]> git.proxmox.com Git - mirror_spl-debian.git/blame - module/spl/spl-vnode.c
Imported Upstream version 0.6.4.2
[mirror_spl-debian.git] / module / spl / spl-vnode.c
CommitLineData
716154c5
BB
1/*****************************************************************************\
2 * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC.
3 * Copyright (C) 2007 The Regents of the University of California.
4 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
5 * Written by Brian Behlendorf <behlendorf1@llnl.gov>.
715f6251 6 * UCRL-CODE-235197
7 *
716154c5 8 * This file is part of the SPL, Solaris Porting Layer.
3d6af2dd 9 * For details, see <http://zfsonlinux.org/>.
716154c5
BB
10 *
11 * The SPL is free software; you can redistribute it and/or modify it
12 * under the terms of the GNU General Public License as published by the
13 * Free Software Foundation; either version 2 of the License, or (at your
14 * option) any later version.
715f6251 15 *
716154c5 16 * The SPL is distributed in the hope that it will be useful, but WITHOUT
715f6251 17 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
18 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
19 * for more details.
20 *
21 * You should have received a copy of the GNU General Public License along
716154c5
BB
22 * with the SPL. If not, see <http://www.gnu.org/licenses/>.
23 *****************************************************************************
24 * Solaris Porting Layer (SPL) Vnode Implementation.
25\*****************************************************************************/
715f6251 26
80093b6f 27#include <sys/cred.h>
4b171585 28#include <sys/vnode.h>
10946b02 29#include <sys/kmem_cache.h>
bbdc6ae4 30#include <linux/falloc.h>
10946b02 31#include <linux/file_compat.h>
937879f1 32
51a727e9 33vnode_t *rootdir = (vnode_t *)0xabcd1234;
4b171585 34EXPORT_SYMBOL(rootdir);
35
7afde631 36static spl_kmem_cache_t *vn_cache;
37static spl_kmem_cache_t *vn_file_cache;
e4f1d29f 38
83c623aa 39static DEFINE_SPINLOCK(vn_file_lock);
e4f1d29f 40static LIST_HEAD(vn_file_list);
af828292 41
4295b530
BB
42vtype_t
43vn_mode_to_vtype(mode_t mode)
4b171585 44{
45 if (S_ISREG(mode))
46 return VREG;
47
48 if (S_ISDIR(mode))
49 return VDIR;
50
51 if (S_ISCHR(mode))
52 return VCHR;
53
54 if (S_ISBLK(mode))
55 return VBLK;
56
57 if (S_ISFIFO(mode))
58 return VFIFO;
59
60 if (S_ISLNK(mode))
61 return VLNK;
62
63 if (S_ISSOCK(mode))
64 return VSOCK;
65
66 if (S_ISCHR(mode))
67 return VCHR;
68
69 return VNON;
4295b530
BB
70} /* vn_mode_to_vtype() */
71EXPORT_SYMBOL(vn_mode_to_vtype);
72
73mode_t
74vn_vtype_to_mode(vtype_t vtype)
75{
76 if (vtype == VREG)
77 return S_IFREG;
78
79 if (vtype == VDIR)
80 return S_IFDIR;
81
82 if (vtype == VCHR)
83 return S_IFCHR;
84
85 if (vtype == VBLK)
86 return S_IFBLK;
87
88 if (vtype == VFIFO)
89 return S_IFIFO;
90
91 if (vtype == VLNK)
92 return S_IFLNK;
93
94 if (vtype == VSOCK)
95 return S_IFSOCK;
96
97 return VNON;
98} /* vn_vtype_to_mode() */
99EXPORT_SYMBOL(vn_vtype_to_mode);
4b171585 100
af828292 101vnode_t *
102vn_alloc(int flag)
103{
104 vnode_t *vp;
105
106 vp = kmem_cache_alloc(vn_cache, flag);
af828292 107 if (vp != NULL) {
e4f1d29f 108 vp->v_file = NULL;
af828292 109 vp->v_type = 0;
110 }
111
