Merge branch 'for-linus' of git://oss.sgi.com:8090/xfs/xfs-2.6
[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_release(
188         struct inode    *inode,
189         struct file     *filp)
190 {
191         bhv_vnode_t     *vp = vn_from_inode(inode);
192
193         if (vp)
194                 return -bhv_vop_release(vp);
195         return 0;
196 }
197
198 STATIC int
199 xfs_file_fsync(
200         struct file     *filp,
201         struct dentry   *dentry,
202         int             datasync)
203 {
204         bhv_vnode_t     *vp = vn_from_inode(dentry->d_inode);
205         int             flags = FSYNC_WAIT;
206
207         if (datasync)
208                 flags |= FSYNC_DATA;
209         if (VN_TRUNC(vp))
210                 VUNTRUNCATE(vp);
211         return -bhv_vop_fsync(vp, flags, NULL, (xfs_off_t)0, (xfs_off_t)-1);
212 }
213
214 #ifdef CONFIG_XFS_DMAPI
215 STATIC int
216 xfs_vm_fault(
217         struct vm_area_struct   *vma,
218         struct vm_fault *vmf)
219 {
220         struct inode    *inode = vma->vm_file->f_path.dentry->d_inode;
221         bhv_vnode_t     *vp = vn_from_inode(inode);
222
223         ASSERT_ALWAYS(vp->v_vfsp->vfs_flag & VFS_DMI);
224         if (XFS_SEND_MMAP(XFS_VFSTOM(vp->v_vfsp), vma, 0))
225                 return VM_FAULT_SIGBUS;
226         return filemap_fault(vma, vmf);
227 }
228 #endif /* CONFIG_XFS_DMAPI */
229
230 STATIC int
231 xfs_file_readdir(
232         struct file     *filp,
233         void            *dirent,
234         filldir_t       filldir)
235 {
236         int             error = 0;
237         bhv_vnode_t     *vp = vn_from_inode(filp->f_path.dentry->d_inode);
238         uio_t           uio;
239         iovec_t         iov;
240         int             eof = 0;
241         caddr_t         read_buf;
242         int             namelen, size = 0;
243         size_t          rlen = PAGE_CACHE_SIZE;
244         xfs_off_t       start_offset, curr_offset;
245         xfs_dirent_t    *dbp = NULL;
246
247         /* Try fairly hard to get memory */
248         do {
249                 if ((read_buf = kmalloc(rlen, GFP_KERNEL)))
250                         break;
251                 rlen >>= 1;
252         } while (rlen >= 1024);
253
254         if (read_buf == NULL)
255                 return -ENOMEM;
256
257         uio.uio_iov = &iov;
258         uio.uio_segflg = UIO_SYSSPACE;
259         curr_offset = filp->f_pos;
260         if (filp->f_pos != 0x7fffffff)
261                 uio.uio_offset = filp->f_pos;
262         else
263                 uio.uio_offset = 0xffffffff;
264
265         while (!eof) {
266                 uio.uio_resid = iov.iov_len = rlen;
267                 iov.iov_base = read_buf;
268                 uio.uio_iovcnt = 1;
269
270                 start_offset = uio.uio_offset;
271
272                 error = bhv_vop_readdir(vp, &uio, NULL, &eof);
273                 if ((uio.uio_offset == start_offset) || error) {
274                         size = 0;
275                         break;
276                 }
277
278                 size = rlen - uio.uio_resid;
279                 dbp = (xfs_dirent_t *)read_buf;
280                 while (size > 0) {
281                         namelen = strlen(dbp->d_name);
282
283                         if (filldir(dirent, dbp->d_name, namelen,
284                                         (loff_t) curr_offset & 0x7fffffff,
285                                         (ino_t) dbp->d_ino,
286                                         DT_UNKNOWN)) {
287                                 goto done;
288                         }
289                         size -= dbp->d_reclen;
