fs/xfs/linux-2.6/xfs_file.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_bit.h"
  20 #include "xfs_log.h"
  21 #include "xfs_inum.h"
  22 #include "xfs_sb.h"
  23 #include "xfs_ag.h"
  24 #include "xfs_dir2.h"
  25 #include "xfs_trans.h"
  26 #include "xfs_dmapi.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_alloc_btree.h"
  30 #include "xfs_ialloc_btree.h"
  31 #include "xfs_alloc.h"
  32 #include "xfs_btree.h"
  33 #include "xfs_attr_sf.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_error.h"
  38 #include "xfs_rw.h"
  39 #include "xfs_ioctl32.h"
  40 #include "xfs_vnodeops.h"
  41
  42 #include <linux/dcache.h>
  43 #include <linux/smp_lock.h>
  44
  45 static struct vm_operations_struct xfs_file_vm_ops;
  46 #ifdef CONFIG_XFS_DMAPI
  47 static struct vm_operations_struct xfs_dmapi_file_vm_ops;
  48 #endif
  49
  50 STATIC_INLINE ssize_t
  51 __xfs_file_read(
  52         struct kiocb            *iocb,
  53         const struct iovec      *iov,
  54         unsigned long           nr_segs,
  55         int                     ioflags,
  56         loff_t                  pos)
  57 {
  58         struct file             *file = iocb->ki_filp;
  59
  60         BUG_ON(iocb->ki_pos != pos);
  61         if (unlikely(file->f_flags & O_DIRECT))
  62                 ioflags |= IO_ISDIRECT;
  63         return xfs_read(XFS_I(file->f_path.dentry->d_inode), iocb, iov,
  64                                 nr_segs, &iocb->ki_pos, ioflags);
  65 }
  66
  67 STATIC ssize_t
  68 xfs_file_aio_read(
  69         struct kiocb            *iocb,
  70         const struct iovec      *iov,
  71         unsigned long           nr_segs,
  72         loff_t                  pos)
  73 {
  74         return __xfs_file_read(iocb, iov, nr_segs, IO_ISAIO, pos);
  75 }
  76
  77 STATIC ssize_t
  78 xfs_file_aio_read_invis(
  79         struct kiocb            *iocb,
  80         const struct iovec      *iov,
  81         unsigned long           nr_segs,
  82         loff_t                  pos)
  83 {
  84         return __xfs_file_read(iocb, iov, nr_segs, IO_ISAIO|IO_INVIS, pos);
  85 }
  86
  87 STATIC_INLINE ssize_t
  88 __xfs_file_write(
  89         struct kiocb            *iocb,
  90         const struct iovec      *iov,
  91         unsigned long           nr_segs,
  92         int                     ioflags,
  93         loff_t                  pos)
  94 {
  95         struct file     *file = iocb->ki_filp;
  96
  97         BUG_ON(iocb->ki_pos != pos);
  98         if (unlikely(file->f_flags & O_DIRECT))
  99                 ioflags |= IO_ISDIRECT;
 100         return xfs_write(XFS_I(file->f_mapping->host), iocb, iov, nr_segs,
 101                                 &iocb->ki_pos, ioflags);
 102 }
 103
 104 STATIC ssize_t
 105 xfs_file_aio_write(
 106         struct kiocb            *iocb,
 107         const struct iovec      *iov,
 108         unsigned long           nr_segs,
 109         loff_t                  pos)
 110 {
 111         return __xfs_file_write(iocb, iov, nr_segs, IO_ISAIO, pos);
 112 }
 113
 114 STATIC ssize_t
 115 xfs_file_aio_write_invis(
 116         struct kiocb            *iocb,
 117         const struct iovec      *iov,
 118         unsigned long           nr_segs,
 119         loff_t                  pos)
 120 {
 121         return __xfs_file_write(iocb, iov, nr_segs, IO_ISAIO|IO_INVIS, pos);
 122 }
 123
 124 STATIC ssize_t
 125 xfs_file_splice_read(
 126         struct file             *infilp,
 127         loff_t                  *ppos,
 128         struct pipe_inode_info  *pipe,
 129         size_t                  len,
 130         unsigned int            flags)
 131 {
 132         return xfs_splice_read(XFS_I(infilp->f_path.dentry->d_inode),
 133                                    infilp, ppos, pipe, len, flags, 0);
 134 }
 135
 136 STATIC ssize_t
 137 xfs_file_splice_read_invis(
 138         struct file             *infilp,