10946b02 112 return (vp);
af828292 113} /* vn_alloc() */
114EXPORT_SYMBOL(vn_alloc);
115
116void
117vn_free(vnode_t *vp)
118{
119 kmem_cache_free(vn_cache, vp);
120} /* vn_free() */
121EXPORT_SYMBOL(vn_free);
122
0b3cf046 123int
af828292 124vn_open(const char *path, uio_seg_t seg, int flags, int mode,
4b171585 125 vnode_t **vpp, int x1, void *x2)
0b3cf046 126{
f7e8739c
RC
127 struct file *fp;
128 struct kstat stat;
129 int rc, saved_umask = 0;
4be55565 130 gfp_t saved_gfp;
0b3cf046 131 vnode_t *vp;
0b3cf046 132
937879f1 133 ASSERT(flags & (FWRITE | FREAD));
134 ASSERT(seg == UIO_SYSSPACE);
135 ASSERT(vpp);
4b171585 136 *vpp = NULL;
137
138 if (!(flags & FCREAT) && (flags & FWRITE))
139 flags |= FEXCL;
140
728b9dd8 141 /* Note for filp_open() the two low bits must be remapped to mean:
142 * 01 - read-only -> 00 read-only
143 * 10 - write-only -> 01 write-only
144 * 11 - read-write -> 10 read-write
145 */
146 flags--;
0b3cf046 147
148 if (flags & FCREAT)
4b171585 149 saved_umask = xchg(&current->fs->umask, 0);
0b3cf046 150
f7e8739c 151 fp = filp_open(path, flags, mode);
0b3cf046 152
153 if (flags & FCREAT)
4b171585 154 (void)xchg(&current->fs->umask, saved_umask);
0b3cf046 155
f7e8739c 156 if (IS_ERR(fp))
10946b02 157 return (-PTR_ERR(fp));
0b3cf046 158
2a305c34
RY
159#ifdef HAVE_2ARGS_VFS_GETATTR
160 rc = vfs_getattr(&fp->f_path, &stat);
161#else
bc90df66 162 rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
2a305c34 163#endif
4b171585 164 if (rc) {
165 filp_close(fp, 0);
10946b02 166 return (-rc);
0b3cf046 167 }
168
af828292 169 vp = vn_alloc(KM_SLEEP);
4b171585 170 if (!vp) {
171 filp_close(fp, 0);
10946b02 172 return (ENOMEM);
4b171585 173 }
0b3cf046 174
4be55565
LW
175 saved_gfp = mapping_gfp_mask(fp->f_mapping);
176 mapping_set_gfp_mask(fp->f_mapping, saved_gfp & ~(__GFP_IO|__GFP_FS));
177
e4f1d29f 178 mutex_enter(&vp->v_lock);
4295b530 179 vp->v_type = vn_mode_to_vtype(stat.mode);
e4f1d29f 180 vp->v_file = fp;
4be55565 181 vp->v_gfp_mask = saved_gfp;
4b171585 182 *vpp = vp;
e4f1d29f 183 mutex_exit(&vp->v_lock);
0b3cf046 184
10946b02 185 return (0);
4b171585 186} /* vn_open() */
187EXPORT_SYMBOL(vn_open);
0b3cf046 188
0b3cf046 189int
af828292 190vn_openat(const char *path, uio_seg_t seg, int flags, int mode,
4b171585 191 vnode_t **vpp, int x1, void *x2, vnode_t *vp, int fd)
0b3cf046 192{
4b171585 193 char *realpath;
12018327 194 int len, rc;
0b3cf046 195
937879f1 196 ASSERT(vp == rootdir);
0b3cf046 197
12018327 198 len = strlen(path) + 2;
10946b02 199 realpath = kmalloc(len, kmem_flags_convert(KM_SLEEP));
4b171585 200 if (!realpath)
10946b02 201 return (ENOMEM);
0b3cf046 202
12018327 203 (void)snprintf(realpath, len, "/%s", path);
4b171585 204 rc = vn_open(realpath, seg, flags, mode, vpp, x1, x2);
4b171585 205 kfree(realpath);
206
10946b02 207 return (rc);
4b171585 208} /* vn_openat() */
209EXPORT_SYMBOL(vn_openat);
0b3cf046 210
0b3cf046 211int
4b171585 212vn_rdwr(uio_rw_t uio, vnode_t *vp, void *addr, ssize_t len, offset_t off,
663e02a1 213 uio_seg_t seg, int ioflag, rlim64_t x2, void *x3, ssize_t *residp)
0b3cf046 214{
4b171585 215 loff_t offset;
216 mm_segment_t saved_fs;
217 struct file *fp;
218 int rc;
219
937879f1 220 ASSERT(uio == UIO_WRITE || uio == UIO_READ);
