikely(left)) { |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1356 | total -= left; |
| 1357 | break; |
| 1358 | } |
| 1359 | if (!bytes) |
| 1360 | break; |
| 1361 | iov++; |
| 1362 | iov_ofs = 0; |
| 1363 | } |
| 1364 | return total; |
| 1365 | } |
| 1366 | |
| 1367 | static inline void ntfs_set_next_iovec(const struct iovec **iovp, |
| 1368 | size_t *iov_ofsp, size_t bytes) |
| 1369 | { |
| 1370 | const struct iovec *iov = *iovp; |
| 1371 | size_t iov_ofs = *iov_ofsp; |
| 1372 | |
| 1373 | while (bytes) { |
| 1374 | unsigned len; |
| 1375 | |
| 1376 | len = iov->iov_len - iov_ofs; |
| 1377 | if (len > bytes) |
| 1378 | len = bytes; |
| 1379 | bytes -= len; |
| 1380 | iov_ofs += len; |
| 1381 | if (iov->iov_len == iov_ofs) { |
| 1382 | iov++; |
| 1383 | iov_ofs = 0; |
| 1384 | } |
| 1385 | } |
| 1386 | *iovp = iov; |
| 1387 | *iov_ofsp = iov_ofs; |
| 1388 | } |
| 1389 | |
| 1390 | /* |
| 1391 | * This has the same side-effects and return value as ntfs_copy_from_user(). |
| 1392 | * The difference is that on a fault we need to memset the remainder of the |
| 1393 | * pages (out to offset + bytes), to emulate ntfs_copy_from_user()'s |
| 1394 | * single-segment behaviour. |
| 1395 | * |
NeilBrown | 01408c4 | 2006-06-25 05:47:58 -0700 | [diff] [blame] | 1396 | * We call the same helper (__ntfs_copy_from_user_iovec_inatomic()) both |
| 1397 | * when atomic and when not atomic. This is ok because |
| 1398 | * __ntfs_copy_from_user_iovec_inatomic() calls __copy_from_user_inatomic() |
| 1399 | * and it is ok to call this when non-atomic. |
| 1400 | * Infact, the only difference between __copy_from_user_inatomic() and |
| 1401 | * __copy_from_user() is that the latter calls might_sleep() and the former |
| 1402 | * should not zero the tail of the buffer on error. And on many |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1403 | * architectures __copy_from_user_inatomic() is just defined to |
| 1404 | * __copy_from_user() so it makes no difference at all on those architectures. |
| 1405 | */ |
| 1406 | static inline size_t ntfs_copy_from_user_iovec(struct page **pages, |
| 1407 | unsigned nr_pages, unsigned ofs, const struct iovec **iov, |
| 1408 | size_t *iov_ofs, size_t bytes) |
| 1409 | { |
| 1410 | struct page **last_page = pages + nr_pages; |
| 1411 | char *kaddr; |
| 1412 | size_t copied, len, total = 0; |
| 1413 | |
| 1414 | do { |
| 1415 | len = PAGE_CACHE_SIZE - ofs; |
| 1416 | if (len > bytes) |
| 1417 | len = bytes; |
| 1418 | kaddr = kmap_atomic(*pages, KM_USER0); |
NeilBrown | 01408c4 | 2006-06-25 05:47:58 -0700 | [diff] [blame] | 1419 | copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs, |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1420 | *iov, *iov_ofs, len); |
| 1421 | kunmap_atomic(kaddr, KM_USER0); |
| 1422 | if (unlikely(copied != len)) { |
| 1423 | /* Do it the slow way. */ |
| 1424 | kaddr = kmap(*pages); |
NeilBrown | 01408c4 | 2006-06-25 05:47:58 -0700 | [diff] [blame] | 1425 | copied = __ntfs_copy_from_user_iovec_inatomic(kaddr + ofs, |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1426 | *iov, *iov_ofs, len); |
NeilBrown | 01408c4 | 2006-06-25 05:47:58 -0700 | [diff] [blame] | 1427 | /* |
| 1428 | * Zero the rest of the target like __copy_from_user(). |
| 1429 | */ |
| 1430 | memset(kaddr + ofs + copied, 0, len - copied); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1431 | kunmap(*pages); |
| 1432 | if (unlikely(copied != len)) |
| 1433 | goto err_out; |
| 1434 | } |
| 1435 | total += len; |
| 1436 | bytes -= len; |
| 1437 | if (!bytes) |
| 1438 | break; |
| 1439 | ntfs_set_next_iovec(iov, iov_ofs, len); |
| 1440 | ofs = 0; |
| 1441 | } while (++pages < last_page); |
| 1442 | out: |
| 1443 | return total; |
| 1444 | err_out: |
| 1445 | total += copied; |
| 1446 | /* Zero the rest of the target like __copy_from_user(). */ |
| 1447 | while (++pages < last_page) { |
| 1448 | bytes -= len; |
| 1449 | if (!bytes) |
| 1450 | break; |
| 1451 | len = PAGE_CACHE_SIZE; |
| 1452 | if (len > bytes) |
| 1453 | len = bytes; |
Nate Diller | e3bf460 | 2007-05-12 10:36:54 -0700 | [diff] [blame] | 1454 | zero_user_page(*pages, 0, len, KM_USER0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1455 | } |
| 1456 | goto out; |
| 1457 | } |
| 1458 | |
| 1459 | static inline void ntfs_flush_dcache_pages(struct page **pages, |
| 1460 | unsigned nr_pages) |
| 1461 | { |
| 1462 | BUG_ON(!nr_pages); |
Anton Altaparmakov | f893afb | 2006-06-22 14:47:15 -0700 | [diff] [blame] | 1463 | /* |
| 1464 | * Warning: Do not do the decrement at the same time as the call to |
| 1465 | * flush_dcache_page() because it is a NULL macro on i386 and hence the |
| 1466 | * decrement never happens so the loop never terminates. |
| 1467 | */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1468 | do { |
Anton Altaparmakov | f893afb | 2006-06-22 14:47:15 -0700 | [diff] [blame] | 1469 | --nr_pages; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1470 | flush_dcache_page(pages[nr_pages]); |
Anton Altaparmakov | f893afb | 2006-06-22 14:47:15 -0700 | [diff] [blame] | 1471 | } while (nr_pages > 0); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1472 | } |