 139         loff_t                  *ppos,
 140         struct pipe_inode_info  *pipe,
 141         size_t                  len,
 142         unsigned int            flags)
 143 {
 144         return xfs_splice_read(XFS_I(infilp->f_path.dentry->d_inode),
 145                                    infilp, ppos, pipe, len, flags, IO_INVIS);
 146 }
 147
 148 STATIC ssize_t
 149 xfs_file_splice_write(
 150         struct pipe_inode_info  *pipe,
 151         struct file             *outfilp,
 152         loff_t                  *ppos,
 153         size_t                  len,
 154         unsigned int            flags)
 155 {
 156         return xfs_splice_write(XFS_I(outfilp->f_path.dentry->d_inode),
 157                                     pipe, outfilp, ppos, len, flags, 0);
 158 }
 159
 160 STATIC ssize_t
 161 xfs_file_splice_write_invis(
 162         struct pipe_inode_info  *pipe,
 163         struct file             *outfilp,
 164         loff_t                  *ppos,
 165         size_t                  len,
 166         unsigned int            flags)
 167 {
 168         return xfs_splice_write(XFS_I(outfilp->f_path.dentry->d_inode),
 169                                     pipe, outfilp, ppos, len, flags, IO_INVIS);
 170 }
 171
 172 STATIC int
 173 xfs_file_open(
 174         struct inode    *inode,
 175         struct file     *filp)
 176 {
 177         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
 178                 return -EFBIG;
 179         return -xfs_open(XFS_I(inode));
 180 }
 181
 182 STATIC int
 183 xfs_file_release(
 184         struct inode    *inode,
 185         struct file     *filp)
 186 {
 187         return -xfs_release(XFS_I(inode));
 188 }
 189
 190 STATIC int
 191 xfs_file_fsync(
 192         struct file     *filp,
 193         struct dentry   *dentry,
 194         int             datasync)
 195 {
 196         int             flags = FSYNC_WAIT;
 197
 198         if (datasync)
 199                 flags |= FSYNC_DATA;
 200         xfs_iflags_clear(XFS_I(dentry->d_inode), XFS_ITRUNCATED);
 201         return -xfs_fsync(XFS_I(dentry->d_inode), flags,
 202                         (xfs_off_t)0, (xfs_off_t)-1);
 203 }
 204
 205 #ifdef CONFIG_XFS_DMAPI
 206 STATIC int
 207 xfs_vm_fault(
 208         struct vm_area_struct   *vma,
 209         struct vm_fault *vmf)
 210 {
 211         struct inode    *inode = vma->vm_file->f_path.dentry->d_inode;
 212         bhv_vnode_t     *vp = vn_from_inode(inode);
 213
 214         ASSERT_ALWAYS(vp->v_vfsp->vfs_flag & VFS_DMI);
 215         if (XFS_SEND_MMAP(XFS_VFSTOM(vp->v_vfsp), vma, 0))
 216                 return VM_FAULT_SIGBUS;
 217         return filemap_fault(vma, vmf);
 218 }
 219 #endif /* CONFIG_XFS_DMAPI */
 220
 221 /*
 222  * Unfortunately we can't just use the clean and simple readdir implementation
 223  * below, because nfs might call back into ->lookup from the filldir callback
 224  * and that will deadlock the low-level btree code.
 225  *
 226  * Hopefully we'll find a better workaround that allows to use the optimal
 227  * version at least for local readdirs for 2.6.25.
 228  */
 229 #if 0
 230 STATIC int
 231 xfs_file_readdir(
 232         struct file     *filp,
 233         void            *dirent,
 234         filldir_t       filldir)
 235 {
 236         struct inode    *inode = filp->f_path.dentry->d_inode;
 237         xfs_inode_t     *ip = XFS_I(inode);
 238         int             error;
 239         size_t          bufsize;
 240
 241         /*
 242          * The Linux API doesn't pass down the total size of the buffer
 243          * we read into down to the filesystem.  With the filldir concept
 244          * it's not needed for correct information, but the XFS dir2 leaf
 245          * code wants an estimate of the buffer size to calculate it's
 246          * readahead window and size the buffers used for mapping to
 247          * physical blocks.