221 ASSERT(vp);
222 ASSERT(vp->v_file);
223 ASSERT(seg == UIO_SYSSPACE);
663e02a1 224 ASSERT((ioflag & ~FAPPEND) == 0);
937879f1 225 ASSERT(x2 == RLIM64_INFINITY);
4b171585 226
e4f1d29f 227 fp = vp->v_file;
4b171585 228
663e02a1
RC
229 offset = off;
230 if (ioflag & FAPPEND)
231 offset = fp->f_pos;
232
4b171585 233 /* Writable user data segment must be briefly increased for this
234 * process so we can use the user space read call paths to write
235 * in to memory allocated by the kernel. */
236 saved_fs = get_fs();
237 set_fs(get_ds());
238
239 if (uio & UIO_WRITE)
240 rc = vfs_write(fp, addr, len, &offset);
241 else
242 rc = vfs_read(fp, addr, len, &offset);
243
244 set_fs(saved_fs);
f3989ed3 245 fp->f_pos = offset;
4b171585 246
247 if (rc < 0)
10946b02 248 return (-rc);
0b3cf046 249
4b171585 250 if (residp) {
251 *residp = len - rc;
0b3cf046 252 } else {
4b171585 253 if (rc != len)
10946b02 254 return (EIO);
0b3cf046 255 }
256
10946b02 257 return (0);
4b171585 258} /* vn_rdwr() */
259EXPORT_SYMBOL(vn_rdwr);
260
261int
2f5d55aa 262vn_close(vnode_t *vp, int flags, int x1, int x2, void *x3, void *x4)
4b171585 263{
264 int rc;
265
937879f1 266 ASSERT(vp);
267 ASSERT(vp->v_file);
4b171585 268
4be55565 269 mapping_set_gfp_mask(vp->v_file->f_mapping, vp->v_gfp_mask);
97735c39
BB
270 rc = filp_close(vp->v_file, 0);
271 vn_free(vp);
4b171585 272
10946b02 273 return (-rc);
4b171585 274} /* vn_close() */
275EXPORT_SYMBOL(vn_close);
276
97735c39
BB
277/* vn_seek() does not actually seek it only performs bounds checking on the
278 * proposed seek. We perform minimal checking and allow vn_rdwr() to catch
279 * anything more serious. */
280int
47995fa6 281vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
97735c39
BB
282{
283 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
284}
285EXPORT_SYMBOL(vn_seek);
286
10946b02
AX
287/*
288 * spl_basename() takes a NULL-terminated string s as input containing a path.
289 * It returns a char pointer to a string and a length that describe the
290 * basename of the path. If the basename is not "." or "/", it will be an index
291 * into the string. While the string should be NULL terminated, the section
292 * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
293 * CC0. Anyone wishing to reuse it in another codebase may pick either license.
294 */
295static void
296spl_basename(const char *s, const char **str, int *len)
297{
298 size_t i, end;
299
300 ASSERT(str);
301 ASSERT(len);
302
303 if (!s || !*s) {
304 *str = ".";
305 *len = 1;
306 return;
307 }
308
309 i = strlen(s) - 1;
310
311 while (i && s[i--] == '/');
312
313 if (i == 0) {
314 *str = "/";
315 *len = 1;
316 return;
317 }
318
319 end = i;
320
321 for (end = i; i; i--) {
322 if (s[i] == '/') {
323 *str = &s[i+1];
324 *len = end - i + 1;
325 return;
326 }
327 }
328
329 *str = s;
330 *len = end + 1;
331}
332
333static struct dentry *
334spl_kern_path_locked(const char *name, struct path *path)
335{
336 struct path parent;
337 struct dentry *dentry;
338 const char *basename;
339 int len;
340 int rc;
341
342 ASSERT(name);
343 ASSERT(path);
344
345 spl_basename(name, &basename, &len);
346
347 /* We do not accept "." or ".." */
348 if (len <= 2 && basename[0] == '.')