| 1473 | |
| 1474 | /** |
| 1475 | * ntfs_commit_pages_after_non_resident_write - commit the received data |
| 1476 | * @pages: array of destination pages |
| 1477 | * @nr_pages: number of pages in @pages |
| 1478 | * @pos: byte position in file at which the write begins |
| 1479 | * @bytes: number of bytes to be written |
| 1480 | * |
| 1481 | * See description of ntfs_commit_pages_after_write(), below. |
| 1482 | */ |
| 1483 | static inline int ntfs_commit_pages_after_non_resident_write( |
| 1484 | struct page **pages, const unsigned nr_pages, |
| 1485 | s64 pos, size_t bytes) |
| 1486 | { |
| 1487 | s64 end, initialized_size; |
| 1488 | struct inode *vi; |
| 1489 | ntfs_inode *ni, *base_ni; |
| 1490 | struct buffer_head *bh, *head; |
| 1491 | ntfs_attr_search_ctx *ctx; |
| 1492 | MFT_RECORD *m; |
| 1493 | ATTR_RECORD *a; |
| 1494 | unsigned long flags; |
| 1495 | unsigned blocksize, u; |
| 1496 | int err; |
| 1497 | |
| 1498 | vi = pages[0]->mapping->host; |
| 1499 | ni = NTFS_I(vi); |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 1500 | blocksize = vi->i_sb->s_blocksize; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1501 | end = pos + bytes; |
| 1502 | u = 0; |
| 1503 | do { |
| 1504 | s64 bh_pos; |
| 1505 | struct page *page; |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1506 | bool partial; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1507 | |
| 1508 | page = pages[u]; |
| 1509 | bh_pos = (s64)page->index << PAGE_CACHE_SHIFT; |
| 1510 | bh = head = page_buffers(page); |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1511 | partial = false; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1512 | do { |
| 1513 | s64 bh_end; |
| 1514 | |
| 1515 | bh_end = bh_pos + blocksize; |
| 1516 | if (bh_end <= pos || bh_pos >= end) { |
| 1517 | if (!buffer_uptodate(bh)) |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1518 | partial = true; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1519 | } else { |
| 1520 | set_buffer_uptodate(bh); |
| 1521 | mark_buffer_dirty(bh); |
| 1522 | } |
| 1523 | } while (bh_pos += blocksize, (bh = bh->b_this_page) != head); |
| 1524 | /* |
| 1525 | * If all buffers are now uptodate but the page is not, set the |
| 1526 | * page uptodate. |
| 1527 | */ |
| 1528 | if (!partial && !PageUptodate(page)) |
| 1529 | SetPageUptodate(page); |
| 1530 | } while (++u < nr_pages); |
| 1531 | /* |
| 1532 | * Finally, if we do not need to update initialized_size or i_size we |
| 1533 | * are finished. |
| 1534 | */ |
| 1535 | read_lock_irqsave(&ni->size_lock, flags); |
| 1536 | initialized_size = ni->initialized_size; |
| 1537 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1538 | if (end <= initialized_size) { |
| 1539 | ntfs_debug("Done."); |
| 1540 | return 0; |
| 1541 | } |
| 1542 | /* |
| 1543 | * Update initialized_size/i_size as appropriate, both in the inode and |
| 1544 | * the mft record. |
| 1545 | */ |
| 1546 | if (!NInoAttr(ni)) |
| 1547 | base_ni = ni; |
| 1548 | else |
| 1549 | base_ni = ni->ext.base_ntfs_ino; |
| 1550 | /* Map, pin, and lock the mft record. */ |
| 1551 | m = map_mft_record(base_ni); |
| 1552 | if (IS_ERR(m)) { |
| 1553 | err = PTR_ERR(m); |
| 1554 | m = NULL; |
| 1555 | ctx = NULL; |
| 1556 | goto err_out; |
| 1557 | } |
| 1558 | BUG_ON(!NInoNonResident(ni)); |
| 1559 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 1560 | if (unlikely(!ctx)) { |
| 1561 | err = -ENOMEM; |
| 1562 | goto err_out; |
| 1563 | } |
| 1564 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 1565 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 1566 | if (unlikely(err)) { |
| 1567 | if (err == -ENOENT) |
| 1568 | err = -EIO; |
| 1569 | goto err_out; |
| 1570 | } |
| 1571 | a = ctx->attr; |
| 1572 | BUG_ON(!a->non_resident); |
| 1573 | write_lock_irqsave(&ni->size_lock, flags); |
| 1574 | BUG_ON(end > ni->allocated_size); |
| 1575 | ni->initialized_size = end; |
| 1576 | a->data.non_resident.initialized_size = cpu_to_sle64(end); |
| 1577 | if (end > i_size_read(vi)) { |
| 1578 | i_size_write(vi, end); |
| 1579 | a->data.non_resident.data_size = |
| 1580 | a->data.non_resident.initialized_size; |
| 1581 | } |
| 1582 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 1583 | /* Mark the mft record dirty, so it gets written back. */ |
| 1584 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1585 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1586 | ntfs_attr_put_search_ctx(ctx); |
| 1587 | unmap_mft_record(base_ni); |
| 1588 | ntfs_debug("Done."); |
| 1589 | return 0; |
| 1590 | err_out: |
| 1591 | if (ctx) |
| 1592 | ntfs_attr_put_search_ctx(ctx); |
| 1593 | if (m) |
| 1594 | unmap_mft_record(base_ni); |
| 1595 | ntfs_error(vi->i_sb, "Failed to update initialized_size/i_size (error " |
| 1596 | "code %i).", err); |
Anton Altaparmakov | f95c401 | 2006-03-23 15:59:32 +0000 | [diff] [blame] | 1597 | if (err != -ENOMEM) |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1598 | NVolSetErrors(ni->vol); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1599 | return err; |
| 1600 | } |
| 1601 | |