 248          *
 249          * Try to give it an estimate that's good enough, maybe at some
 250          * point we can change the ->readdir prototype to include the
 251          * buffer size.
 252          */
 253         bufsize = (size_t)min_t(loff_t, PAGE_SIZE, inode->i_size);
 254
 255         error = xfs_readdir(ip, dirent, bufsize,
 256                                 (xfs_off_t *)&filp->f_pos, filldir);
 257         if (error)
 258                 return -error;
 259         return 0;
 260 }
 261 #else
 262
 263 struct hack_dirent {
 264         u64             ino;
 265         loff_t          offset;
 266         int             namlen;
 267         unsigned int    d_type;
 268         char            name[];
 269 };
 270
 271 struct hack_callback {
 272         char            *dirent;
 273         size_t          len;
 274         size_t          used;
 275 };
 276
 277 STATIC int
 278 xfs_hack_filldir(
 279         void            *__buf,
 280         const char      *name,
 281         int             namlen,
 282         loff_t          offset,
 283         u64             ino,
 284         unsigned int    d_type)
 285 {
 286         struct hack_callback *buf = __buf;
 287         struct hack_dirent *de = (struct hack_dirent *)(buf->dirent + buf->used);
 288         unsigned int reclen;
 289
 290         reclen = ALIGN(sizeof(struct hack_dirent) + namlen, sizeof(u64));
 291         if (buf->used + reclen > buf->len)
 292                 return -EINVAL;
 293
 294         de->namlen = namlen;
 295         de->offset = offset;
 296         de->ino = ino;
 297         de->d_type = d_type;
 298         memcpy(de->name, name, namlen);
 299         buf->used += reclen;
 300         return 0;
 301 }
 302
 303 STATIC int
 304 xfs_file_readdir(
 305         struct file     *filp,
 306         void            *dirent,
 307         filldir_t       filldir)
 308 {
 309         struct inode    *inode = filp->f_path.dentry->d_inode;
 310         xfs_inode_t     *ip = XFS_I(inode);
 311         struct hack_callback buf;
 312         struct hack_dirent *de;
 313         int             error;
 314         loff_t          size;
 315         int             eof = 0;
 316         xfs_off_t       start_offset, curr_offset, offset;
 317
 318         /*
 319          * Try fairly hard to get memory
 320          */
 321         buf.len = PAGE_CACHE_SIZE;
 322         do {
 323                 buf.dirent = kmalloc(buf.len, GFP_KERNEL);
 324                 if (buf.dirent)
 325                         break;
 326                 buf.len >>= 1;
 327         } while (buf.len >= 1024);
 328
 329         if (!buf.dirent)
 330                 return -ENOMEM;
 331
 332         curr_offset = filp->f_pos;
 333         if (curr_offset == 0x7fffffff)
 334                 offset = 0xffffffff;
 335         else
 336                 offset = filp->f_pos;
 337
 338         while (!eof) {
 339                 unsigned int reclen;
 340
 341                 start_offset = offset;
 342
 343                 buf.used = 0;
 344                 error = -xfs_readdir(ip, &buf, buf.len, &offset,
 345                                      xfs_hack_filldir);
 346                 if (error || offset == start_offset) {
 347                         size = 0;
 348                         break;
 349                 }
 350
 351                 size = buf.used;
 352                 de = (struct hack_dirent *)buf.dirent;
 353                 curr_offset = de->offset /* & 0x7fffffff */;
 354                 while (size > 0) {
 355                         if (filldir(dirent, de->name, de->namlen,
 356                                         curr_offset & 0x7fffffff,
 357                                         de->ino, de->d_type)) {
 358                                 goto done;
 359                         }
 360
