Merge branch 'for-2.6.23' into merge
[sfrench/cifs-2.6.git] / 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
41 #include <linux/dcache.h>
42 #include <linux/smp_lock.h>
43
44 static struct vm_operations_struct xfs_file_vm_ops;
45 #ifdef CONFIG_XFS_DMAPI
46 static struct vm_operations_struct xfs_dmapi_file_vm_ops;
47 #endif
48
49 STATIC_INLINE ssize_t
50 __xfs_file_read(
51         struct kiocb            *iocb,
52         const struct iovec      *iov,
53         unsigned long           nr_segs,
54         int                     ioflags,
55         loff_t                  pos)
56 {
57         struct file             *file = iocb->ki_filp;
58         bhv_vnode_t             *vp = vn_from_inode(file->f_path.dentry->d_inode);
59
60         BUG_ON(iocb->ki_pos != pos);
61         if (unlikely(file->f_flags & O_DIRECT))
62                 ioflags |= IO_ISDIRECT;
63         return bhv_vop_read(vp, iocb, iov, nr_segs, &iocb->ki_pos,
64                                 ioflags, NULL);
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         struct inode    *inode = file->f_mapping->host;
97         bhv_vnode_t     *vp = vn_from_inode(inode);
98
99         BUG_ON(iocb->ki_pos != pos);
100         if (unlikely(file->f_flags & O_DIRECT))
101                 ioflags |= IO_ISDIRECT;
102         return bhv_vop_write(vp, iocb, iov, nr_segs, &iocb->ki_pos,
103                                 ioflags, NULL);
104 }
105
106 STATIC ssize_t
107 xfs_file_aio_write(
108         struct kiocb            *iocb,
109         const struct iovec      *iov,
110         unsigned long           nr_segs,
111         loff_t                  pos)
112 {
113         return __xfs_file_write(iocb, iov, nr_segs, IO_ISAIO, pos);
114 }
115
116 STATIC ssize_t
117 xfs_file_aio_write_invis(
118         struct kiocb            *iocb,
119         const struct iovec      *iov,
120         unsigned long           nr_segs,
121         loff_t                  pos)
122 {
123         return __xfs_file_write(iocb, iov, nr_segs, IO_ISAIO|IO_INVIS, pos);
124 }
125
126 STATIC ssize_t
127 xfs_file_splice_read(
128         struct file             *infilp,
129         loff_t                  *ppos,
130         struct pipe_inode_info  *pipe,
131         size_t                  len,
132         unsigned int            flags)
133 {
134         return bhv_vop_splice_read(vn_from_inode(infilp->f_path.dentry->d_inode),
135                                    infilp, ppos, pipe, len, flags, 0, NULL);
136 }
137
138 STATIC ssize_t
139 xfs_file_splice_read_invis(
140         struct file             *infilp,
141         loff_t                  *ppos,
142         struct pipe_inode_info  *pipe,
143         size_t                  len,
144         unsigned int            flags)
145 {
146         return bhv_vop_splice_read(vn_from_inode(infilp->f_path.dentry->d_inode),
147                                    infilp, ppos, pipe, len, flags, IO_INVIS,
148                                    NULL);
149 }
150
151 STATIC ssize_t
152 xfs_file_splice_write(
153         struct pipe_inode_info  *pipe,
154         struct file             *outfilp,
155         loff_t                  *ppos,
156         size_t                  len,
157         unsigned int            flags)
158 {
159         return bhv_vop_splice_write(vn_from_inode(outfilp->f_path.dentry->d_inode),
160                                     pipe, outfilp, ppos, len, flags, 0, NULL);
161 }
162
163 STATIC ssize_t
164 xfs_file_splice_write_invis(
165         struct pipe_inode_info  *pipe,
166         struct file             *outfilp,
167         loff_t                  *ppos,
168         size_t                  len,
169         unsigned int            flags)
170 {
171         return bhv_vop_splice_write(vn_from_inode(outfilp->f_path.dentry->d_inode),
172                                     pipe, outfilp, ppos, len, flags, IO_INVIS,
173                                     NULL);
