pipe: change the ->pin() operation to ->confirm()
[linux-2.6/mini2440.git] / fs / nfsd / vfs.c
blob8604e35bd48e2d5966a244f1763f62def303dc0a
1 #define MSNFS /* HACK HACK */
2 /*
3 * linux/fs/nfsd/vfs.c
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
22 #include <linux/fs.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/in.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
41 #ifdef CONFIG_NFSD_V3
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
51 #ifdef CONFIG_NFSD_V4
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
64 /* We must ignore files (but only files) which might have mandatory
65 * locks on them because there is no way to know if the accesser has
66 * the lock.
68 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
71 * This is a cache of readahead params that help us choose the proper
72 * readahead strategy. Initially, we set all readahead parameters to 0
73 * and let the VFS handle things.
74 * If you increase the number of cached files very much, you'll need to
75 * add a hash table here.
77 struct raparms {
78 struct raparms *p_next;
79 unsigned int p_count;
80 ino_t p_ino;
81 dev_t p_dev;
82 int p_set;
83 struct file_ra_state p_ra;
84 unsigned int p_hindex;
87 struct raparm_hbucket {
88 struct raparms *pb_head;
89 spinlock_t pb_lock;
90 } ____cacheline_aligned_in_smp;
92 static struct raparms * raparml;
93 #define RAPARM_HASH_BITS 4
94 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
95 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
96 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
98 /*
99 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
100 * a mount point.
101 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
102 * or nfs_ok having possibly changed *dpp and *expp
105 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
106 struct svc_export **expp)
108 struct svc_export *exp = *expp, *exp2 = NULL;
109 struct dentry *dentry = *dpp;
110 struct vfsmount *mnt = mntget(exp->ex_mnt);
111 struct dentry *mounts = dget(dentry);
112 int err = 0;
114 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
116 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
117 if (IS_ERR(exp2)) {
118 err = PTR_ERR(exp2);
119 dput(mounts);
120 mntput(mnt);
121 goto out;
123 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
124 /* successfully crossed mount point */
125 exp_put(exp);
126 *expp = exp2;
127 dput(dentry);
128 *dpp = mounts;
129 } else {
130 if (exp2) exp_put(exp2);
131 dput(mounts);
133 mntput(mnt);
134 out:
135 return err;
139 * Look up one component of a pathname.
140 * N.B. After this call _both_ fhp and resfh need an fh_put
142 * If the lookup would cross a mountpoint, and the mounted filesystem
143 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
144 * accepted as it stands and the mounted directory is
145 * returned. Otherwise the covered directory is returned.
146 * NOTE: this mountpoint crossing is not supported properly by all
147 * clients and is explicitly disallowed for NFSv3
148 * NeilBrown <neilb@cse.unsw.edu.au>
150 __be32
151 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
152 int len, struct svc_fh *resfh)
154 struct svc_export *exp;
155 struct dentry *dparent;
156 struct dentry *dentry;
157 __be32 err;
158 int host_err;
160 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
162 /* Obtain dentry and export. */
163 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
164 if (err)
165 return err;
167 dparent = fhp->fh_dentry;
168 exp = fhp->fh_export;
169 exp_get(exp);
171 err = nfserr_acces;
173 /* Lookup the name, but don't follow links */
174 if (isdotent(name, len)) {
175 if (len==1)
176 dentry = dget(dparent);
177 else if (dparent != exp->ex_dentry) {
178 dentry = dget_parent(dparent);
179 } else if (!EX_NOHIDE(exp))
180 dentry = dget(dparent); /* .. == . just like at / */
181 else {
182 /* checking mountpoint crossing is very different when stepping up */
183 struct svc_export *exp2 = NULL;
184 struct dentry *dp;
185 struct vfsmount *mnt = mntget(exp->ex_mnt);
186 dentry = dget(dparent);
187 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
189 dp = dget_parent(dentry);
190 dput(dentry);
191 dentry = dp;
193 exp2 = exp_parent(exp->ex_client, mnt, dentry,
194 &rqstp->rq_chandle);
195 if (IS_ERR(exp2)) {
196 host_err = PTR_ERR(exp2);
197 dput(dentry);
198 mntput(mnt);
199 goto out_nfserr;
201 if (!exp2) {
202 dput(dentry);
203 dentry = dget(dparent);
204 } else {
205 exp_put(exp);
206 exp = exp2;
208 mntput(mnt);
210 } else {
211 fh_lock(fhp);
212 dentry = lookup_one_len(name, dparent, len);
213 host_err = PTR_ERR(dentry);
214 if (IS_ERR(dentry))
215 goto out_nfserr;
217 * check if we have crossed a mount point ...
219 if (d_mountpoint(dentry)) {
220 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
221 dput(dentry);
222 goto out_nfserr;
227 * Note: we compose the file handle now, but as the
228 * dentry may be negative, it may need to be updated.
230 err = fh_compose(resfh, exp, dentry, fhp);
231 if (!err && !dentry->d_inode)
232 err = nfserr_noent;
233 dput(dentry);
234 out:
235 exp_put(exp);
236 return err;
238 out_nfserr:
239 err = nfserrno(host_err);
240 goto out;
244 * Set various file attributes.
245 * N.B. After this call fhp needs an fh_put
247 __be32
248 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
249 int check_guard, time_t guardtime)
251 struct dentry *dentry;
252 struct inode *inode;
253 int accmode = MAY_SATTR;
254 int ftype = 0;
255 int imode;
256 __be32 err;
257 int host_err;
258 int size_change = 0;
260 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
261 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
262 if (iap->ia_valid & ATTR_SIZE)
263 ftype = S_IFREG;
265 /* Get inode */
266 err = fh_verify(rqstp, fhp, ftype, accmode);
267 if (err)
268 goto out;
270 dentry = fhp->fh_dentry;
271 inode = dentry->d_inode;
273 /* Ignore any mode updates on symlinks */
274 if (S_ISLNK(inode->i_mode))
275 iap->ia_valid &= ~ATTR_MODE;
277 if (!iap->ia_valid)
278 goto out;
280 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
281 * which only requires access, and "set-[ac]time-to-X" which
282 * requires ownership.