| 1602 | /** |
| 1603 | * ntfs_commit_pages_after_write - commit the received data |
| 1604 | * @pages: array of destination pages |
| 1605 | * @nr_pages: number of pages in @pages |
| 1606 | * @pos: byte position in file at which the write begins |
| 1607 | * @bytes: number of bytes to be written |
| 1608 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1609 | * This is called from ntfs_file_buffered_write() with i_mutex held on the inode |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1610 | * (@pages[0]->mapping->host). There are @nr_pages pages in @pages which are |
| 1611 | * locked but not kmap()ped. The source data has already been copied into the |
| 1612 | * @page. ntfs_prepare_pages_for_non_resident_write() has been called before |
| 1613 | * the data was copied (for non-resident attributes only) and it returned |
| 1614 | * success. |
| 1615 | * |
| 1616 | * Need to set uptodate and mark dirty all buffers within the boundary of the |
| 1617 | * write. If all buffers in a page are uptodate we set the page uptodate, too. |
| 1618 | * |
| 1619 | * Setting the buffers dirty ensures that they get written out later when |
| 1620 | * ntfs_writepage() is invoked by the VM. |
| 1621 | * |
| 1622 | * Finally, we need to update i_size and initialized_size as appropriate both |
| 1623 | * in the inode and the mft record. |
| 1624 | * |
| 1625 | * This is modelled after fs/buffer.c::generic_commit_write(), which marks |
| 1626 | * buffers uptodate and dirty, sets the page uptodate if all buffers in the |
| 1627 | * page are uptodate, and updates i_size if the end of io is beyond i_size. In |
| 1628 | * that case, it also marks the inode dirty. |
| 1629 | * |
| 1630 | * If things have gone as outlined in |
| 1631 | * ntfs_prepare_pages_for_non_resident_write(), we do not need to do any page |
| 1632 | * content modifications here for non-resident attributes. For resident |
| 1633 | * attributes we need to do the uptodate bringing here which we combine with |
| 1634 | * the copying into the mft record which means we save one atomic kmap. |
| 1635 | * |
| 1636 | * Return 0 on success or -errno on error. |
| 1637 | */ |
| 1638 | static int ntfs_commit_pages_after_write(struct page **pages, |
| 1639 | const unsigned nr_pages, s64 pos, size_t bytes) |
| 1640 | { |
| 1641 | s64 end, initialized_size; |
| 1642 | loff_t i_size; |
| 1643 | struct inode *vi; |
| 1644 | ntfs_inode *ni, *base_ni; |
| 1645 | struct page *page; |
| 1646 | ntfs_attr_search_ctx *ctx; |
| 1647 | MFT_RECORD *m; |
| 1648 | ATTR_RECORD *a; |
| 1649 | char *kattr, *kaddr; |
| 1650 | unsigned long flags; |
| 1651 | u32 attr_len; |
| 1652 | int err; |
| 1653 | |
| 1654 | BUG_ON(!nr_pages); |
| 1655 | BUG_ON(!pages); |
| 1656 | page = pages[0]; |
| 1657 | BUG_ON(!page); |
| 1658 | vi = page->mapping->host; |
| 1659 | ni = NTFS_I(vi); |
| 1660 | ntfs_debug("Entering for inode 0x%lx, attribute type 0x%x, start page " |
Anton Altaparmakov | d04bd1f | 2005-10-24 08:41:24 +0100 | [diff] [blame] | 1661 | "index 0x%lx, nr_pages 0x%x, pos 0x%llx, bytes 0x%zx.", |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1662 | vi->i_ino, ni->type, page->index, nr_pages, |
| 1663 | (long long)pos, bytes); |
| 1664 | if (NInoNonResident(ni)) |
| 1665 | return ntfs_commit_pages_after_non_resident_write(pages, |
| 1666 | nr_pages, pos, bytes); |
| 1667 | BUG_ON(nr_pages > 1); |
| 1668 | /* |
| 1669 | * Attribute is resident, implying it is not compressed, encrypted, or |
| 1670 | * sparse. |
| 1671 | */ |
| 1672 | if (!NInoAttr(ni)) |
| 1673 | base_ni = ni; |
| 1674 | else |
| 1675 | base_ni = ni->ext.base_ntfs_ino; |
| 1676 | BUG_ON(NInoNonResident(ni)); |
| 1677 | /* Map, pin, and lock the mft record. */ |
| 1678 | m = map_mft_record(base_ni); |
| 1679 | if (IS_ERR(m)) { |
| 1680 | err = PTR_ERR(m); |
| 1681 | m = NULL; |
| 1682 | ctx = NULL; |
| 1683 | goto err_out; |
| 1684 | } |
| 1685 | ctx = ntfs_attr_get_search_ctx(base_ni, m); |
| 1686 | if (unlikely(!ctx)) { |
| 1687 | err = -ENOMEM; |
| 1688 | goto err_out; |
| 1689 | } |
| 1690 | err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len, |
| 1691 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 1692 | if (unlikely(err)) { |
| 1693 | if (err == -ENOENT) |
| 1694 | err = -EIO; |
| 1695 | goto err_out; |
| 1696 | } |
| 1697 | a = ctx->attr; |
| 1698 | BUG_ON(a->non_resident); |
| 1699 | /* The total length of the attribute value. */ |
| 1700 | attr_len = le32_to_cpu(a->data.resident.value_length); |
| 1701 | i_size = i_size_read(vi); |
| 1702 | BUG_ON(attr_len != i_size); |
| 1703 | BUG_ON(pos > attr_len); |
| 1704 | end = pos + bytes; |
| 1705 | BUG_ON(end > le32_to_cpu(a->length) - |
| 1706 | le16_to_cpu(a->data.resident.value_offset)); |
| 1707 | kattr = (u8*)a + le16_to_cpu(a->data.resident.value_offset); |
| 1708 | kaddr = kmap_atomic(page, KM_USER0); |
| 1709 | /* Copy the received data from the page to the mft record. */ |
| 1710 | memcpy(kattr + pos, kaddr + pos, bytes); |
| 1711 | /* Update the attribute length if necessary. */ |
| 1712 | if (end > attr_len) { |
| 1713 | attr_len = end; |