174 }
175
176 STATIC int
177 xfs_file_open(
178         struct inode    *inode,
179         struct file     *filp)
180 {
181         if (!(filp->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
182                 return -EFBIG;
183         return -bhv_vop_open(vn_from_inode(inode), NULL);
184 }
185
186 STATIC int
187 xfs_file_close(
188         struct file     *filp,
189         fl_owner_t      id)
190 {
191         return -bhv_vop_close(vn_from_inode(filp->f_path.dentry->d_inode), 0,
192                                 file_count(filp) > 1 ? L_FALSE : L_TRUE, NULL);
193 }
194
195 STATIC int
196 xfs_file_release(
197         struct inode    *inode,
198         struct file     *filp)
199 {
200         bhv_vnode_t     *vp = vn_from_inode(inode);
201
202         if (vp)
203                 return -bhv_vop_release(vp);
204         return 0;
205 }
206
207 STATIC int
208 xfs_file_fsync(
209         struct file     *filp,
210         struct dentry   *dentry,
211         int             datasync)
212 {
213         bhv_vnode_t     *vp = vn_from_inode(dentry->d_inode);
214         int             flags = FSYNC_WAIT;
215
216         if (datasync)
217                 flags |= FSYNC_DATA;
218         if (VN_TRUNC(vp))
219                 VUNTRUNCATE(vp);
220         return -bhv_vop_fsync(vp, flags, NULL, (xfs_off_t)0, (xfs_off_t)-1);
221 }
222
223 #ifdef CONFIG_XFS_DMAPI
224 STATIC struct page *
225 xfs_vm_nopage(
226         struct vm_area_struct   *area,
227         unsigned long           address,
228         int                     *type)
229 {
230         struct inode    *inode = area->vm_file->f_path.dentry->d_inode;
231         bhv_vnode_t     *vp = vn_from_inode(inode);
232
233         ASSERT_ALWAYS(vp->v_vfsp->vfs_flag & VFS_DMI);
234         if (XFS_SEND_MMAP(XFS_VFSTOM(vp->v_vfsp), area, 0))
235                 return NULL;
236         return filemap_nopage(area, address, type);
237 }
238 #endif /* CONFIG_XFS_DMAPI */
239
240 STATIC int
241 xfs_file_readdir(
242         struct file     *filp,
243         void            *dirent,
244         filldir_t       filldir)
245 {
246         int             error = 0;
247         bhv_vnode_t     *vp = vn_from_inode(filp->f_path.dentry->d_inode);
248         uio_t           uio;
249         iovec_t         iov;
250         int             eof = 0;
251         caddr_t         read_buf;
252         int             namelen, size = 0;
253         size_t          rlen = PAGE_CACHE_SIZE;
254         xfs_off_t       start_offset, curr_offset;
255         xfs_dirent_t    *dbp = NULL;
256
257         /* Try fairly hard to get memory */
258         do {
259                 if ((read_buf = kmalloc(rlen, GFP_KERNEL)))
260                         break;
261                 rlen >>= 1;
262         } while (rlen >= 1024);
263
264         if (read_buf == NULL)
265                 return -ENOMEM;
266
267         uio.uio_iov = &iov;
268         uio.uio_segflg = UIO_SYSSPACE;
269         curr_offset = filp->f_pos;
270         if (filp->f_pos != 0x7fffffff)
271                 uio.uio_offset = filp->f_pos;
272         else
273                 uio.uio_offset = 0xffffffff;
274
275         while (!eof) {
276                 uio.uio_resid = iov.iov_len = rlen;
277                 iov.iov_base = read_buf;
278                 uio.uio_iovcnt = 1;
279
280                 start_offset = uio.uio_offset;
281
282                 error = bhv_vop_readdir(vp, &uio, NULL, &eof);
283                 if ((uio.uio_offset == start_offset) || error) {
284                         size = 0;
285                         break;
286                 }
287
288                 size = rlen - uio.uio_resid;
289                 dbp = (xfs_dirent_t *)read_buf;
290                 while (size > 0) {
291                         namelen = strlen(dbp->d_name);
292
293                         if (filldir(dirent, dbp->d_name, namelen,