283 * So if it looks like it might be "set both to the same time which
284 * is close to now", and if inode_change_ok fails, then we
285 * convert to "set to now" instead of "set to explicit time"
287 * We only call inode_change_ok as the last test as technically
288 * it is not an interface that we should be using. It is only
289 * valid if the filesystem does not define it's own i_op->setattr.
291 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
292 #define MAX_TOUCH_TIME_ERROR (30*60)
293 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
294 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
296 /* Looks probable. Now just make sure time is in the right ballpark.
297 * Solaris, at least, doesn't seem to care what the time request is.
298 * We require it be within 30 minutes of now.
300 time_t delta = iap->ia_atime.tv_sec - get_seconds();
301 if (delta<0) delta = -delta;
302 if (delta < MAX_TOUCH_TIME_ERROR &&
303 inode_change_ok(inode, iap) != 0) {
304 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
305 * this will cause notify_change to set these times to "now"
307 iap->ia_valid &= ~BOTH_TIME_SET;
311 /* The size case is special. It changes the file as well as the attributes. */
312 if (iap->ia_valid & ATTR_SIZE) {
313 if (iap->ia_size < inode->i_size) {
314 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
315 if (err)
316 goto out;
320 * If we are changing the size of the file, then
321 * we need to break all leases.
323 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
324 if (host_err == -EWOULDBLOCK)
325 host_err = -ETIMEDOUT;
326 if (host_err) /* ENOMEM or EWOULDBLOCK */
327 goto out_nfserr;
329 host_err = get_write_access(inode);
330 if (host_err)
331 goto out_nfserr;
333 size_change = 1;
334 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
335 if (host_err) {
336 put_write_access(inode);
337 goto out_nfserr;
339 DQUOT_INIT(inode);
342 imode = inode->i_mode;
343 if (iap->ia_valid & ATTR_MODE) {
344 iap->ia_mode &= S_IALLUGO;
345 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
348 /* Revoke setuid/setgid bit on chown/chgrp */
349 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
350 iap->ia_valid |= ATTR_KILL_SUID;
351 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
352 iap->ia_valid |= ATTR_KILL_SGID;
354 /* Change the attributes. */
356 iap->ia_valid |= ATTR_CTIME;
358 err = nfserr_notsync;
359 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
360 fh_lock(fhp);
361 host_err = notify_change(dentry, iap);
362 err = nfserrno(host_err);
363 fh_unlock(fhp);
365 if (size_change)
366 put_write_access(inode);
367 if (!err)
368 if (EX_ISSYNC(fhp->fh_export))
369 write_inode_now(inode, 1);
370 out:
371 return err;
373 out_nfserr:
374 err = nfserrno(host_err);
375 goto out;
378 #if defined(CONFIG_NFSD_V2_ACL) || \
379 defined(CONFIG_NFSD_V3_ACL) || \
380 defined(CONFIG_NFSD_V4)
381 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
383 ssize_t buflen;
385 buflen = vfs_getxattr(dentry, key, NULL, 0);
386 if (buflen <= 0)
387 return buflen;
389 *buf = kmalloc(buflen, GFP_KERNEL);
390 if (!*buf)
391 return -ENOMEM;
393 return vfs_getxattr(dentry, key, *buf, buflen);
395 #endif
397 #if defined(CONFIG_NFSD_V4)
398 static int
399 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
401 int len;
402 size_t buflen;
403 char *buf = NULL;
404 int error = 0;
406 buflen = posix_acl_xattr_size(pacl->a_count);
407 buf = kmalloc(buflen, GFP_KERNEL);
408 error = -ENOMEM;
409 if (buf == NULL)
410 goto out;
412 len = posix_acl_to_xattr(pacl, buf, buflen);
413 if (len < 0) {
414 error = len;
415 goto out;
418 error = vfs_setxattr(dentry, key, buf, len, 0);
419 out:
420 kfree(buf);
421 return error;
424 __be32
425 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
426 struct nfs4_acl *acl)
428 __be32 error;
429 int host_error;
430 struct dentry *dentry;
431 struct inode *inode;
432 struct posix_acl *pacl = NULL, *dpacl = NULL;
433 unsigned int flags = 0;
435 /* Get inode */
436 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
437 if (error)
438 goto out;
440 dentry = fhp->fh_dentry;
441 inode = dentry->d_inode;
442 if (S_ISDIR(inode->i_mode))
443 flags = NFS4_ACL_DIR;
445 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
446 if (host_error == -EINVAL) {
447 error = nfserr_attrnotsupp;
448 goto out;
449 } else if (host_error < 0)
450 goto out_nfserr;
452 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
453 if (host_error < 0)
454 goto out_nfserr;
456 if (S_ISDIR(inode->i_mode)) {
457 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
458 if (host_error < 0)
459 goto out_nfserr;
462 error = nfs_ok;
464 out:
465 posix_acl_release(pacl);
466 posix_acl_release(dpacl);
467 return (error);
468 out_nfserr:
469 if (host_error == -EOPNOTSUPP)
470 error = nfserr_attrnotsupp;
471 else
472 error = nfserrno(host_error);
473 goto out;
476 static struct posix_acl *
477 _get_posix_acl(struct dentry *dentry, char *key)
479 void *buf = NULL;
480 struct posix_acl *pacl = NULL;
481 int buflen;
483 buflen = nfsd_getxattr(dentry, key, &buf);
484 if (!buflen)
485 buflen = -ENODATA;
486 if (buflen <= 0)
487 return ERR_PTR(buflen);
489 pacl = posix_acl_from_xattr(buf, buflen);
490 kfree(buf);
491 return pacl;
495 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
497 struct inode *inode = dentry->d_inode;
498 int error = 0;
499 struct posix_acl *pacl = NULL, *dpacl = NULL;
500 unsigned int flags = 0;
502 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
503 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
504 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
505 if (IS_ERR(pacl)) {
506 error = PTR_ERR(pacl);
507 pacl = NULL;
508 goto out;
511 if (S_ISDIR(inode->i_mode)) {