| 1714 | a->data.resident.value_length = cpu_to_le32(attr_len); |
| 1715 | } |
| 1716 | /* |
| 1717 | * If the page is not uptodate, bring the out of bounds area(s) |
| 1718 | * uptodate by copying data from the mft record to the page. |
| 1719 | */ |
| 1720 | if (!PageUptodate(page)) { |
| 1721 | if (pos > 0) |
| 1722 | memcpy(kaddr, kattr, pos); |
| 1723 | if (end < attr_len) |
| 1724 | memcpy(kaddr + end, kattr + end, attr_len - end); |
| 1725 | /* Zero the region outside the end of the attribute value. */ |
| 1726 | memset(kaddr + attr_len, 0, PAGE_CACHE_SIZE - attr_len); |
| 1727 | flush_dcache_page(page); |
| 1728 | SetPageUptodate(page); |
| 1729 | } |
| 1730 | kunmap_atomic(kaddr, KM_USER0); |
| 1731 | /* Update initialized_size/i_size if necessary. */ |
| 1732 | read_lock_irqsave(&ni->size_lock, flags); |
| 1733 | initialized_size = ni->initialized_size; |
| 1734 | BUG_ON(end > ni->allocated_size); |
| 1735 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1736 | BUG_ON(initialized_size != i_size); |
| 1737 | if (end > initialized_size) { |
| 1738 | unsigned long flags; |
| 1739 | |
| 1740 | write_lock_irqsave(&ni->size_lock, flags); |
| 1741 | ni->initialized_size = end; |
| 1742 | i_size_write(vi, end); |
| 1743 | write_unlock_irqrestore(&ni->size_lock, flags); |
| 1744 | } |
| 1745 | /* Mark the mft record dirty, so it gets written back. */ |
| 1746 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1747 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1748 | ntfs_attr_put_search_ctx(ctx); |
| 1749 | unmap_mft_record(base_ni); |
| 1750 | ntfs_debug("Done."); |
| 1751 | return 0; |
| 1752 | err_out: |
| 1753 | if (err == -ENOMEM) { |
| 1754 | ntfs_warning(vi->i_sb, "Error allocating memory required to " |
| 1755 | "commit the write."); |
| 1756 | if (PageUptodate(page)) { |
| 1757 | ntfs_warning(vi->i_sb, "Page is uptodate, setting " |
| 1758 | "dirty so the write will be retried " |
| 1759 | "later on by the VM."); |
| 1760 | /* |
| 1761 | * Put the page on mapping->dirty_pages, but leave its |
| 1762 | * buffers' dirty state as-is. |
| 1763 | */ |
| 1764 | __set_page_dirty_nobuffers(page); |
| 1765 | err = 0; |
| 1766 | } else |
| 1767 | ntfs_error(vi->i_sb, "Page is not uptodate. Written " |
| 1768 | "data has been lost."); |
| 1769 | } else { |
| 1770 | ntfs_error(vi->i_sb, "Resident attribute commit write failed " |
| 1771 | "with error %i.", err); |
| 1772 | NVolSetErrors(ni->vol); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1773 | } |
| 1774 | if (ctx) |
| 1775 | ntfs_attr_put_search_ctx(ctx); |
| 1776 | if (m) |
| 1777 | unmap_mft_record(base_ni); |
| 1778 | return err; |
| 1779 | } |
| 1780 | |
| 1781 | /** |
| 1782 | * ntfs_file_buffered_write - |
| 1783 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 1784 | * Locking: The vfs is holding ->i_mutex on the inode. |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1785 | */ |
| 1786 | static ssize_t ntfs_file_buffered_write(struct kiocb *iocb, |
| 1787 | const struct iovec *iov, unsigned long nr_segs, |
| 1788 | loff_t pos, loff_t *ppos, size_t count) |
| 1789 | { |
| 1790 | struct file *file = iocb->ki_filp; |
| 1791 | struct address_space *mapping = file->f_mapping; |
| 1792 | struct inode *vi = mapping->host; |
| 1793 | ntfs_inode *ni = NTFS_I(vi); |
| 1794 | ntfs_volume *vol = ni->vol; |
| 1795 | struct page *pages[NTFS_MAX_PAGES_PER_CLUSTER]; |
| 1796 | struct page *cached_page = NULL; |
| 1797 | char __user *buf = NULL; |
| 1798 | s64 end, ll; |
| 1799 | VCN last_vcn; |
| 1800 | LCN lcn; |
| 1801 | unsigned long flags; |
Anton Altaparmakov | dda65b9 | 2005-10-24 08:57:59 +0100 | [diff] [blame] | 1802 | size_t bytes, iov_ofs = 0; /* Offset in the current iovec. */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1803 | ssize_t status, written; |
| 1804 | unsigned nr_pages; |
| 1805 | int err; |
| 1806 | struct pagevec lru_pvec; |
| 1807 | |
| 1808 | ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, " |
| 1809 | "pos 0x%llx, count 0x%lx.", |
| 1810 | vi->i_ino, (unsigned)le32_to_cpu(ni->type), |
| 1811 | (unsigned long long)pos, (unsigned long)count); |
| 1812 | if (unlikely(!count)) |
| 1813 | return 0; |
| 1814 | BUG_ON(NInoMstProtected(ni)); |
| 1815 | /* |
| 1816 | * If the attribute is not an index root and it is encrypted or |
| 1817 | * compressed, we cannot write to it yet. Note we need to check for |
| 1818 | * AT_INDEX_ALLOCATION since this is the type of both directory and |
| 1819 | * index inodes. |
| 1820 | */ |
| 1821 | if (ni->type != AT_INDEX_ALLOCATION) { |
| 1822 | /* If file is encrypted, deny access, just like NT4. */ |
| 1823 | if (NInoEncrypted(ni)) { |
Anton Altaparmakov | 7d0ffdb | 2005-10-19 12:21:19 +0100 | [diff] [blame] | 1824 | /* |
| 1825 | * Reminder for later: Encrypted files are _always_ |
| 1826 | * non-resident so that the content can always be |
| 1827 | * encrypted. |
| 1828 | */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1829 | ntfs_debug("Denying write access to encrypted file."); |
| 1830 | return -EACCES; |
| 1831 | } |
| 1832 | if (NInoCompressed(ni)) { |