294                                         (loff_t) curr_offset & 0x7fffffff,
295                                         (ino_t) dbp->d_ino,
296                                         DT_UNKNOWN)) {
297                                 goto done;
298                         }
299                         size -= dbp->d_reclen;
300                         curr_offset = (loff_t)dbp->d_off /* & 0x7fffffff */;
301                         dbp = (xfs_dirent_t *)((char *)dbp + dbp->d_reclen);
302                 }
303         }
304 done:
305         if (!error) {
306                 if (size == 0)
307                         filp->f_pos = uio.uio_offset & 0x7fffffff;
308                 else if (dbp)
309                         filp->f_pos = curr_offset;
310         }
311
312         kfree(read_buf);
313         return -error;
314 }
315
316 STATIC int
317 xfs_file_mmap(
318         struct file     *filp,
319         struct vm_area_struct *vma)
320 {
321         vma->vm_ops = &xfs_file_vm_ops;
322
323 #ifdef CONFIG_XFS_DMAPI
324         if (vn_from_inode(filp->f_path.dentry->d_inode)->v_vfsp->vfs_flag & VFS_DMI)
325                 vma->vm_ops = &xfs_dmapi_file_vm_ops;
326 #endif /* CONFIG_XFS_DMAPI */
327
328         file_accessed(filp);
329         return 0;
330 }
331
332 STATIC long
333 xfs_file_ioctl(
334         struct file     *filp,
335         unsigned int    cmd,
336         unsigned long   p)
337 {
338         int             error;
339         struct inode    *inode = filp->f_path.dentry->d_inode;
340         bhv_vnode_t     *vp = vn_from_inode(inode);
341
342         error = bhv_vop_ioctl(vp, inode, filp, 0, cmd, (void __user *)p);
343         VMODIFY(vp);
344
345         /* NOTE:  some of the ioctl's return positive #'s as a
346          *        byte count indicating success, such as
347          *        readlink_by_handle.  So we don't "sign flip"
348          *        like most other routines.  This means true
349          *        errors need to be returned as a negative value.
350          */
351         return error;
352 }
353
354 STATIC long
355 xfs_file_ioctl_invis(
356         struct file     *filp,
357         unsigned int    cmd,
358         unsigned long   p)
359 {
360         int             error;
361         struct inode    *inode = filp->f_path.dentry->d_inode;
362         bhv_vnode_t     *vp = vn_from_inode(inode);
363
364         error = bhv_vop_ioctl(vp, inode, filp, IO_INVIS, cmd, (void __user *)p);
365         VMODIFY(vp);
366
367         /* NOTE:  some of the ioctl's return positive #'s as a
368          *        byte count indicating success, such as
369          *        readlink_by_handle.  So we don't "sign flip"
370          *        like most other routines.  This means true
371          *        errors need to be returned as a negative value.
372          */
373         return error;
374 }
375
376 #ifdef CONFIG_XFS_DMAPI
377 #ifdef HAVE_VMOP_MPROTECT
378 STATIC int
379 xfs_vm_mprotect(
380         struct vm_area_struct *vma,
381         unsigned int    newflags)
382 {
383         bhv_vnode_t     *vp = vn_from_inode(vma->vm_file->f_path.dentry->d_inode);
384         int             error = 0;
385
386         if (vp->v_vfsp->vfs_flag & VFS_DMI) {
387                 if ((vma->vm_flags & VM_MAYSHARE) &&
388                     (newflags & VM_WRITE) && !(vma->vm_flags & VM_WRITE)) {
389                         xfs_mount_t     *mp = XFS_VFSTOM(vp->v_vfsp);
390
391                         error = XFS_SEND_MMAP(mp, vma, VM_WRITE);
392                     }
393         }
394         return error;
395 }
396 #endif /* HAVE_VMOP_MPROTECT */
397 #endif /* CONFIG_XFS_DMAPI */
398
399 #ifdef HAVE_FOP_OPEN_EXEC
400 /* If the user is attempting to execute a file that is offline then
401  * we have to trigger a DMAPI READ event before the file is marked as busy
402  * otherwise the invisible I/O will not be able to write to the file to bring
403  * it back online.