512 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
513 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
514 dpacl = NULL;
515 else if (IS_ERR(dpacl)) {
516 error = PTR_ERR(dpacl);
517 dpacl = NULL;
518 goto out;
520 flags = NFS4_ACL_DIR;
523 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
524 if (IS_ERR(*acl)) {
525 error = PTR_ERR(*acl);
526 *acl = NULL;
528 out:
529 posix_acl_release(pacl);
530 posix_acl_release(dpacl);
531 return error;
534 #endif /* defined(CONFIG_NFS_V4) */
536 #ifdef CONFIG_NFSD_V3
538 * Check server access rights to a file system object
540 struct accessmap {
541 u32 access;
542 int how;
544 static struct accessmap nfs3_regaccess[] = {
545 { NFS3_ACCESS_READ, MAY_READ },
546 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
547 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
548 { NFS3_ACCESS_EXTEND, MAY_WRITE },
550 { 0, 0 }
553 static struct accessmap nfs3_diraccess[] = {
554 { NFS3_ACCESS_READ, MAY_READ },
555 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
556 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
557 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
558 { NFS3_ACCESS_DELETE, MAY_REMOVE },
560 { 0, 0 }
563 static struct accessmap nfs3_anyaccess[] = {
564 /* Some clients - Solaris 2.6 at least, make an access call
565 * to the server to check for access for things like /dev/null
566 * (which really, the server doesn't care about). So
567 * We provide simple access checking for them, looking
568 * mainly at mode bits, and we make sure to ignore read-only
569 * filesystem checks
571 { NFS3_ACCESS_READ, MAY_READ },
572 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
573 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
574 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
576 { 0, 0 }
579 __be32
580 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
582 struct accessmap *map;
583 struct svc_export *export;
584 struct dentry *dentry;
585 u32 query, result = 0, sresult = 0;
586 __be32 error;
588 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
589 if (error)
590 goto out;
592 export = fhp->fh_export;
593 dentry = fhp->fh_dentry;
595 if (S_ISREG(dentry->d_inode->i_mode))
596 map = nfs3_regaccess;
597 else if (S_ISDIR(dentry->d_inode->i_mode))
598 map = nfs3_diraccess;
599 else
600 map = nfs3_anyaccess;
603 query = *access;
604 for (; map->access; map++) {
605 if (map->access & query) {
606 __be32 err2;
608 sresult |= map->access;
610 err2 = nfsd_permission(export, dentry, map->how);
611 switch (err2) {
612 case nfs_ok:
613 result |= map->access;
614 break;
616 /* the following error codes just mean the access was not allowed,
617 * rather than an error occurred */
618 case nfserr_rofs:
619 case nfserr_acces:
620 case nfserr_perm:
621 /* simply don't "or" in the access bit. */
622 break;
623 default:
624 error = err2;
625 goto out;
629 *access = result;
630 if (supported)
631 *supported = sresult;
633 out:
634 return error;
636 #endif /* CONFIG_NFSD_V3 */
641 * Open an existing file or directory.
642 * The access argument indicates the type of open (read/write/lock)
643 * N.B. After this call fhp needs an fh_put
645 __be32
646 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
647 int access, struct file **filp)
649 struct dentry *dentry;
650 struct inode *inode;
651 int flags = O_RDONLY|O_LARGEFILE;
652 __be32 err;
653 int host_err;
656 * If we get here, then the client has already done an "open",
657 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
658 * in case a chmod has now revoked permission.
660 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
661 if (err)
662 goto out;
664 dentry = fhp->fh_dentry;
665 inode = dentry->d_inode;
667 /* Disallow write access to files with the append-only bit set
668 * or any access when mandatory locking enabled
670 err = nfserr_perm;
671 if (IS_APPEND(inode) && (access & MAY_WRITE))
672 goto out;
673 if (IS_ISMNDLK(inode))
674 goto out;
676 if (!inode->i_fop)
677 goto out;
680 * Check to see if there are any leases on this file.
681 * This may block while leases are broken.
683 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
684 if (host_err == -EWOULDBLOCK)
685 host_err = -ETIMEDOUT;
686 if (host_err) /* NOMEM or WOULDBLOCK */
687 goto out_nfserr;
689 if (access & MAY_WRITE) {
690 if (access & MAY_READ)
691 flags = O_RDWR|O_LARGEFILE;
692 else
693 flags = O_WRONLY|O_LARGEFILE;
695 DQUOT_INIT(inode);
697 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
698 if (IS_ERR(*filp))
699 host_err = PTR_ERR(*filp);
700 out_nfserr:
701 err = nfserrno(host_err);
702 out:
703 return err;
707 * Close a file.
709 void
710 nfsd_close(struct file *filp)
712 fput(filp);
716 * Sync a file
717 * As this calls fsync (not fdatasync) there is no need for a write_inode
718 * after it.
720 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
721 const struct file_operations *fop)
723 struct inode *inode = dp->d_inode;
724 int (*fsync) (struct file *, struct dentry *, int);
725 int err;
727 err = filemap_fdatawrite(inode->i_mapping);
728 if (err == 0 && fop && (fsync = fop->fsync))
729 err = fsync(filp, dp, 0);
730 if (err == 0)
731 err = filemap_fdatawait(inode->i_mapping);
733 return err;
737 static int
738 nfsd_sync(struct file *filp)
740 int err;
741 struct inode *inode = filp->f_path.dentry->d_inode;
742 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
743 mutex_lock(&inode->i_mutex);
744 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
745 mutex_unlock(&inode->i_mutex);
747 return err;
751 nfsd_sync_dir(struct dentry *dp)
753 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
757 * Obtain the readahead parameters for the file
758 * specified by (dev, ino).