Anton Altaparmakov | 7d0ffdb | 2005-10-19 12:21:19 +0100 | [diff] [blame] | 1833 | /* Only unnamed $DATA attribute can be compressed. */ |
| 1834 | BUG_ON(ni->type != AT_DATA); |
| 1835 | BUG_ON(ni->name_len); |
| 1836 | /* |
| 1837 | * Reminder for later: If resident, the data is not |
| 1838 | * actually compressed. Only on the switch to non- |
| 1839 | * resident does compression kick in. This is in |
| 1840 | * contrast to encrypted files (see above). |
| 1841 | */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1842 | ntfs_error(vi->i_sb, "Writing to compressed files is " |
| 1843 | "not implemented yet. Sorry."); |
| 1844 | return -EOPNOTSUPP; |
| 1845 | } |
| 1846 | } |
| 1847 | /* |
| 1848 | * If a previous ntfs_truncate() failed, repeat it and abort if it |
| 1849 | * fails again. |
| 1850 | */ |
| 1851 | if (unlikely(NInoTruncateFailed(ni))) { |
| 1852 | down_write(&vi->i_alloc_sem); |
| 1853 | err = ntfs_truncate(vi); |
| 1854 | up_write(&vi->i_alloc_sem); |
| 1855 | if (err || NInoTruncateFailed(ni)) { |
| 1856 | if (!err) |
| 1857 | err = -EIO; |
| 1858 | ntfs_error(vol->sb, "Cannot perform write to inode " |
| 1859 | "0x%lx, attribute type 0x%x, because " |
| 1860 | "ntfs_truncate() failed (error code " |
| 1861 | "%i).", vi->i_ino, |
| 1862 | (unsigned)le32_to_cpu(ni->type), err); |
| 1863 | return err; |
| 1864 | } |
| 1865 | } |
| 1866 | /* The first byte after the write. */ |
| 1867 | end = pos + count; |
| 1868 | /* |
| 1869 | * If the write goes beyond the allocated size, extend the allocation |
| 1870 | * to cover the whole of the write, rounded up to the nearest cluster. |
| 1871 | */ |
| 1872 | read_lock_irqsave(&ni->size_lock, flags); |
| 1873 | ll = ni->allocated_size; |
| 1874 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1875 | if (end > ll) { |
| 1876 | /* Extend the allocation without changing the data size. */ |
| 1877 | ll = ntfs_attr_extend_allocation(ni, end, -1, pos); |
| 1878 | if (likely(ll >= 0)) { |
| 1879 | BUG_ON(pos >= ll); |
| 1880 | /* If the extension was partial truncate the write. */ |
| 1881 | if (end > ll) { |
| 1882 | ntfs_debug("Truncating write to inode 0x%lx, " |
| 1883 | "attribute type 0x%x, because " |
| 1884 | "the allocation was only " |
| 1885 | "partially extended.", |
| 1886 | vi->i_ino, (unsigned) |
| 1887 | le32_to_cpu(ni->type)); |
| 1888 | end = ll; |
| 1889 | count = ll - pos; |
| 1890 | } |
| 1891 | } else { |
| 1892 | err = ll; |
| 1893 | read_lock_irqsave(&ni->size_lock, flags); |
| 1894 | ll = ni->allocated_size; |
| 1895 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1896 | /* Perform a partial write if possible or fail. */ |
| 1897 | if (pos < ll) { |
| 1898 | ntfs_debug("Truncating write to inode 0x%lx, " |
| 1899 | "attribute type 0x%x, because " |
| 1900 | "extending the allocation " |
| 1901 | "failed (error code %i).", |
| 1902 | vi->i_ino, (unsigned) |
| 1903 | le32_to_cpu(ni->type), err); |
| 1904 | end = ll; |
| 1905 | count = ll - pos; |
| 1906 | } else { |
| 1907 | ntfs_error(vol->sb, "Cannot perform write to " |
| 1908 | "inode 0x%lx, attribute type " |
| 1909 | "0x%x, because extending the " |
| 1910 | "allocation failed (error " |
| 1911 | "code %i).", vi->i_ino, |
| 1912 | (unsigned) |
| 1913 | le32_to_cpu(ni->type), err); |
| 1914 | return err; |
| 1915 | } |
| 1916 | } |
| 1917 | } |
| 1918 | pagevec_init(&lru_pvec, 0); |
| 1919 | written = 0; |
| 1920 | /* |
| 1921 | * If the write starts beyond the initialized size, extend it up to the |
| 1922 | * beginning of the write and initialize all non-sparse space between |
| 1923 | * the old initialized size and the new one. This automatically also |
| 1924 | * increments the vfs inode->i_size to keep it above or equal to the |
| 1925 | * initialized_size. |
| 1926 | */ |
| 1927 | read_lock_irqsave(&ni->size_lock, flags); |
| 1928 | ll = ni->initialized_size; |
| 1929 | read_unlock_irqrestore(&ni->size_lock, flags); |
| 1930 | if (pos > ll) { |
| 1931 | err = ntfs_attr_extend_initialized(ni, pos, &cached_page, |
| 1932 | &lru_pvec); |
| 1933 | if (err < 0) { |
| 1934 | ntfs_error(vol->sb, "Cannot perform write to inode " |
| 1935 | "0x%lx, attribute type 0x%x, because " |
| 1936 | "extending the initialized size " |
| 1937 | "failed (error code %i).", vi->i_ino, |
| 1938 | (unsigned)le32_to_cpu(ni->type), err); |
| 1939 | status = err; |
| 1940 | goto err_out; |
| 1941 | } |
| 1942 | } |
| 1943 | /* |
| 1944 | * Determine the number of pages per cluster for non-resident |
| 1945 | * attributes. |
| 1946 | */ |
| 1947 | nr_pages = 1; |
| 1948 | if (vol->cluster_size > PAGE_CACHE_SIZE && NInoNonResident(ni)) |
| 1949 | nr_pages = vol->cluster_size >> PAGE_CACHE_SHIFT; |
| 1950 | /* Finally, perform the actual write. */ |
| 1951 | last_vcn = -1; |
| 1952 | if (likely(nr_segs == 1)) |
| 1953 | buf = iov->iov_base; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1954 | do { |
| 1955 | VCN vcn; |
| 1956 | pgoff_t idx, start_idx; |
| 1957 | unsigned ofs, do_pages, u; |
| 1958 | size_t copied; |
| 1959 | |
| 1960 | start_idx = idx = pos >> PAGE_CACHE_SHIFT; |
| 1961 | ofs = pos & ~PAGE_CACHE_MASK; |
| 1962 | bytes = PAGE_CACHE_SIZE - ofs; |