404  */
405 STATIC int
406 xfs_file_open_exec(
407         struct inode    *inode)
408 {
409         bhv_vnode_t     *vp = vn_from_inode(inode);
410
411         if (unlikely(vp->v_vfsp->vfs_flag & VFS_DMI)) {
412                 xfs_mount_t     *mp = XFS_VFSTOM(vp->v_vfsp);
413                 xfs_inode_t     *ip = xfs_vtoi(vp);
414
415                 if (!ip)
416                         return -EINVAL;
417                 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ))
418                         return -XFS_SEND_DATA(mp, DM_EVENT_READ, vp,
419                                                0, 0, 0, NULL);
420         }
421         return 0;
422 }
423 #endif /* HAVE_FOP_OPEN_EXEC */
424
425 const struct file_operations xfs_file_operations = {
426         .llseek         = generic_file_llseek,
427         .read           = do_sync_read,
428         .write          = do_sync_write,
429         .aio_read       = xfs_file_aio_read,
430         .aio_write      = xfs_file_aio_write,
431         .splice_read    = xfs_file_splice_read,
432         .splice_write   = xfs_file_splice_write,
433         .unlocked_ioctl = xfs_file_ioctl,
434 #ifdef CONFIG_COMPAT
435         .compat_ioctl   = xfs_file_compat_ioctl,
436 #endif
437         .mmap           = xfs_file_mmap,
438         .open           = xfs_file_open,
439         .flush          = xfs_file_close,
440         .release        = xfs_file_release,
441         .fsync          = xfs_file_fsync,
442 #ifdef HAVE_FOP_OPEN_EXEC
443         .open_exec      = xfs_file_open_exec,
444 #endif
445 };
446
447 const struct file_operations xfs_invis_file_operations = {
448         .llseek         = generic_file_llseek,
449         .read           = do_sync_read,
450         .write          = do_sync_write,
451         .aio_read       = xfs_file_aio_read_invis,
452         .aio_write      = xfs_file_aio_write_invis,
453         .splice_read    = xfs_file_splice_read_invis,
454         .splice_write   = xfs_file_splice_write_invis,
455         .unlocked_ioctl = xfs_file_ioctl_invis,
456 #ifdef CONFIG_COMPAT
457         .compat_ioctl   = xfs_file_compat_invis_ioctl,
458 #endif
459         .mmap           = xfs_file_mmap,
460         .open           = xfs_file_open,
461         .flush          = xfs_file_close,
462         .release        = xfs_file_release,
463         .fsync          = xfs_file_fsync,
464 };
465
466
467 const struct file_operations xfs_dir_file_operations = {
468         .read           = generic_read_dir,
469         .readdir        = xfs_file_readdir,
470         .unlocked_ioctl = xfs_file_ioctl,
471 #ifdef CONFIG_COMPAT
472         .compat_ioctl   = xfs_file_compat_ioctl,
473 #endif
474         .fsync          = xfs_file_fsync,
475 };
476
477 static struct vm_operations_struct xfs_file_vm_ops = {
478         .nopage         = filemap_nopage,
479         .populate       = filemap_populate,
480 };
481
482 #ifdef CONFIG_XFS_DMAPI
483 static struct vm_operations_struct xfs_dmapi_file_vm_ops = {
484         .nopage         = xfs_vm_nopage,
485         .populate       = filemap_populate,
486 #ifdef HAVE_VMOP_MPROTECT
487         .mprotect       = xfs_vm_mprotect,
488 #endif
489 };
490 #endif /* CONFIG_XFS_DMAPI */