761 static inline struct raparms *
762 nfsd_get_raparms(dev_t dev, ino_t ino)
764 struct raparms *ra, **rap, **frap = NULL;
765 int depth = 0;
766 unsigned int hash;
767 struct raparm_hbucket *rab;
769 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
770 rab = &raparm_hash[hash];
772 spin_lock(&rab->pb_lock);
773 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
774 if (ra->p_ino == ino && ra->p_dev == dev)
775 goto found;
776 depth++;
777 if (ra->p_count == 0)
778 frap = rap;
780 depth = nfsdstats.ra_size*11/10;
781 if (!frap) {
782 spin_unlock(&rab->pb_lock);
783 return NULL;
785 rap = frap;
786 ra = *frap;
787 ra->p_dev = dev;
788 ra->p_ino = ino;
789 ra->p_set = 0;
790 ra->p_hindex = hash;
791 found:
792 if (rap != &rab->pb_head) {
793 *rap = ra->p_next;
794 ra->p_next = rab->pb_head;
795 rab->pb_head = ra;
797 ra->p_count++;
798 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
799 spin_unlock(&rab->pb_lock);
800 return ra;
804 * Grab and keep cached pages associated with a file in the svc_rqst
805 * so that they can be passed to the network sendmsg/sendpage routines
806 * directly. They will be released after the sending has completed.
808 static int
809 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
810 struct splice_desc *sd)
812 struct svc_rqst *rqstp = sd->u.data;
813 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
814 struct page *page = buf->page;
815 size_t size;
816 int ret;
818 ret = buf->ops->confirm(pipe, buf);
819 if (unlikely(ret))
820 return ret;
822 size = sd->len;
824 if (rqstp->rq_res.page_len == 0) {
825 get_page(page);
826 put_page(*pp);
827 *pp = page;
828 rqstp->rq_resused++;
829 rqstp->rq_res.page_base = buf->offset;
830 rqstp->rq_res.page_len = size;
831 } else if (page != pp[-1]) {
832 get_page(page);
833 if (*pp)
834 put_page(*pp);
835 *pp = page;
836 rqstp->rq_resused++;
837 rqstp->rq_res.page_len += size;
838 } else
839 rqstp->rq_res.page_len += size;
841 return size;
844 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
845 struct splice_desc *sd)
847 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
850 static __be32
851 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
852 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
854 struct inode *inode;
855 struct raparms *ra;
856 mm_segment_t oldfs;
857 __be32 err;
858 int host_err;
860 err = nfserr_perm;
861 inode = file->f_path.dentry->d_inode;
862 #ifdef MSNFS
863 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
864 (!lock_may_read(inode, offset, *count)))
865 goto out;
866 #endif
868 /* Get readahead parameters */
869 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
871 if (ra && ra->p_set)
872 file->f_ra = ra->p_ra;
874 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
875 struct splice_desc sd = {
876 .len = 0,
877 .total_len = *count,
878 .pos = offset,
879 .u.data = rqstp,
882 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
883 } else {
884 oldfs = get_fs();
885 set_fs(KERNEL_DS);
886 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
887 set_fs(oldfs);
890 /* Write back readahead params */
891 if (ra) {
892 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
893 spin_lock(&rab->pb_lock);
894 ra->p_ra = file->f_ra;
895 ra->p_set = 1;
896 ra->p_count--;
897 spin_unlock(&rab->pb_lock);
900 if (host_err >= 0) {
901 nfsdstats.io_read += host_err;
902 *count = host_err;
903 err = 0;
904 fsnotify_access(file->f_path.dentry);
905 } else
906 err = nfserrno(host_err);
907 out:
908 return err;
911 static void kill_suid(struct dentry *dentry)
913 struct iattr ia;
914 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
916 mutex_lock(&dentry->d_inode->i_mutex);
917 notify_change(dentry, &ia);
918 mutex_unlock(&dentry->d_inode->i_mutex);
921 static __be32
922 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
923 loff_t offset, struct kvec *vec, int vlen,
924 unsigned long cnt, int *stablep)
926 struct svc_export *exp;
927 struct dentry *dentry;
928 struct inode *inode;
929 mm_segment_t oldfs;
930 __be32 err = 0;
931 int host_err;
932 int stable = *stablep;
934 #ifdef MSNFS
935 err = nfserr_perm;
937 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
938 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
939 goto out;
940 #endif
942 dentry = file->f_path.dentry;
943 inode = dentry->d_inode;
944 exp = fhp->fh_export;
947 * Request sync writes if
948 * - the sync export option has been set, or
949 * - the client requested O_SYNC behavior (NFSv3 feature).
950 * - The file system doesn't support fsync().
951 * When gathered writes have been configured for this volume,
952 * flushing the data to disk is handled separately below.
955 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
956 stable = 2;
957 *stablep = 2; /* FILE_SYNC */
960 if (!EX_ISSYNC(exp))
961 stable = 0;
962 if (stable && !EX_WGATHER(exp))
963 file->f_flags |= O_SYNC;
965 /* Write the data. */
966 oldfs = get_fs(); set_fs(KERNEL_DS);
967 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
968 set_fs(oldfs);
969 if (host_err >= 0) {
970 nfsdstats.io_write += cnt;
971 fsnotify_modify(file->f_path.dentry);
974 /* clear setuid/setgid flag after write */
975 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
976 kill_suid(dentry);
978 if (host_err >= 0 && stable) {
979 static ino_t last_ino;
980 static dev_t last_dev;
983 * Gathered writes: If another process is currently
984 * writing to the file, there's a high chance
985 * this is another nfsd (triggered by a bulk write
986 * from a client's biod). Rather than syncing the
987 * file with each write request, we sleep for 10 msec.
989 * I don't know if this roughly approximates
990 * C. Juszak's idea of gathered writes, but it's a
991 * nice and simple solution (IMHO), and it seems to
992 * work:-)
994 if (EX_WGATHER(exp)) {
995 if (atomic_read(&inode->i_writecount) > 1
996 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
997 dprintk("nfsd: write defer %d\n", current->pid);
998 msleep(10);
999 dprintk("nfsd: write resume %d\n", current->pid);
1002 if (inode->i_state & I_DIRTY) {
1003 dprintk("nfsd: write sync %d\n", current->pid);
1004 host_err=nfsd_sync(file);
1006 #if 0
1007 wake_up(&inode->i_wait);
1008 #endif
1010 last_ino = inode->i_ino;
1011 last_dev = inode->i_sb->s_dev;
1014 dprintk("nfsd: write complete host_err=%d\n", host_err);
1015 if (host_err >= 0)
1016 err = 0;
1017 else
1018 err = nfserrno(host_err);
1019 out:
1020 return err;
1024 * Read data from a file. count must contain the requested read count
1025 * on entry. On return, *count contains the number of bytes actually read.