290                         curr_offset = (loff_t)dbp->d_off /* & 0x7fffffff */;
291                         dbp = (xfs_dirent_t *)((char *)dbp + dbp->d_reclen);
292                 }
293         }
294 done:
295         if (!error) {
296                 if (size == 0)
297                         filp->f_pos = uio.uio_offset & 0x7fffffff;
298                 else if (dbp)
299                         filp->f_pos = curr_offset;
300         }
301
302         kfree(read_buf);
303         return -error;
304 }
305
306 STATIC int
307 xfs_file_mmap(
308         struct file     *filp,
309         struct vm_area_struct *vma)
310 {
311         vma->vm_ops = &xfs_file_vm_ops;
312         vma->vm_flags |= VM_CAN_NONLINEAR;
313
314 #ifdef CONFIG_XFS_DMAPI
315         if (vn_from_inode(filp->f_path.dentry->d_inode)->v_vfsp->vfs_flag & VFS_DMI)
316                 vma->vm_ops = &xfs_dmapi_file_vm_ops;
317 #endif /* CONFIG_XFS_DMAPI */
318
319         file_accessed(filp);
320         return 0;
321 }
322
323 STATIC long
324 xfs_file_ioctl(
325         struct file     *filp,
326         unsigned int    cmd,
327         unsigned long   p)
328 {
329         int             error;
330         struct inode    *inode = filp->f_path.dentry->d_inode;
331         bhv_vnode_t     *vp = vn_from_inode(inode);
332
333         error = bhv_vop_ioctl(vp, inode, filp, 0, cmd, (void __user *)p);
334         VMODIFY(vp);
335
336         /* NOTE:  some of the ioctl's return positive #'s as a
337          *        byte count indicating success, such as
338          *        readlink_by_handle.  So we don't "sign flip"
339          *        like most other routines.  This means true
340          *        errors need to be returned as a negative value.
341          */
342         return error;
343 }
344
345 STATIC long
346 xfs_file_ioctl_invis(
347         struct file     *filp,
348         unsigned int    cmd,
349         unsigned long   p)
350 {
351         int             error;
352         struct inode    *inode = filp->f_path.dentry->d_inode;
353         bhv_vnode_t     *vp = vn_from_inode(inode);
354
355         error = bhv_vop_ioctl(vp, inode, filp, IO_INVIS, cmd, (void __user *)p);
356         VMODIFY(vp);
357
358         /* NOTE:  some of the ioctl's return positive #'s as a
359          *        byte count indicating success, such as
360          *        readlink_by_handle.  So we don't "sign flip"
361          *        like most other routines.  This means true
362          *        errors need to be returned as a negative value.
363          */
364         return error;
365 }
366
367 #ifdef CONFIG_XFS_DMAPI
368 #ifdef HAVE_VMOP_MPROTECT
369 STATIC int
370 xfs_vm_mprotect(
371         struct vm_area_struct *vma,
372         unsigned int    newflags)
373 {
374         bhv_vnode_t     *vp = vn_from_inode(vma->vm_file->f_path.dentry->d_inode);
375         int             error = 0;
376
377         if (vp->v_vfsp->vfs_flag & VFS_DMI) {
378                 if ((vma->vm_flags & VM_MAYSHARE) &&
379                     (newflags & VM_WRITE) && !(vma->vm_flags & VM_WRITE)) {
380                         xfs_mount_t     *mp = XFS_VFSTOM(vp->v_vfsp);
381
382                         error = XFS_SEND_MMAP(mp, vma, VM_WRITE);
383                     }
384         }
385         return error;
386 }
387 #endif /* HAVE_VMOP_MPROTECT */
388 #endif /* CONFIG_XFS_DMAPI */
389
390 #ifdef HAVE_FOP_OPEN_EXEC
391 /* If the user is attempting to execute a file that is offline then
392  * we have to trigger a DMAPI READ event before the file is marked as busy
393  * otherwise the invisible I/O will not be able to write to the file to bring
394  * it back online.