 361                         reclen = ALIGN(sizeof(struct hack_dirent) + de->namlen,
 362                                        sizeof(u64));
 363                         size -= reclen;
 364                         de = (struct hack_dirent *)((char *)de + reclen);
 365                         curr_offset = de->offset /* & 0x7fffffff */;
 366                 }
 367         }
 368
 369  done:
 370         if (!error) {
 371                 if (size == 0)
 372                         filp->f_pos = offset & 0x7fffffff;
 373                 else if (de)
 374                         filp->f_pos = curr_offset;
 375         }
 376
 377         kfree(buf.dirent);
 378         return error;
 379 }
 380 #endif
 381
 382 STATIC int
 383 xfs_file_mmap(
 384         struct file     *filp,
 385         struct vm_area_struct *vma)
 386 {
 387         vma->vm_ops = &xfs_file_vm_ops;
 388         vma->vm_flags |= VM_CAN_NONLINEAR;
 389
 390 #ifdef CONFIG_XFS_DMAPI
 391         if (XFS_M(filp->f_path.dentry->d_inode->i_sb)->m_flags & XFS_MOUNT_DMAPI)
 392                 vma->vm_ops = &xfs_dmapi_file_vm_ops;
 393 #endif /* CONFIG_XFS_DMAPI */
 394
 395         file_accessed(filp);
 396         return 0;
 397 }
 398
 399 STATIC long
 400 xfs_file_ioctl(
 401         struct file     *filp,
 402         unsigned int    cmd,
 403         unsigned long   p)
 404 {
 405         int             error;
 406         struct inode    *inode = filp->f_path.dentry->d_inode;
 407
 408         error = xfs_ioctl(XFS_I(inode), filp, 0, cmd, (void __user *)p);
 409         xfs_iflags_set(XFS_I(inode), XFS_IMODIFIED);
 410
 411         /* NOTE:  some of the ioctl's return positive #'s as a
 412          *        byte count indicating success, such as
 413          *        readlink_by_handle.  So we don't "sign flip"
 414          *        like most other routines.  This means true
 415          *        errors need to be returned as a negative value.
 416          */
 417         return error;
 418 }
 419
 420 STATIC long
 421 xfs_file_ioctl_invis(
 422         struct file     *filp,
 423         unsigned int    cmd,
 424         unsigned long   p)
 425 {
 426         int             error;
 427         struct inode    *inode = filp->f_path.dentry->d_inode;
 428
 429         error = xfs_ioctl(XFS_I(inode), filp, IO_INVIS, cmd, (void __user *)p);
 430         xfs_iflags_set(XFS_I(inode), XFS_IMODIFIED);
 431
 432         /* NOTE:  some of the ioctl's return positive #'s as a
 433          *        byte count indicating success, such as
 434          *        readlink_by_handle.  So we don't "sign flip"
 435          *        like most other routines.  This means true
 436          *        errors need to be returned as a negative value.
 437          */
 438         return error;
 439 }
 440
 441 #ifdef CONFIG_XFS_DMAPI
 442 #ifdef HAVE_VMOP_MPROTECT
 443 STATIC int
 444 xfs_vm_mprotect(
 445         struct vm_area_struct *vma,
 446         unsigned int    newflags)
 447 {
 448         struct inode    *inode = vma->vm_file->f_path.dentry->d_inode;
 449         struct xfs_mount *mp = XFS_M(inode->i_sb);
 450         int             error = 0;
 451
 452         if (mp->m_flags & XFS_MOUNT_DMAPI) {
 453                 if ((vma->vm_flags & VM_MAYSHARE) &&
 454                     (newflags & VM_WRITE) && !(vma->vm_flags & VM_WRITE))
 455                         error = XFS_SEND_MMAP(mp, vma, VM_WRITE);
 456         }
 457         return error;
 458 }
 459 #endif /* HAVE_VMOP_MPROTECT */
 460 #endif /* CONFIG_XFS_DMAPI */
 461
 462 #ifdef HAVE_FOP_OPEN_EXEC
 463 /* If the user is attempting to execute a file that is offline then
 464  * we have to trigger a DMAPI READ event before the file is marked as busy
 465  * otherwise the invisible I/O will not be able to write to the file to bring
 466  * it back online.