349 if (len == 1 || basename[1] == '.')
350 return (ERR_PTR(-EACCES));
351
352 rc = kern_path(name, LOOKUP_PARENT, &parent);
353 if (rc)
354 return (ERR_PTR(rc));
355
356 spl_inode_lock(parent.dentry->d_inode);
357
358 dentry = lookup_one_len(basename, parent.dentry, len);
359 if (IS_ERR(dentry)) {
360 spl_inode_unlock(parent.dentry->d_inode);
361 path_put(&parent);
362 } else {
363 *path = parent;
364 }
365
366 return (dentry);
367}
368
bcb15891
YS
369/* Based on do_unlinkat() from linux/fs/namei.c */
370int
371vn_remove(const char *path, uio_seg_t seg, int flags)
372{
373 struct dentry *dentry;
374 struct path parent;
375 struct inode *inode = NULL;
376 int rc = 0;
bcb15891
YS
377
378 ASSERT(seg == UIO_SYSSPACE);
379 ASSERT(flags == RMFILE);
380
381 dentry = spl_kern_path_locked(path, &parent);
382 rc = PTR_ERR(dentry);
383 if (!IS_ERR(dentry)) {
10946b02
AX
384 if (parent.dentry->d_name.name[parent.dentry->d_name.len]) {
385 rc = 0;
386 goto slashes;
387 }
bcb15891
YS
388
389 inode = dentry->d_inode;
10946b02
AX
390 if (inode) {
391 atomic_inc(&inode->i_count);
392 } else {
393 rc = 0;
394 goto slashes;
395 }
bcb15891 396
33a20369 397#ifdef HAVE_2ARGS_VFS_UNLINK
bcb15891 398 rc = vfs_unlink(parent.dentry->d_inode, dentry);
33a20369
LG
399#else
400 rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
401#endif /* HAVE_2ARGS_VFS_UNLINK */
bcb15891
YS
402exit1:
403 dput(dentry);
053678f3
BB
404 } else {
405 return (-rc);
bcb15891
YS
406 }
407
408 spl_inode_unlock(parent.dentry->d_inode);
409 if (inode)
410 iput(inode); /* truncate the inode here */
411
412 path_put(&parent);
10946b02 413 return (-rc);
bcb15891
YS
414
415slashes:
416 rc = !dentry->d_inode ? -ENOENT :
417 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
10946b02 418 goto exit1;
bcb15891
YS
419} /* vn_remove() */
420EXPORT_SYMBOL(vn_remove);
421
422/* Based on do_rename() from linux/fs/namei.c */
423int
424vn_rename(const char *oldname, const char *newname, int x1)
425{
426 struct dentry *old_dir, *new_dir;
427 struct dentry *old_dentry, *new_dentry;
428 struct dentry *trap;
429 struct path old_parent, new_parent;
430 int rc = 0;
bcb15891
YS
431
432 old_dentry = spl_kern_path_locked(oldname, &old_parent);
10946b02
AX
433 if (IS_ERR(old_dentry)) {
434 rc = PTR_ERR(old_dentry);
435 goto exit;
436 }
bcb15891
YS
437
438 spl_inode_unlock(old_parent.dentry->d_inode);
439
440 new_dentry = spl_kern_path_locked(newname, &new_parent);
10946b02
AX
441 if (IS_ERR(new_dentry)) {
442 rc = PTR_ERR(new_dentry);
443 goto exit2;
444 }
bcb15891
YS
445
446 spl_inode_unlock(new_parent.dentry->d_inode);
447
448 rc = -EXDEV;
449 if (old_parent.mnt != new_parent.mnt)
10946b02 450 goto exit3;
bcb15891
YS
451
452 old_dir = old_parent.dentry;
453 new_dir = new_parent.dentry;
454 trap = lock_rename(new_dir, old_dir);
455
456 /* source should not be ancestor of target */
457 rc = -EINVAL;
458 if (old_dentry == trap)