| 1963 | do_pages = 1; |
| 1964 | if (nr_pages > 1) { |
| 1965 | vcn = pos >> vol->cluster_size_bits; |
| 1966 | if (vcn != last_vcn) { |
| 1967 | last_vcn = vcn; |
| 1968 | /* |
| 1969 | * Get the lcn of the vcn the write is in. If |
| 1970 | * it is a hole, need to lock down all pages in |
| 1971 | * the cluster. |
| 1972 | */ |
| 1973 | down_read(&ni->runlist.lock); |
| 1974 | lcn = ntfs_attr_vcn_to_lcn_nolock(ni, pos >> |
Richard Knutsson | c49c311 | 2006-09-30 23:27:12 -0700 | [diff] [blame] | 1975 | vol->cluster_size_bits, false); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 1976 | up_read(&ni->runlist.lock); |
| 1977 | if (unlikely(lcn < LCN_HOLE)) { |
| 1978 | status = -EIO; |
| 1979 | if (lcn == LCN_ENOMEM) |
| 1980 | status = -ENOMEM; |
| 1981 | else |
| 1982 | ntfs_error(vol->sb, "Cannot " |
| 1983 | "perform write to " |
| 1984 | "inode 0x%lx, " |
| 1985 | "attribute type 0x%x, " |
| 1986 | "because the attribute " |
| 1987 | "is corrupt.", |
| 1988 | vi->i_ino, (unsigned) |
| 1989 | le32_to_cpu(ni->type)); |
| 1990 | break; |
| 1991 | } |
| 1992 | if (lcn == LCN_HOLE) { |
| 1993 | start_idx = (pos & ~(s64) |
| 1994 | vol->cluster_size_mask) |
| 1995 | >> PAGE_CACHE_SHIFT; |
| 1996 | bytes = vol->cluster_size - (pos & |
| 1997 | vol->cluster_size_mask); |
| 1998 | do_pages = nr_pages; |
| 1999 | } |
| 2000 | } |
| 2001 | } |
| 2002 | if (bytes > count) |
| 2003 | bytes = count; |
| 2004 | /* |
| 2005 | * Bring in the user page(s) that we will copy from _first_. |
| 2006 | * Otherwise there is a nasty deadlock on copying from the same |
| 2007 | * page(s) as we are writing to, without it/them being marked |
| 2008 | * up-to-date. Note, at present there is nothing to stop the |
| 2009 | * pages being swapped out between us bringing them into memory |
| 2010 | * and doing the actual copying. |
| 2011 | */ |
| 2012 | if (likely(nr_segs == 1)) |
| 2013 | ntfs_fault_in_pages_readable(buf, bytes); |
| 2014 | else |
| 2015 | ntfs_fault_in_pages_readable_iovec(iov, iov_ofs, bytes); |
| 2016 | /* Get and lock @do_pages starting at index @start_idx. */ |
| 2017 | status = __ntfs_grab_cache_pages(mapping, start_idx, do_pages, |
| 2018 | pages, &cached_page, &lru_pvec); |
| 2019 | if (unlikely(status)) |
| 2020 | break; |
| 2021 | /* |
| 2022 | * For non-resident attributes, we need to fill any holes with |
| 2023 | * actual clusters and ensure all bufferes are mapped. We also |
| 2024 | * need to bring uptodate any buffers that are only partially |
| 2025 | * being written to. |
| 2026 | */ |
| 2027 | if (NInoNonResident(ni)) { |
| 2028 | status = ntfs_prepare_pages_for_non_resident_write( |
| 2029 | pages, do_pages, pos, bytes); |
| 2030 | if (unlikely(status)) { |
| 2031 | loff_t i_size; |
| 2032 | |
| 2033 | do { |
| 2034 | unlock_page(pages[--do_pages]); |
| 2035 | page_cache_release(pages[do_pages]); |
| 2036 | } while (do_pages); |
| 2037 | /* |
| 2038 | * The write preparation may have instantiated |
| 2039 | * allocated space outside i_size. Trim this |
| 2040 | * off again. We can ignore any errors in this |
| 2041 | * case as we will just be waisting a bit of |
| 2042 | * allocated space, which is not a disaster. |
| 2043 | */ |
| 2044 | i_size = i_size_read(vi); |
| 2045 | if (pos + bytes > i_size) |
| 2046 | vmtruncate(vi, i_size); |
| 2047 | break; |
| 2048 | } |
| 2049 | } |
| 2050 | u = (pos >> PAGE_CACHE_SHIFT) - pages[0]->index; |
| 2051 | if (likely(nr_segs == 1)) { |
| 2052 | copied = ntfs_copy_from_user(pages + u, do_pages - u, |
| 2053 | ofs, buf, bytes); |
| 2054 | buf += copied; |
| 2055 | } else |
| 2056 | copied = ntfs_copy_from_user_iovec(pages + u, |
| 2057 | do_pages - u, ofs, &iov, &iov_ofs, |
| 2058 | bytes); |
| 2059 | ntfs_flush_dcache_pages(pages + u, do_pages - u); |
| 2060 | status = ntfs_commit_pages_after_write(pages, do_pages, pos, |
| 2061 | bytes); |
| 2062 | if (likely(!status)) { |
| 2063 | written += copied; |
| 2064 | count -= copied; |
| 2065 | pos += copied; |
| 2066 | if (unlikely(copied != bytes)) |
| 2067 | status = -EFAULT; |
| 2068 | } |
| 2069 | do { |
| 2070 | unlock_page(pages[--do_pages]); |
| 2071 | mark_page_accessed(pages[do_pages]); |
| 2072 | page_cache_release(pages[do_pages]); |
| 2073 | } while (do_pages); |
| 2074 | if (unlikely(status)) |
| 2075 | break; |
| 2076 | balance_dirty_pages_ratelimited(mapping); |
| 2077 | cond_resched(); |
| 2078 | } while (count); |
| 2079 | err_out: |
| 2080 | *ppos = pos; |
| 2081 | if (cached_page) |
| 2082 | page_cache_release(cached_page); |
| 2083 | /* For now, when the user asks for O_SYNC, we actually give O_DSYNC. */ |
| 2084 | if (likely(!status)) { |
| 2085 | if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(vi))) { |
| 2086 | if (!mapping->a_ops->writepage || !is_sync_kiocb(iocb)) |
| 2087 | status = generic_osync_inode(vi, mapping, |
| 2088 | OSYNC_METADATA|OSYNC_DATA); |
| 2089 | } |
| 2090 | } |
| 2091 | pagevec_lru_add(&lru_pvec); |
| 2092 | ntfs_debug("Done. Returning %s (written 0x%lx, status %li).", |
| 2093 | written ? "written" : "status", (unsigned long)written, |
| 2094 | (long)status); |