1026 * N.B. After this call fhp needs an fh_put
1028 __be32
1029 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1030 loff_t offset, struct kvec *vec, int vlen,
1031 unsigned long *count)
1033 __be32 err;
1035 if (file) {
1036 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1037 MAY_READ|MAY_OWNER_OVERRIDE);
1038 if (err)
1039 goto out;
1040 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1041 } else {
1042 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1043 if (err)
1044 goto out;
1045 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1046 nfsd_close(file);
1048 out:
1049 return err;
1053 * Write data to a file.
1054 * The stable flag requests synchronous writes.
1055 * N.B. After this call fhp needs an fh_put
1057 __be32
1058 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1059 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1060 int *stablep)
1062 __be32 err = 0;
1064 if (file) {
1065 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1066 MAY_WRITE|MAY_OWNER_OVERRIDE);
1067 if (err)
1068 goto out;
1069 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1070 stablep);
1071 } else {
1072 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1073 if (err)
1074 goto out;
1076 if (cnt)
1077 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1078 cnt, stablep);
1079 nfsd_close(file);
1081 out:
1082 return err;
1085 #ifdef CONFIG_NFSD_V3
1087 * Commit all pending writes to stable storage.
1088 * Strictly speaking, we could sync just the indicated file region here,
1089 * but there's currently no way we can ask the VFS to do so.
1091 * Unfortunately we cannot lock the file to make sure we return full WCC
1092 * data to the client, as locking happens lower down in the filesystem.
1094 __be32
1095 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1096 loff_t offset, unsigned long count)
1098 struct file *file;
1099 __be32 err;
1101 if ((u64)count > ~(u64)offset)
1102 return nfserr_inval;
1104 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1105 return err;
1106 if (EX_ISSYNC(fhp->fh_export)) {
1107 if (file->f_op && file->f_op->fsync) {
1108 err = nfserrno(nfsd_sync(file));
1109 } else {
1110 err = nfserr_notsupp;
1114 nfsd_close(file);
1115 return err;
1117 #endif /* CONFIG_NFSD_V3 */
1120 * Create a file (regular, directory, device, fifo); UNIX sockets
1121 * not yet implemented.
1122 * If the response fh has been verified, the parent directory should
1123 * already be locked. Note that the parent directory is left locked.
1125 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1127 __be32
1128 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1129 char *fname, int flen, struct iattr *iap,
1130 int type, dev_t rdev, struct svc_fh *resfhp)
1132 struct dentry *dentry, *dchild = NULL;
1133 struct inode *dirp;
1134 __be32 err;
1135 int host_err;
1137 err = nfserr_perm;
1138 if (!flen)
1139 goto out;
1140 err = nfserr_exist;
1141 if (isdotent(fname, flen))
1142 goto out;
1144 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1145 if (err)
1146 goto out;
1148 dentry = fhp->fh_dentry;
1149 dirp = dentry->d_inode;
1151 err = nfserr_notdir;
1152 if(!dirp->i_op || !dirp->i_op->lookup)
1153 goto out;
1155 * Check whether the response file handle has been verified yet.
1156 * If it has, the parent directory should already be locked.
1158 if (!resfhp->fh_dentry) {
1159 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1160 fh_lock_nested(fhp, I_MUTEX_PARENT);
1161 dchild = lookup_one_len(fname, dentry, flen);
1162 host_err = PTR_ERR(dchild);
1163 if (IS_ERR(dchild))
1164 goto out_nfserr;
1165 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1166 if (err)
1167 goto out;
1168 } else {
1169 /* called from nfsd_proc_create */
1170 dchild = dget(resfhp->fh_dentry);
1171 if (!fhp->fh_locked) {
1172 /* not actually possible */
1173 printk(KERN_ERR
1174 "nfsd_create: parent %s/%s not locked!\n",
1175 dentry->d_parent->d_name.name,
1176 dentry->d_name.name);
1177 err = nfserr_io;
1178 goto out;
1182 * Make sure the child dentry is still negative ...
1184 err = nfserr_exist;
1185 if (dchild->d_inode) {
1186 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1187 dentry->d_name.name, dchild->d_name.name);
1188 goto out;
1191 if (!(iap->ia_valid & ATTR_MODE))
1192 iap->ia_mode = 0;
1193 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1196 * Get the dir op function pointer.
1198 err = 0;
1199 switch (type) {
1200 case S_IFREG:
1201 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1202 break;
1203 case S_IFDIR:
1204 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1205 break;
1206 case S_IFCHR:
1207 case S_IFBLK:
1208 case S_IFIFO:
1209 case S_IFSOCK:
1210 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1211 break;
1212 default:
1213 printk("nfsd: bad file type %o in nfsd_create\n", type);
1214 host_err = -EINVAL;
1216 if (host_err < 0)
1217 goto out_nfserr;
1219 if (EX_ISSYNC(fhp->fh_export)) {
1220 err = nfserrno(nfsd_sync_dir(dentry));
1221 write_inode_now(dchild->d_inode, 1);
1225 /* Set file attributes. Mode has already been set and
1226 * setting uid/gid works only for root. Irix appears to
1227 * send along the gid when it tries to implement setgid
1228 * directories via NFS.
1230 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1231 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1232 if (err2)
1233 err = err2;
1236 * Update the file handle to get the new inode info.