10946b02 459 goto exit4;
bcb15891
YS
460
461 /* target should not be an ancestor of source */
462 rc = -ENOTEMPTY;
463 if (new_dentry == trap)
10946b02 464 goto exit4;
bcb15891
YS
465
466 /* source must exist */
467 rc = -ENOENT;
468 if (!old_dentry->d_inode)
10946b02 469 goto exit4;
bcb15891
YS
470
471 /* unless the source is a directory trailing slashes give -ENOTDIR */
472 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
473 rc = -ENOTDIR;
474 if (old_dentry->d_name.name[old_dentry->d_name.len])
10946b02 475 goto exit4;
bcb15891 476 if (new_dentry->d_name.name[new_dentry->d_name.len])
10946b02 477 goto exit4;
bcb15891
YS
478 }
479
9e4fb5c2 480#if defined(HAVE_4ARGS_VFS_RENAME)
bcb15891 481 rc = vfs_rename(old_dir->d_inode, old_dentry,
33a20369 482 new_dir->d_inode, new_dentry);
9e4fb5c2 483#elif defined(HAVE_5ARGS_VFS_RENAME)
33a20369
LG
484 rc = vfs_rename(old_dir->d_inode, old_dentry,
485 new_dir->d_inode, new_dentry, NULL);
9e4fb5c2
LG
486#else
487 rc = vfs_rename(old_dir->d_inode, old_dentry,
488 new_dir->d_inode, new_dentry, NULL, 0);
489#endif
bcb15891
YS
490exit4:
491 unlock_rename(new_dir, old_dir);
492exit3:
493 dput(new_dentry);
494 path_put(&new_parent);
495exit2:
496 dput(old_dentry);
497 path_put(&old_parent);
498exit:
10946b02 499 return (-rc);
0b3cf046 500}
4b171585 501EXPORT_SYMBOL(vn_rename);
0b3cf046 502
4b171585 503int
36e6f861 504vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
0b3cf046 505{
4b171585 506 struct file *fp;
dcd9cb5a 507 struct kstat stat;
4b171585 508 int rc;
509
937879f1 510 ASSERT(vp);
511 ASSERT(vp->v_file);
512 ASSERT(vap);
4b171585 513
e4f1d29f 514 fp = vp->v_file;
4b171585 515
2a305c34
RY
516#ifdef HAVE_2ARGS_VFS_GETATTR
517 rc = vfs_getattr(&fp->f_path, &stat);
518#else
519 rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
520#endif
4b171585 521 if (rc)
10946b02 522 return (-rc);
4b171585 523
4295b530 524 vap->va_type = vn_mode_to_vtype(stat.mode);
4b171585 525 vap->va_mode = stat.mode;
80093b6f
AX
526 vap->va_uid = KUID_TO_SUID(stat.uid);
527 vap->va_gid = KGID_TO_SGID(stat.gid);
4b171585 528 vap->va_fsid = 0;
529 vap->va_nodeid = stat.ino;
530 vap->va_nlink = stat.nlink;
531 vap->va_size = stat.size;
47995fa6 532 vap->va_blksize = stat.blksize;
dcd9cb5a
BB
533 vap->va_atime = stat.atime;
534 vap->va_mtime = stat.mtime;
535 vap->va_ctime = stat.ctime;
4b171585 536 vap->va_rdev = stat.rdev;
dcd9cb5a 537 vap->va_nblocks = stat.blocks;
4b171585 538
10946b02 539 return (0);
0b3cf046 540}
4b171585 541EXPORT_SYMBOL(vn_getattr);
542
2f5d55aa 543int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
4b171585 544{
36e6f861 545 int datasync = 0;
10946b02
AX
546 int error;
547 int fstrans;
36e6f861 548
937879f1 549 ASSERT(vp);
550 ASSERT(vp->v_file);
4b171585 551
36e6f861 552 if (flags & FDSYNC)
553 datasync = 1;
554
10946b02
AX
555 /*
556 * May enter XFS which generates a warning when PF_FSTRANS is set.
