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         int             namlen;
 265         loff_t          offset;
 266         u64             ino;
 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
 289         if (buf->used + sizeof(struct hack_dirent) + namlen > buf->len)
 290                 return -EINVAL;
 291
 292         de->namlen = namlen;
 293         de->offset = offset;
 294         de->ino = ino;
 295         de->d_type = d_type;
 296         memcpy(de->name, name, namlen);
 297         buf->used += sizeof(struct hack_dirent) + namlen;
 298         return 0;
 299 }
 300
 301 STATIC int
 302 xfs_file_readdir(
 303         struct file     *filp,
 304         void            *dirent,
 305         filldir_t       filldir)
 306 {
 307         struct inode    *inode = filp->f_path.dentry->d_inode;
 308         xfs_inode_t     *ip = XFS_I(inode);
 309         struct hack_callback buf;
 310         struct hack_dirent *de;
 311         int             error;
 312         loff_t          size;
 313         int             eof = 0;
 314         xfs_off_t       start_offset, curr_offset, offset;
 315
 316         /*
 317          * Try fairly hard to get memory
 318          */
 319         buf.len = PAGE_CACHE_SIZE;
 320         do {
 321                 buf.dirent = kmalloc(buf.len, GFP_KERNEL);
 322                 if (buf.dirent)
 323                         break;
 324                 buf.len >>= 1;
 325         } while (buf.len >= 1024);
 326
 327         if (!buf.dirent)
 328                 return -ENOMEM;
 329
 330         curr_offset = filp->f_pos;
 331         if (curr_offset == 0x7fffffff)
 332                 offset = 0xffffffff;
 333         else
 334                 offset = filp->f_pos;
 335
 336         while (!eof) {
 337                 int reclen;
 338                 start_offset = offset;
 339
 340                 buf.used = 0;
 341                 error = -xfs_readdir(ip, &buf, buf.len, &offset,
 342                                      xfs_hack_filldir);
 343                 if (error || offset == start_offset) {
 344                         size = 0;
 345                         break;
 346                 }
 347
 348                 size = buf.used;
 349                 de = (struct hack_dirent *)buf.dirent;
 350                 while (size > 0) {
 351                         if (filldir(dirent, de->name, de->namlen,
 352                                         curr_offset & 0x7fffffff,
 353                                         de->ino, de->d_type)) {
 354                                 goto done;
 355                         }
 356
 357                         reclen = sizeof(struct hack_dirent) + de->namlen;
 358                         size -= reclen;
 359                         de = (struct hack_dirent *)((char *)de + reclen);
 360                         curr_offset = de->offset /* & 0x7fffffff */;
 361                 }
 362         }
 363
 364  done:
 365         if (!error) {
 366                 if (size == 0)
 367                         filp->f_pos = offset & 0x7fffffff;
 368                 else if (de)
 369                         filp->f_pos = curr_offset;
 370         }
 371
 372         kfree(buf.dirent);
 373         return error;
 374 }
 375 #endif
 376
 377 STATIC int
 378 xfs_file_mmap(
 379         struct file     *filp,
 380         struct vm_area_struct *vma)
 381 {
 382         vma->vm_ops = &xfs_file_vm_ops;
 383         vma->vm_flags |= VM_CAN_NONLINEAR;
 384
 385 #ifdef CONFIG_XFS_DMAPI
 386         if (XFS_M(filp->f_path.dentry->d_inode->i_sb)->m_flags & XFS_MOUNT_DMAPI)
 387                 vma->vm_ops = &xfs_dmapi_file_vm_ops;
 388 #endif /* CONFIG_XFS_DMAPI */
 389
 390         file_accessed(filp);
 391         return 0;
 392 }
 393
 394 STATIC long
 395 xfs_file_ioctl(
 396         struct file     *filp,
 397         unsigned int    cmd,
 398         unsigned long   p)
 399 {
 400         int             error;
 401         struct inode    *inode = filp->f_path.dentry->d_inode;
 402
 403         error = xfs_ioctl(XFS_I(inode), filp, 0, cmd, (void __user *)p);
 404         xfs_iflags_set(XFS_I(inode), XFS_IMODIFIED);
 405
 406         /* NOTE:  some of the ioctl's return positive #'s as a
 407          *        byte count indicating success, such as
 408          *        readlink_by_handle.  So we don't "sign flip"
 409          *        like most other routines.  This means true
 410          *        errors need to be returned as a negative value.
 411          */
 412         return error;
 413 }
 414
 415 STATIC long
 416 xfs_file_ioctl_invis(
 417         struct file     *filp,
 418         unsigned int    cmd,
 419         unsigned long   p)
 420 {
 421         int             error;
 422         struct inode    *inode = filp->f_path.dentry->d_inode;
 423
 424         error = xfs_ioctl(XFS_I(inode), filp, IO_INVIS, cmd, (void __user *)p);
 425         xfs_iflags_set(XFS_I(inode), XFS_IMODIFIED);
 426
 427         /* NOTE:  some of the ioctl's return positive #'s as a
 428          *        byte count indicating success, such as
 429          *        readlink_by_handle.  So we don't "sign flip"
 430          *        like most other routines.  This means true
 431          *        errors need to be returned as a negative value.