1238 if (!err)
1239 err = fh_update(resfhp);
1240 out:
1241 if (dchild && !IS_ERR(dchild))
1242 dput(dchild);
1243 return err;
1245 out_nfserr:
1246 err = nfserrno(host_err);
1247 goto out;
1250 #ifdef CONFIG_NFSD_V3
1252 * NFSv3 version of nfsd_create
1254 __be32
1255 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1256 char *fname, int flen, struct iattr *iap,
1257 struct svc_fh *resfhp, int createmode, u32 *verifier,
1258 int *truncp, int *created)
1260 struct dentry *dentry, *dchild = NULL;
1261 struct inode *dirp;
1262 __be32 err;
1263 int host_err;
1264 __u32 v_mtime=0, v_atime=0;
1266 err = nfserr_perm;
1267 if (!flen)
1268 goto out;
1269 err = nfserr_exist;
1270 if (isdotent(fname, flen))
1271 goto out;
1272 if (!(iap->ia_valid & ATTR_MODE))
1273 iap->ia_mode = 0;
1274 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1275 if (err)
1276 goto out;
1278 dentry = fhp->fh_dentry;
1279 dirp = dentry->d_inode;
1281 /* Get all the sanity checks out of the way before
1282 * we lock the parent. */
1283 err = nfserr_notdir;
1284 if(!dirp->i_op || !dirp->i_op->lookup)
1285 goto out;
1286 fh_lock_nested(fhp, I_MUTEX_PARENT);
1289 * Compose the response file handle.
1291 dchild = lookup_one_len(fname, dentry, flen);
1292 host_err = PTR_ERR(dchild);
1293 if (IS_ERR(dchild))
1294 goto out_nfserr;
1296 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1297 if (err)
1298 goto out;
1300 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1301 /* solaris7 gets confused (bugid 4218508) if these have
1302 * the high bit set, so just clear the high bits.
1304 v_mtime = verifier[0]&0x7fffffff;
1305 v_atime = verifier[1]&0x7fffffff;
1308 if (dchild->d_inode) {
1309 err = 0;
1311 switch (createmode) {
1312 case NFS3_CREATE_UNCHECKED:
1313 if (! S_ISREG(dchild->d_inode->i_mode))
1314 err = nfserr_exist;
1315 else if (truncp) {
1316 /* in nfsv4, we need to treat this case a little
1317 * differently. we don't want to truncate the
1318 * file now; this would be wrong if the OPEN
1319 * fails for some other reason. furthermore,
1320 * if the size is nonzero, we should ignore it
1321 * according to spec!
1323 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1325 else {
1326 iap->ia_valid &= ATTR_SIZE;
1327 goto set_attr;
1329 break;
1330 case NFS3_CREATE_EXCLUSIVE:
1331 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1332 && dchild->d_inode->i_atime.tv_sec == v_atime
1333 && dchild->d_inode->i_size == 0 )
1334 break;
1335 /* fallthru */
1336 case NFS3_CREATE_GUARDED:
1337 err = nfserr_exist;
1339 goto out;
1342 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1343 if (host_err < 0)
1344 goto out_nfserr;
1345 if (created)
1346 *created = 1;
1348 if (EX_ISSYNC(fhp->fh_export)) {
1349 err = nfserrno(nfsd_sync_dir(dentry));
1350 /* setattr will sync the child (or not) */
1353 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1354 /* Cram the verifier into atime/mtime */
1355 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1356 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1357 /* XXX someone who knows this better please fix it for nsec */
1358 iap->ia_mtime.tv_sec = v_mtime;
1359 iap->ia_atime.tv_sec = v_atime;
1360 iap->ia_mtime.tv_nsec = 0;
1361 iap->ia_atime.tv_nsec = 0;
1364 /* Set file attributes.
1365 * Irix appears to send along the gid when it tries to
1366 * implement setgid directories via NFS. Clear out all that cruft.
1368 set_attr:
1369 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1370 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1371 if (err2)
1372 err = err2;
1376 * Update the filehandle to get the new inode info.
1378 if (!err)
1379 err = fh_update(resfhp);
1381 out:
1382 fh_unlock(fhp);
1383 if (dchild && !IS_ERR(dchild))
1384 dput(dchild);
1385 return err;
1387 out_nfserr:
1388 err = nfserrno(host_err);
1389 goto out;
1391 #endif /* CONFIG_NFSD_V3 */
1394 * Read a symlink. On entry, *lenp must contain the maximum path length that
1395 * fits into the buffer. On return, it contains the true length.
1396 * N.B. After this call fhp needs an fh_put
1398 __be32
1399 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1401 struct dentry *dentry;
1402 struct inode *inode;
1403 mm_segment_t oldfs;
1404 __be32 err;
1405 int host_err;
1407 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1408 if (err)
1409 goto out;
1411 dentry = fhp->fh_dentry;
1412 inode = dentry->d_inode;
1414 err = nfserr_inval;
1415 if (!inode->i_op || !inode->i_op->readlink)
1416 goto out;
1418 touch_atime(fhp->fh_export->ex_mnt, dentry);
1419 /* N.B. Why does this call need a get_fs()??