557 * To avoid this the flag is cleared over vfs_sync() and then reset.
558 */
559 fstrans = spl_fstrans_check();
560 if (fstrans)
561 current->flags &= ~(PF_FSTRANS);
562
563 error = -spl_filp_fsync(vp->v_file, datasync);
564 if (fstrans)
565 current->flags |= PF_FSTRANS;
566
567 return (error);
4b171585 568} /* vn_fsync() */
569EXPORT_SYMBOL(vn_fsync);
af828292 570
bbdc6ae4
ED
571int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
572 offset_t offset, void *x6, void *x7)
573{
574 int error = EOPNOTSUPP;
bbdc6ae4
ED
575
576 if (cmd != F_FREESP || bfp->l_whence != 0)
10946b02 577 return (EOPNOTSUPP);
bbdc6ae4
ED
578
579 ASSERT(vp);
580 ASSERT(vp->v_file);
581 ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
582
583#ifdef FALLOC_FL_PUNCH_HOLE
1c7b3eaf
BB
584 /*
585 * When supported by the underlying file system preferentially
586 * use the fallocate() callback to preallocate the space.
587 */
588 error = -spl_filp_fallocate(vp->v_file,
589 FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
590 bfp->l_start, bfp->l_len);
591 if (error == 0)
10946b02 592 return (0);
bbdc6ae4
ED
593#endif
594
595#ifdef HAVE_INODE_TRUNCATE_RANGE
596 if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
597 vp->v_file->f_dentry->d_inode->i_op &&
598 vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
599 off_t end = bfp->l_start + bfp->l_len;
600 /*
601 * Judging from the code in shmem_truncate_range(),
602 * it seems the kernel expects the end offset to be
603 * inclusive and aligned to the end of a page.
604 */
605 if (end % PAGE_SIZE != 0) {
606 end &= ~(off_t)(PAGE_SIZE - 1);
607 if (end <= bfp->l_start)
10946b02 608 return (0);
bbdc6ae4
ED
609 }
610 --end;
611
612 vp->v_file->f_dentry->d_inode->i_op->truncate_range(
613 vp->v_file->f_dentry->d_inode,
614 bfp->l_start, end
615 );
10946b02 616 return (0);
bbdc6ae4
ED
617 }
618#endif
619
10946b02 620 return (error);
bbdc6ae4
ED
621}
622EXPORT_SYMBOL(vn_space);
623
e4f1d29f 624/* Function must be called while holding the vn_file_lock */
625static file_t *
626file_find(int fd)
627{
628 file_t *fp;
629
937879f1 630 ASSERT(spin_is_locked(&vn_file_lock));
e4f1d29f 631
632 list_for_each_entry(fp, &vn_file_list, f_list) {
763b2f3b 633 if (fd == fp->f_fd && fp->f_task == current) {
937879f1 634 ASSERT(atomic_read(&fp->f_ref) != 0);
e4f1d29f 635 return fp;
636 }
637 }
638
639 return NULL;
640} /* file_find() */
641
642file_t *
643vn_getf(int fd)
644{
645 struct kstat stat;
646 struct file *lfp;
647 file_t *fp;
648 vnode_t *vp;
937879f1 649 int rc = 0;
e4f1d29f 650
651 /* Already open just take an extra reference */
652 spin_lock(&vn_file_lock);
653
654 fp = file_find(fd);
655 if (fp) {
656 atomic_inc(&fp->f_ref);
657 spin_unlock(&vn_file_lock);
10946b02 658 return (fp);
e4f1d29f 659 }
660
661 spin_unlock(&vn_file_lock);
662
663 /* File was not yet opened create the object and setup */
4afaaefa 664 fp = kmem_cache_alloc(vn_file_cache, KM_SLEEP);
e4f1d29f 665 if (fp == NULL)
10946b02 666 goto out;
e4f1d29f 667
668 mutex_enter(&fp->f_lock);