| 2095 | return written ? written : status; |
| 2096 | } |
| 2097 | |
| 2098 | /** |
| 2099 | * ntfs_file_aio_write_nolock - |
| 2100 | */ |
| 2101 | static ssize_t ntfs_file_aio_write_nolock(struct kiocb *iocb, |
| 2102 | const struct iovec *iov, unsigned long nr_segs, loff_t *ppos) |
| 2103 | { |
| 2104 | struct file *file = iocb->ki_filp; |
| 2105 | struct address_space *mapping = file->f_mapping; |
| 2106 | struct inode *inode = mapping->host; |
| 2107 | loff_t pos; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2108 | size_t count; /* after file limit checks */ |
| 2109 | ssize_t written, err; |
| 2110 | |
| 2111 | count = 0; |
Dmitriy Monakhov | 0ceb331 | 2007-05-08 00:23:02 -0700 | [diff] [blame] | 2112 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); |
| 2113 | if (err) |
| 2114 | return err; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2115 | pos = *ppos; |
| 2116 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
| 2117 | /* We can write back this queue in page reclaim. */ |
| 2118 | current->backing_dev_info = mapping->backing_dev_info; |
| 2119 | written = 0; |
| 2120 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
| 2121 | if (err) |
| 2122 | goto out; |
| 2123 | if (!count) |
| 2124 | goto out; |
Josef "Jeff" Sipek | 20d2937 | 2006-12-08 02:36:41 -0800 | [diff] [blame] | 2125 | err = remove_suid(file->f_path.dentry); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2126 | if (err) |
| 2127 | goto out; |
Christoph Hellwig | 870f481 | 2006-01-09 20:52:01 -0800 | [diff] [blame] | 2128 | file_update_time(file); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2129 | written = ntfs_file_buffered_write(iocb, iov, nr_segs, pos, ppos, |
| 2130 | count); |
| 2131 | out: |
| 2132 | current->backing_dev_info = NULL; |
| 2133 | return written ? written : err; |
| 2134 | } |
| 2135 | |
| 2136 | /** |
| 2137 | * ntfs_file_aio_write - |
| 2138 | */ |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2139 | static ssize_t ntfs_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
| 2140 | unsigned long nr_segs, loff_t pos) |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2141 | { |
| 2142 | struct file *file = iocb->ki_filp; |
| 2143 | struct address_space *mapping = file->f_mapping; |
| 2144 | struct inode *inode = mapping->host; |
| 2145 | ssize_t ret; |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2146 | |
| 2147 | BUG_ON(iocb->ki_pos != pos); |
| 2148 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2149 | mutex_lock(&inode->i_mutex); |
Badari Pulavarty | 027445c | 2006-09-30 23:28:46 -0700 | [diff] [blame] | 2150 | ret = ntfs_file_aio_write_nolock(iocb, iov, nr_segs, &iocb->ki_pos); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2151 | mutex_unlock(&inode->i_mutex); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2152 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { |
| 2153 | int err = sync_page_range(inode, mapping, pos, ret); |
| 2154 | if (err < 0) |
| 2155 | ret = err; |
| 2156 | } |
| 2157 | return ret; |
| 2158 | } |
| 2159 | |
| 2160 | /** |
| 2161 | * ntfs_file_writev - |
| 2162 | * |
| 2163 | * Basically the same as generic_file_writev() except that it ends up calling |
| 2164 | * ntfs_file_aio_write_nolock() instead of __generic_file_aio_write_nolock(). |
| 2165 | */ |
| 2166 | static ssize_t ntfs_file_writev(struct file *file, const struct iovec *iov, |
| 2167 | unsigned long nr_segs, loff_t *ppos) |
| 2168 | { |
| 2169 | struct address_space *mapping = file->f_mapping; |
| 2170 | struct inode *inode = mapping->host; |
| 2171 | struct kiocb kiocb; |
| 2172 | ssize_t ret; |
| 2173 | |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2174 | mutex_lock(&inode->i_mutex); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2175 | init_sync_kiocb(&kiocb, file); |
| 2176 | ret = ntfs_file_aio_write_nolock(&kiocb, iov, nr_segs, ppos); |
| 2177 | if (ret == -EIOCBQUEUED) |
| 2178 | ret = wait_on_sync_kiocb(&kiocb); |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2179 | mutex_unlock(&inode->i_mutex); |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2180 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { |
| 2181 | int err = sync_page_range(inode, mapping, *ppos - ret, ret); |
| 2182 | if (err < 0) |
| 2183 | ret = err; |
| 2184 | } |
| 2185 | return ret; |
| 2186 | } |
| 2187 | |
| 2188 | /** |
| 2189 | * ntfs_file_write - simple wrapper for ntfs_file_writev() |
| 2190 | */ |
| 2191 | static ssize_t ntfs_file_write(struct file *file, const char __user *buf, |
| 2192 | size_t count, loff_t *ppos) |
| 2193 | { |
| 2194 | struct iovec local_iov = { .iov_base = (void __user *)buf, |
| 2195 | .iov_len = count }; |
| 2196 | |
| 2197 | return ntfs_file_writev(file, &local_iov, 1, ppos); |
| 2198 | } |
| 2199 | |
| 2200 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2201 | * ntfs_file_fsync - sync a file to disk |
| 2202 | * @filp: file to be synced |
| 2203 | * @dentry: dentry describing the file to sync |
| 2204 | * @datasync: if non-zero only flush user data and not metadata |
| 2205 | * |
| 2206 | * Data integrity sync of a file to disk. Used for fsync, fdatasync, and msync |