1420 * Remove the set_fs and watch the fireworks:-) --okir
1423 oldfs = get_fs(); set_fs(KERNEL_DS);
1424 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1425 set_fs(oldfs);
1427 if (host_err < 0)
1428 goto out_nfserr;
1429 *lenp = host_err;
1430 err = 0;
1431 out:
1432 return err;
1434 out_nfserr:
1435 err = nfserrno(host_err);
1436 goto out;
1440 * Create a symlink and look up its inode
1441 * N.B. After this call _both_ fhp and resfhp need an fh_put
1443 __be32
1444 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1445 char *fname, int flen,
1446 char *path, int plen,
1447 struct svc_fh *resfhp,
1448 struct iattr *iap)
1450 struct dentry *dentry, *dnew;
1451 __be32 err, cerr;
1452 int host_err;
1453 umode_t mode;
1455 err = nfserr_noent;
1456 if (!flen || !plen)
1457 goto out;
1458 err = nfserr_exist;
1459 if (isdotent(fname, flen))
1460 goto out;
1462 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1463 if (err)
1464 goto out;
1465 fh_lock(fhp);
1466 dentry = fhp->fh_dentry;
1467 dnew = lookup_one_len(fname, dentry, flen);
1468 host_err = PTR_ERR(dnew);
1469 if (IS_ERR(dnew))
1470 goto out_nfserr;
1472 mode = S_IALLUGO;
1473 /* Only the MODE ATTRibute is even vaguely meaningful */
1474 if (iap && (iap->ia_valid & ATTR_MODE))
1475 mode = iap->ia_mode & S_IALLUGO;
1477 if (unlikely(path[plen] != 0)) {
1478 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1479 if (path_alloced == NULL)
1480 host_err = -ENOMEM;
1481 else {
1482 strncpy(path_alloced, path, plen);
1483 path_alloced[plen] = 0;
1484 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1485 kfree(path_alloced);
1487 } else
1488 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1490 if (!host_err) {
1491 if (EX_ISSYNC(fhp->fh_export))
1492 host_err = nfsd_sync_dir(dentry);
1494 err = nfserrno(host_err);
1495 fh_unlock(fhp);
1497 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1498 dput(dnew);
1499 if (err==0) err = cerr;
1500 out:
1501 return err;
1503 out_nfserr:
1504 err = nfserrno(host_err);
1505 goto out;
1509 * Create a hardlink
1510 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1512 __be32
1513 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1514 char *name, int len, struct svc_fh *tfhp)
1516 struct dentry *ddir, *dnew, *dold;
1517 struct inode *dirp, *dest;
1518 __be32 err;
1519 int host_err;
1521 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1522 if (err)
1523 goto out;
1524 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1525 if (err)
1526 goto out;
1528 err = nfserr_perm;
1529 if (!len)
1530 goto out;
1531 err = nfserr_exist;
1532 if (isdotent(name, len))
1533 goto out;
1535 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1536 ddir = ffhp->fh_dentry;
1537 dirp = ddir->d_inode;
1539 dnew = lookup_one_len(name, ddir, len);
1540 host_err = PTR_ERR(dnew);
1541 if (IS_ERR(dnew))
1542 goto out_nfserr;
1544 dold = tfhp->fh_dentry;
1545 dest = dold->d_inode;
1547 host_err = vfs_link(dold, dirp, dnew);
1548 if (!host_err) {
1549 if (EX_ISSYNC(ffhp->fh_export)) {
1550 err = nfserrno(nfsd_sync_dir(ddir));
1551 write_inode_now(dest, 1);
1553 err = 0;
1554 } else {
1555 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1556 err = nfserr_acces;
1557 else
1558 err = nfserrno(host_err);
1561 dput(dnew);
1562 out_unlock:
1563 fh_unlock(ffhp);
1564 out:
1565 return err;
1567 out_nfserr:
1568 err = nfserrno(host_err);
1569 goto out_unlock;
1573 * Rename a file
1574 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1576 __be32
1577 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1578 struct svc_fh *tfhp, char *tname, int tlen)
1580 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1581 struct inode *fdir, *tdir;
1582 __be32 err;
1583 int host_err;
1585 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1586 if (err)
1587 goto out;
1588 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1589 if (err)
1590 goto out;
1592 fdentry = ffhp->fh_dentry;
1593 fdir = fdentry->d_inode;
1595 tdentry = tfhp->fh_dentry;
1596 tdir = tdentry->d_inode;
1598 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1599 if (ffhp->fh_export != tfhp->fh_export)
1600 goto out;
1602 err = nfserr_perm;
1603 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1604 goto out;
1606 /* cannot use fh_lock as we need deadlock protective ordering
1607 * so do it by hand */
1608 trap = lock_rename(tdentry, fdentry);
1609 ffhp->fh_locked = tfhp->fh_locked = 1;
1610 fill_pre_wcc(ffhp);
1611 fill_pre_wcc(tfhp);
1613 odentry = lookup_one_len(fname, fdentry, flen);
1614 host_err = PTR_ERR(odentry);
1615 if (IS_ERR(odentry))
1616 goto out_nfserr;
1618 host_err = -ENOENT;
1619 if (!odentry->d_inode)
1620 goto out_dput_old;
1621 host_err = -EINVAL;
1622 if (odentry == trap)
1623 goto out_dput_old;
1625 ndentry = lookup_one_len(tname, tdentry, tlen);
1626 host_err = PTR_ERR(ndentry);
1627 if (IS_ERR(ndentry))
1628 goto out_dput_old;
1629 host_err = -ENOTEMPTY;
1630 if (ndentry == trap)
1631 goto out_dput_new;
1633 #ifdef MSNFS
1634 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1635 ((atomic_read(&odentry->d_count) > 1)
1636 || (atomic_read(&ndentry->d_count) > 1))) {
1637 host_err = -EPERM;
1638 } else
1639 #endif
1640 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1641 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1642 host_err = nfsd_sync_dir(tdentry);
1643 if (!host_err)
1644 host_err = nfsd_sync_dir(fdentry);
1647 out_dput_new:
1648 dput(ndentry);
1649 out_dput_old:
1650 dput(odentry);
1651 out_nfserr:
1652 err = nfserrno(host_err);
1654 /* we cannot reply on fh_unlock on the two filehandles,
1655 * as that would do the wrong thing if the two directories
1656 * were the same, so again we do it by hand
1658 fill_post_wcc(ffhp);
1659 fill_post_wcc(tfhp);
1660 unlock_rename(tdentry, fdentry);
1661 ffhp->fh_locked = tfhp->fh_locked = 0;
1663 out:
1664 return err;
1668 * Unlink a file or directory
1669 * N.B. After this call fhp needs an fh_put
1671 __be32
1672 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1673 char *fname, int flen)
1675 struct dentry *dentry, *rdentry;
1676 struct inode *dirp;
1677 __be32 err;
1678 int host_err;
1680 err = nfserr_acces;
1681 if (!flen || isdotent(fname, flen))
1682 goto out;
1683 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1684 if (err)
1685 goto out;
1687 fh_lock_nested(fhp, I_MUTEX_PARENT);
1688 dentry = fhp->fh_dentry;
1689 dirp = dentry->d_inode;
1691 rdentry = lookup_one_len(fname, dentry, flen);
1692 host_err = PTR_ERR(rdentry);
1693 if (IS_ERR(rdentry))
1694 goto out_nfserr;
1696 if (!rdentry->d_inode) {
1697 dput(rdentry);
1698 err = nfserr_noent;
1699 goto out;
1702 if (!type)
1703 type = rdentry->d_inode->i_mode & S_IFMT;
1705 if (type != S_IFDIR) { /* It's UNLINK */
1706 #ifdef MSNFS
1707 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1708 (atomic_read(&rdentry->d_count) > 1)) {
1709 host_err = -EPERM;
1710 } else
1711 #endif
1712 host_err = vfs_unlink(dirp, rdentry);
1713 } else { /* It's RMDIR */
1714 host_err = vfs_rmdir(dirp, rdentry);
1717 dput(rdentry);
1719 if (host_err)
1720 goto out_nfserr;
1721 if (EX_ISSYNC(fhp->fh_export))
1722 host_err = nfsd_sync_dir(dentry);
1724 out_nfserr:
1725 err = nfserrno(host_err);
1726 out:
1727 return err;
1731 * Read entries from a directory.