395  */
396 STATIC int
397 xfs_file_open_exec(
398         struct inode    *inode)
399 {
400         bhv_vnode_t     *vp = vn_from_inode(inode);
401
402         if (unlikely(vp->v_vfsp->vfs_flag & VFS_DMI)) {
403                 xfs_mount_t     *mp = XFS_VFSTOM(vp->v_vfsp);
404                 xfs_inode_t     *ip = xfs_vtoi(vp);
405
406                 if (!ip)
407                         return -EINVAL;
408                 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ))
409                         return -XFS_SEND_DATA(mp, DM_EVENT_READ, vp,
410                                                0, 0, 0, NULL);
411         }
412         return 0;
413 }
414 #endif /* HAVE_FOP_OPEN_EXEC */
415
416 /*
417  * mmap()d file has taken write protection fault and is being made
418  * writable. We can set the page state up correctly for a writable
419  * page, which means we can do correct delalloc accounting (ENOSPC
420  * checking!) and unwritten extent mapping.
421  */
422 STATIC int
423 xfs_vm_page_mkwrite(
424         struct vm_area_struct   *vma,
425         struct page             *page)
426 {
427         return block_page_mkwrite(vma, page, xfs_get_blocks);
428 }
429
430 const struct file_operations xfs_file_operations = {
431         .llseek         = generic_file_llseek,
432         .read           = do_sync_read,
433         .write          = do_sync_write,
434         .aio_read       = xfs_file_aio_read,
435         .aio_write      = xfs_file_aio_write,
436         .splice_read    = xfs_file_splice_read,
437         .splice_write   = xfs_file_splice_write,
438         .unlocked_ioctl = xfs_file_ioctl,
439 #ifdef CONFIG_COMPAT
440         .compat_ioctl   = xfs_file_compat_ioctl,
441 #endif
442         .mmap           = xfs_file_mmap,
443         .open           = xfs_file_open,
444         .release        = xfs_file_release,
445         .fsync          = xfs_file_fsync,
446 #ifdef HAVE_FOP_OPEN_EXEC
447         .open_exec      = xfs_file_open_exec,
448 #endif
449 };
450
451 const struct file_operations xfs_invis_file_operations = {
452         .llseek         = generic_file_llseek,
453         .read           = do_sync_read,
454         .write          = do_sync_write,
455         .aio_read       = xfs_file_aio_read_invis,
456         .aio_write      = xfs_file_aio_write_invis,
457         .splice_read    = xfs_file_splice_read_invis,
458         .splice_write   = xfs_file_splice_write_invis,
459         .unlocked_ioctl = xfs_file_ioctl_invis,
460 #ifdef CONFIG_COMPAT
461         .compat_ioctl   = xfs_file_compat_invis_ioctl,
462 #endif
463         .mmap           = xfs_file_mmap,
464         .open           = xfs_file_open,
465         .release        = xfs_file_release,
466         .fsync          = xfs_file_fsync,
467 };
468
469
470 const struct file_operations xfs_dir_file_operations = {
471         .read           = generic_read_dir,
472         .readdir        = xfs_file_readdir,
473         .unlocked_ioctl = xfs_file_ioctl,
474 #ifdef CONFIG_COMPAT
475         .compat_ioctl   = xfs_file_compat_ioctl,
476 #endif
477         .fsync          = xfs_file_fsync,
478 };
479
480 static struct vm_operations_struct xfs_file_vm_ops = {
481         .fault          = filemap_fault,
482         .page_mkwrite   = xfs_vm_page_mkwrite,
483 };
484
485 #ifdef CONFIG_XFS_DMAPI
486 static struct vm_operations_struct xfs_dmapi_file_vm_ops = {
487         .fault          = xfs_vm_fault,
488         .page_mkwrite   = xfs_vm_page_mkwrite,
489 #ifdef HAVE_VMOP_MPROTECT
490         .mprotect       = xfs_vm_mprotect,
491 #endif
492 };
493 #endif /* CONFIG_XFS_DMAPI */