669
670 fp->f_fd = fd;
763b2f3b 671 fp->f_task = current;
e4f1d29f 672 fp->f_offset = 0;
673 atomic_inc(&fp->f_ref);
674
675 lfp = fget(fd);
676 if (lfp == NULL)
10946b02 677 goto out_mutex;
e4f1d29f 678
679 vp = vn_alloc(KM_SLEEP);
680 if (vp == NULL)
10946b02 681 goto out_fget;
e4f1d29f 682
2a305c34
RY
683#ifdef HAVE_2ARGS_VFS_GETATTR
684 rc = vfs_getattr(&lfp->f_path, &stat);
685#else
686 rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
687#endif
688 if (rc)
10946b02 689 goto out_vnode;
e4f1d29f 690
691 mutex_enter(&vp->v_lock);
4295b530 692 vp->v_type = vn_mode_to_vtype(stat.mode);
e4f1d29f 693 vp->v_file = lfp;
694 mutex_exit(&vp->v_lock);
695
696 fp->f_vnode = vp;
697 fp->f_file = lfp;
698
699 /* Put it on the tracking list */
700 spin_lock(&vn_file_lock);
701 list_add(&fp->f_list, &vn_file_list);
702 spin_unlock(&vn_file_lock);
703
704 mutex_exit(&fp->f_lock);
10946b02 705 return (fp);
e4f1d29f 706
707out_vnode:
e4f1d29f 708 vn_free(vp);
709out_fget:
e4f1d29f 710 fput(lfp);
711out_mutex:
e4f1d29f 712 mutex_exit(&fp->f_lock);
713 kmem_cache_free(vn_file_cache, fp);
714out:
10946b02 715 return (NULL);
e4f1d29f 716} /* getf() */
717EXPORT_SYMBOL(getf);
718
719static void releasef_locked(file_t *fp)
720{
937879f1 721 ASSERT(fp->f_file);
722 ASSERT(fp->f_vnode);
e4f1d29f 723
724 /* Unlinked from list, no refs, safe to free outside mutex */
725 fput(fp->f_file);
726 vn_free(fp->f_vnode);
727
728 kmem_cache_free(vn_file_cache, fp);
729}
730
731void
732vn_releasef(int fd)
733{
734 file_t *fp;
735
736 spin_lock(&vn_file_lock);
737 fp = file_find(fd);
738 if (fp) {
739 atomic_dec(&fp->f_ref);
740 if (atomic_read(&fp->f_ref) > 0) {
741 spin_unlock(&vn_file_lock);
742 return;
743 }
744
745 list_del(&fp->f_list);
746 releasef_locked(fp);
747 }
748 spin_unlock(&vn_file_lock);
749
750 return;
751} /* releasef() */
752EXPORT_SYMBOL(releasef);
753
10946b02
AX
754static void
755#ifdef HAVE_SET_FS_PWD_WITH_CONST
756vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
757#else
758vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
759#endif /* HAVE_SET_FS_PWD_WITH_CONST */
51a727e9 760{
9b2048c2
BB
761 struct path old_pwd;
762
10946b02 763#ifdef HAVE_FS_STRUCT_SPINLOCK
9b2048c2
BB
764 spin_lock(&fs->lock);
765 old_pwd = fs->pwd;
766 fs->pwd = *path;
767 path_get(path);
768 spin_unlock(&fs->lock);
10946b02 769#else
9b2048c2
BB
770 write_lock(&fs->lock);
771 old_pwd = fs->pwd;
772 fs->pwd = *path;
773 path_get(path);
774 write_unlock(&fs->lock);
10946b02 775#endif /* HAVE_FS_STRUCT_SPINLOCK */
9b2048c2
BB
776
777 if (old_pwd.dentry)
778 path_put(&old_pwd);
51a727e9 779}
51a727e9
BB
780
781int
782vn_set_pwd(const char *filename)
783{
51a727e9 784 struct path path;
82a358d9 785 mm_segment_t saved_fs;
51a727e9 786 int rc;
51a727e9 787
82a358d9
BB
788 /*
789 * user_path_dir() and __user_walk() both expect 'filename' to be
790 * a user space address so we must briefly increase the data segment
791 * size to ensure strncpy_from_user() does not fail with -EFAULT.