| 2207 | * system calls. This function is inspired by fs/buffer.c::file_fsync(). |
| 2208 | * |
| 2209 | * If @datasync is false, write the mft record and all associated extent mft |
| 2210 | * records as well as the $DATA attribute and then sync the block device. |
| 2211 | * |
| 2212 | * If @datasync is true and the attribute is non-resident, we skip the writing |
| 2213 | * of the mft record and all associated extent mft records (this might still |
| 2214 | * happen due to the write_inode_now() call). |
| 2215 | * |
| 2216 | * Also, if @datasync is true, we do not wait on the inode to be written out |
| 2217 | * but we always wait on the page cache pages to be written out. |
| 2218 | * |
| 2219 | * Note: In the past @filp could be NULL so we ignore it as we don't need it |
| 2220 | * anyway. |
| 2221 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 2222 | * Locking: Caller must hold i_mutex on the inode. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2223 | * |
| 2224 | * TODO: We should probably also write all attribute/index inodes associated |
| 2225 | * with this inode but since we have no simple way of getting to them we ignore |
| 2226 | * this problem for now. |
| 2227 | */ |
| 2228 | static int ntfs_file_fsync(struct file *filp, struct dentry *dentry, |
| 2229 | int datasync) |
| 2230 | { |
| 2231 | struct inode *vi = dentry->d_inode; |
| 2232 | int err, ret = 0; |
| 2233 | |
| 2234 | ntfs_debug("Entering for inode 0x%lx.", vi->i_ino); |
| 2235 | BUG_ON(S_ISDIR(vi->i_mode)); |
| 2236 | if (!datasync || !NInoNonResident(NTFS_I(vi))) |
| 2237 | ret = ntfs_write_inode(vi, 1); |
| 2238 | write_inode_now(vi, !datasync); |
Anton Altaparmakov | f25dfb5 | 2005-09-08 20:35:33 +0100 | [diff] [blame] | 2239 | /* |
| 2240 | * NOTE: If we were to use mapping->private_list (see ext2 and |
| 2241 | * fs/buffer.c) for dirty blocks then we could optimize the below to be |
| 2242 | * sync_mapping_buffers(vi->i_mapping). |
| 2243 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2244 | err = sync_blockdev(vi->i_sb->s_bdev); |
| 2245 | if (unlikely(err && !ret)) |
| 2246 | ret = err; |
| 2247 | if (likely(!ret)) |
| 2248 | ntfs_debug("Done."); |
| 2249 | else |
| 2250 | ntfs_warning(vi->i_sb, "Failed to f%ssync inode 0x%lx. Error " |
| 2251 | "%u.", datasync ? "data" : "", vi->i_ino, -ret); |
| 2252 | return ret; |
| 2253 | } |
| 2254 | |
| 2255 | #endif /* NTFS_RW */ |
| 2256 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2257 | const struct file_operations ntfs_file_ops = { |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2258 | .llseek = generic_file_llseek, /* Seek inside file. */ |
Badari Pulavarty | 543ade1 | 2006-09-30 23:28:48 -0700 | [diff] [blame] | 2259 | .read = do_sync_read, /* Read from file. */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2260 | .aio_read = generic_file_aio_read, /* Async read from file. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2261 | #ifdef NTFS_RW |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2262 | .write = ntfs_file_write, /* Write to file. */ |
| 2263 | .aio_write = ntfs_file_aio_write, /* Async write to file. */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2264 | /*.release = ,*/ /* Last file is closed. See |
| 2265 | fs/ext2/file.c:: |
| 2266 | ext2_release_file() for |
| 2267 | how to use this to discard |
| 2268 | preallocated space for |
| 2269 | write opened files. */ |
| 2270 | .fsync = ntfs_file_fsync, /* Sync a file to disk. */ |
| 2271 | /*.aio_fsync = ,*/ /* Sync all outstanding async |
| 2272 | i/o operations on a |
| 2273 | kiocb. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2274 | #endif /* NTFS_RW */ |
Anton Altaparmakov | 98b2703 | 2005-10-11 15:40:40 +0100 | [diff] [blame] | 2275 | /*.ioctl = ,*/ /* Perform function on the |
| 2276 | mounted filesystem. */ |
| 2277 | .mmap = generic_file_mmap, /* Mmap file. */ |
| 2278 | .open = ntfs_file_open, /* Open file. */ |
| 2279 | .sendfile = generic_file_sendfile, /* Zero-copy data send with |
| 2280 | the data source being on |
| 2281 | the ntfs partition. We do |
| 2282 | not need to care about the |
| 2283 | data destination. */ |
| 2284 | /*.sendpage = ,*/ /* Zero-copy data send with |
| 2285 | the data destination being |
| 2286 | on the ntfs partition. We |
| 2287 | do not need to care about |
| 2288 | the data source. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2289 | }; |
| 2290 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2291 | const struct inode_operations ntfs_file_inode_ops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2292 | #ifdef NTFS_RW |
| 2293 | .truncate = ntfs_truncate_vfs, |
| 2294 | .setattr = ntfs_setattr, |
| 2295 | #endif /* NTFS_RW */ |
| 2296 | }; |
| 2297 | |
Arjan van de Ven | 4b6f5d2 | 2006-03-28 01:56:42 -0800 | [diff] [blame] | 2298 | const struct file_operations ntfs_empty_file_ops = {}; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2299 | |
Arjan van de Ven | 92e1d5b | 2007-02-12 00:55:39 -0800 | [diff] [blame] | 2300 | const struct inode_operations ntfs_empty_inode_ops = {}; |