 432          */
 433         return error;
 434 }
 435
 436 #ifdef CONFIG_XFS_DMAPI
 437 #ifdef HAVE_VMOP_MPROTECT
 438 STATIC int
 439 xfs_vm_mprotect(
 440         struct vm_area_struct *vma,
 441         unsigned int    newflags)
 442 {
 443         struct inode    *inode = vma->vm_file->f_path.dentry->d_inode;
 444         struct xfs_mount *mp = XFS_M(inode->i_sb);
 445         int             error = 0;
 446
 447         if (mp->m_flags & XFS_MOUNT_DMAPI) {
 448                 if ((vma->vm_flags & VM_MAYSHARE) &&
 449                     (newflags & VM_WRITE) && !(vma->vm_flags & VM_WRITE))
 450                         error = XFS_SEND_MMAP(mp, vma, VM_WRITE);
 451         }
 452         return error;
 453 }
 454 #endif /* HAVE_VMOP_MPROTECT */
 455 #endif /* CONFIG_XFS_DMAPI */
 456
 457 #ifdef HAVE_FOP_OPEN_EXEC
 458 /* If the user is attempting to execute a file that is offline then
 459  * we have to trigger a DMAPI READ event before the file is marked as busy
 460  * otherwise the invisible I/O will not be able to write to the file to bring
 461  * it back online.
 462  */
 463 STATIC int
 464 xfs_file_open_exec(
 465         struct inode    *inode)
 466 {
 467         struct xfs_mount *mp = XFS_M(inode->i_sb);
 468
 469         if (unlikely(mp->m_flags & XFS_MOUNT_DMAPI)) {
 470                 if (DM_EVENT_ENABLED(XFS_I(inode), DM_EVENT_READ)) {
 471                         bhv_vnode_t *vp = vn_from_inode(inode);
 472
 473                         return -XFS_SEND_DATA(mp, DM_EVENT_READ,
 474                                                 vp, 0, 0, 0, NULL);
 475                 }
 476         }
 477
 478         return 0;
 479 }
 480 #endif /* HAVE_FOP_OPEN_EXEC */
 481
 482 /*
 483  * mmap()d file has taken write protection fault and is being made
 484  * writable. We can set the page state up correctly for a writable
 485  * page, which means we can do correct delalloc accounting (ENOSPC
 486  * checking!) and unwritten extent mapping.
 487  */
 488 STATIC int
 489 xfs_vm_page_mkwrite(
 490         struct vm_area_struct   *vma,
 491         struct page             *page)
 492 {
 493         return block_page_mkwrite(vma, page, xfs_get_blocks);
 494 }
 495
 496 const struct file_operations xfs_file_operations = {
 497         .llseek         = generic_file_llseek,
 498         .read           = do_sync_read,
 499         .write          = do_sync_write,
 500         .aio_read       = xfs_file_aio_read,
 501         .aio_write      = xfs_file_aio_write,
 502         .splice_read    = xfs_file_splice_read,
 503         .splice_write   = xfs_file_splice_write,
 504         .unlocked_ioctl = xfs_file_ioctl,
 505 #ifdef CONFIG_COMPAT
 506         .compat_ioctl   = xfs_file_compat_ioctl,
 507 #endif
 508         .mmap           = xfs_file_mmap,
 509         .open           = xfs_file_open,
 510         .release        = xfs_file_release,
 511         .fsync          = xfs_file_fsync,
 512 #ifdef HAVE_FOP_OPEN_EXEC
 513         .open_exec      = xfs_file_open_exec,
 514 #endif
 515 };
 516
 517 const struct file_operations xfs_invis_file_operations = {
 518         .llseek         = generic_file_llseek,
 519         .read           = do_sync_read,
 520         .write          = do_sync_write,
 521         .aio_read       = xfs_file_aio_read_invis,
 522         .aio_write      = xfs_file_aio_write_invis,
 523         .splice_read    = xfs_file_splice_read_invis,
 524         .splice_write   = xfs_file_splice_write_invis,
 525         .unlocked_ioctl = xfs_file_ioctl_invis,
 526 #ifdef CONFIG_COMPAT
 527         .compat_ioctl   = xfs_file_compat_invis_ioctl,
 528 #endif
 529         .mmap           = xfs_file_mmap,
 530         .open           = xfs_file_open,
 531         .release        = xfs_file_release,
 532         .fsync          = xfs_file_fsync,
 533 };
 534
 535
 536 const struct file_operations xfs_dir_file_operations = {
 537         .read           = generic_read_dir,
 538         .readdir        = xfs_file_readdir,
 539         .unlocked_ioctl = xfs_file_ioctl,
 540 #ifdef CONFIG_COMPAT
 541         .compat_ioctl   = xfs_file_compat_ioctl,
 542 #endif
 543         .fsync          = xfs_file_fsync,
 544 };
 545
 546 static struct vm_operations_struct xfs_file_vm_ops = {
 547         .fault          = filemap_fault,
 548         .page_mkwrite   = xfs_vm_page_mkwrite,
 549 };
 550
 551 #ifdef CONFIG_XFS_DMAPI
 552 static struct vm_operations_struct xfs_dmapi_file_vm_ops = {
 553         .fault          = xfs_vm_fault,
 554         .page_mkwrite   = xfs_vm_page_mkwrite,
 555 #ifdef HAVE_VMOP_MPROTECT
 556         .mprotect       = xfs_vm_mprotect,
 557 #endif
 558 };
 559 #endif /* CONFIG_XFS_DMAPI */