792 */
793 saved_fs = get_fs();
794 set_fs(get_ds());
795
51a727e9
BB
796 rc = user_path_dir(filename, &path);
797 if (rc)
10946b02 798 goto out;
51a727e9
BB
799
800 rc = inode_permission(path.dentry->d_inode, MAY_EXEC | MAY_ACCESS);
801 if (rc)
10946b02 802 goto dput_and_out;
51a727e9 803
10946b02 804 vn_set_fs_pwd(current->fs, &path);
51a727e9
BB
805
806dput_and_out:
807 path_put(&path);
51a727e9 808out:
82a358d9
BB
809 set_fs(saved_fs);
810
10946b02 811 return (-rc);
51a727e9
BB
812} /* vn_set_pwd() */
813EXPORT_SYMBOL(vn_set_pwd);
814
af828292 815static int
816vn_cache_constructor(void *buf, void *cdrarg, int kmflags)
817{
818 struct vnode *vp = buf;
819
820 mutex_init(&vp->v_lock, NULL, MUTEX_DEFAULT, NULL);
821
822 return (0);
823} /* vn_cache_constructor() */
824
825static void
826vn_cache_destructor(void *buf, void *cdrarg)
827{
828 struct vnode *vp = buf;
829
830 mutex_destroy(&vp->v_lock);
831} /* vn_cache_destructor() */
832
e4f1d29f 833static int
834vn_file_cache_constructor(void *buf, void *cdrarg, int kmflags)
835{
836 file_t *fp = buf;
837
838 atomic_set(&fp->f_ref, 0);
839 mutex_init(&fp->f_lock, NULL, MUTEX_DEFAULT, NULL);
4e62fd41 840 INIT_LIST_HEAD(&fp->f_list);
e4f1d29f 841
842 return (0);
843} /* file_cache_constructor() */
844
845static void
846vn_file_cache_destructor(void *buf, void *cdrarg)
847{
848 file_t *fp = buf;
849
850 mutex_destroy(&fp->f_lock);
851} /* vn_file_cache_destructor() */
852
af828292 853int
12ff95ff 854spl_vn_init(void)
af828292 855{
57d86234 856 vn_cache = kmem_cache_create("spl_vn_cache",
857 sizeof(struct vnode), 64,
5d86345d 858 vn_cache_constructor,
859 vn_cache_destructor,
a5b40eed 860 NULL, NULL, NULL, KMC_KMEM);
e4f1d29f 861
862 vn_file_cache = kmem_cache_create("spl_vn_file_cache",
863 sizeof(file_t), 64,
864 vn_file_cache_constructor,
865 vn_file_cache_destructor,
a5b40eed 866 NULL, NULL, NULL, KMC_KMEM);
10946b02 867 return (0);
af828292 868} /* vn_init() */
869
870void
12ff95ff 871spl_vn_fini(void)
af828292 872{
e4f1d29f 873 file_t *fp, *next_fp;
2fb9b26a 874 int leaked = 0;
e4f1d29f 875
876 spin_lock(&vn_file_lock);
877
878 list_for_each_entry_safe(fp, next_fp, &vn_file_list, f_list) {
879 list_del(&fp->f_list);
880 releasef_locked(fp);
881 leaked++;
882 }
883
e4f1d29f 884 spin_unlock(&vn_file_lock);
885
886 if (leaked > 0)
10946b02 887 printk(KERN_WARNING "WARNING: %d vnode files leaked\n", leaked);
e4f1d29f 888
2371321e 889 kmem_cache_destroy(vn_file_cache);
2fb9b26a 890 kmem_cache_destroy(vn_cache);
e4f1d29f 891
892 return;
af828292 893} /* vn_fini() */