 467  */
 468 STATIC int
 469 xfs_file_open_exec(
 470         struct inode    *inode)
 471 {
 472         struct xfs_mount *mp = XFS_M(inode->i_sb);
 473
 474         if (unlikely(mp->m_flags & XFS_MOUNT_DMAPI)) {
 475                 if (DM_EVENT_ENABLED(XFS_I(inode), DM_EVENT_READ)) {
 476                         bhv_vnode_t *vp = vn_from_inode(inode);
 477
 478                         return -XFS_SEND_DATA(mp, DM_EVENT_READ,
 479                                                 vp, 0, 0, 0, NULL);
 480                 }
 481         }
 482
 483         return 0;
 484 }
 485 #endif /* HAVE_FOP_OPEN_EXEC */
 486
 487 /*
 488  * mmap()d file has taken write protection fault and is being made
 489  * writable. We can set the page state up correctly for a writable
 490  * page, which means we can do correct delalloc accounting (ENOSPC
 491  * checking!) and unwritten extent mapping.
 492  */
 493 STATIC int
 494 xfs_vm_page_mkwrite(
 495         struct vm_area_struct   *vma,
 496         struct page             *page)
 497 {
 498         return block_page_mkwrite(vma, page, xfs_get_blocks);
 499 }
 500
 501 const struct file_operations xfs_file_operations = {
 502         .llseek         = generic_file_llseek,
 503         .read           = do_sync_read,
 504         .write          = do_sync_write,
 505         .aio_read       = xfs_file_aio_read,
 506         .aio_write      = xfs_file_aio_write,
 507         .splice_read    = xfs_file_splice_read,
 508         .splice_write   = xfs_file_splice_write,
 509         .unlocked_ioctl = xfs_file_ioctl,
 510 #ifdef CONFIG_COMPAT
 511         .compat_ioctl   = xfs_file_compat_ioctl,
 512 #endif
 513         .mmap           = xfs_file_mmap,
 514         .open           = xfs_file_open,
 515         .release        = xfs_file_release,
 516         .fsync          = xfs_file_fsync,
 517 #ifdef HAVE_FOP_OPEN_EXEC
 518         .open_exec      = xfs_file_open_exec,
 519 #endif
 520 };
 521
 522 const struct file_operations xfs_invis_file_operations = {
 523         .llseek         = generic_file_llseek,
 524         .read           = do_sync_read,
 525         .write          = do_sync_write,
 526         .aio_read       = xfs_file_aio_read_invis,
 527         .aio_write      = xfs_file_aio_write_invis,
 528         .splice_read    = xfs_file_splice_read_invis,
 529         .splice_write   = xfs_file_splice_write_invis,
 530         .unlocked_ioctl = xfs_file_ioctl_invis,
 531 #ifdef CONFIG_COMPAT
 532         .compat_ioctl   = xfs_file_compat_invis_ioctl,
 533 #endif
 534         .mmap           = xfs_file_mmap,
 535         .open           = xfs_file_open,
 536         .release        = xfs_file_release,
 537         .fsync          = xfs_file_fsync,
 538 };
 539
 540
 541 const struct file_operations xfs_dir_file_operations = {
 542         .read           = generic_read_dir,
 543         .readdir        = xfs_file_readdir,
 544         .unlocked_ioctl = xfs_file_ioctl,
 545 #ifdef CONFIG_COMPAT
 546         .compat_ioctl   = xfs_file_compat_ioctl,
 547 #endif
 548         .fsync          = xfs_file_fsync,
 549 };
 550
 551 static struct vm_operations_struct xfs_file_vm_ops = {
 552         .fault          = filemap_fault,
 553         .page_mkwrite   = xfs_vm_page_mkwrite,
 554 };
 555
 556 #ifdef CONFIG_XFS_DMAPI
 557 static struct vm_operations_struct xfs_dmapi_file_vm_ops = {
 558         .fault          = xfs_vm_fault,
 559         .page_mkwrite   = xfs_vm_page_mkwrite,
 560 #ifdef HAVE_VMOP_MPROTECT
 561         .mprotect       = xfs_vm_mprotect,
 562 #endif
 563 };
 564 #endif /* CONFIG_XFS_DMAPI */