1732 * The NFSv3/4 verifier we ignore for now.
1734 __be32
1735 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1736 struct readdir_cd *cdp, filldir_t func)
1738 __be32 err;
1739 int host_err;
1740 struct file *file;
1741 loff_t offset = *offsetp;
1743 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1744 if (err)
1745 goto out;
1747 offset = vfs_llseek(file, offset, 0);
1748 if (offset < 0) {
1749 err = nfserrno((int)offset);
1750 goto out_close;
1754 * Read the directory entries. This silly loop is necessary because
1755 * readdir() is not guaranteed to fill up the entire buffer, but
1756 * may choose to do less.
1759 do {
1760 cdp->err = nfserr_eof; /* will be cleared on successful read */
1761 host_err = vfs_readdir(file, func, cdp);
1762 } while (host_err >=0 && cdp->err == nfs_ok);
1763 if (host_err)
1764 err = nfserrno(host_err);
1765 else
1766 err = cdp->err;
1767 *offsetp = vfs_llseek(file, 0, 1);
1769 if (err == nfserr_eof || err == nfserr_toosmall)
1770 err = nfs_ok; /* can still be found in ->err */
1771 out_close:
1772 nfsd_close(file);
1773 out:
1774 return err;
1778 * Get file system stats
1779 * N.B. After this call fhp needs an fh_put
1781 __be32
1782 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1784 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1785 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1786 err = nfserr_io;
1787 return err;
1791 * Check for a user's access permissions to this inode.
1793 __be32
1794 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1796 struct inode *inode = dentry->d_inode;
1797 int err;
1799 if (acc == MAY_NOP)
1800 return 0;
1801 #if 0
1802 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1803 acc,
1804 (acc & MAY_READ)? " read" : "",
1805 (acc & MAY_WRITE)? " write" : "",
1806 (acc & MAY_EXEC)? " exec" : "",
1807 (acc & MAY_SATTR)? " sattr" : "",
1808 (acc & MAY_TRUNC)? " trunc" : "",
1809 (acc & MAY_LOCK)? " lock" : "",
1810 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1811 inode->i_mode,
1812 IS_IMMUTABLE(inode)? " immut" : "",
1813 IS_APPEND(inode)? " append" : "",
1814 IS_RDONLY(inode)? " ro" : "");
1815 dprintk(" owner %d/%d user %d/%d\n",
1816 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1817 #endif
1819 /* Normally we reject any write/sattr etc access on a read-only file
1820 * system. But if it is IRIX doing check on write-access for a
1821 * device special file, we ignore rofs.
1823 if (!(acc & MAY_LOCAL_ACCESS))
1824 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1825 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1826 return nfserr_rofs;
1827 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1828 return nfserr_perm;
1830 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1831 return nfserr_perm;
1833 if (acc & MAY_LOCK) {
1834 /* If we cannot rely on authentication in NLM requests,
1835 * just allow locks, otherwise require read permission, or
1836 * ownership
1838 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1839 return 0;
1840 else
1841 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1844 * The file owner always gets access permission for accesses that
1845 * would normally be checked at open time. This is to make
1846 * file access work even when the client has done a fchmod(fd, 0).
1848 * However, `cp foo bar' should fail nevertheless when bar is
1849 * readonly. A sensible way to do this might be to reject all
1850 * attempts to truncate a read-only file, because a creat() call
1851 * always implies file truncation.
1852 * ... but this isn't really fair. A process may reasonably call
1853 * ftruncate on an open file descriptor on a file with perm 000.
1854 * We must trust the client to do permission checking - using "ACCESS"
1855 * with NFSv3.
1857 if ((acc & MAY_OWNER_OVERRIDE) &&
1858 inode->i_uid == current->fsuid)
1859 return 0;
1861 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1863 /* Allow read access to binaries even when mode 111 */
1864 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1865 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1866 err = permission(inode, MAY_EXEC, NULL);
1868 return err? nfserrno(err) : 0;
1871 void
1872 nfsd_racache_shutdown(void)
1874 if (!raparml)
1875 return;
1876 dprintk("nfsd: freeing readahead buffers.\n");
1877 kfree(raparml);
1878 raparml = NULL;
1881 * Initialize readahead param cache
1884 nfsd_racache_init(int cache_size)
1886 int i;
1887 int j = 0;
1888 int nperbucket;
1891 if (raparml)
1892 return 0;
1893 if (cache_size < 2*RAPARM_HASH_SIZE)
1894 cache_size = 2*RAPARM_HASH_SIZE;
1895 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1897 if (!raparml) {
1898 printk(KERN_WARNING
1899 "nfsd: Could not allocate memory read-ahead cache.\n");
1900 return -ENOMEM;
1903 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1904 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1905 raparm_hash[i].pb_head = NULL;
1906 spin_lock_init(&raparm_hash[i].pb_lock);
1908 nperbucket = cache_size >> RAPARM_HASH_BITS;
1909 for (i = 0; i < cache_size - 1; i++) {
1910 if (i % nperbucket == 0)
1911 raparm_hash[j++].pb_head = raparml + i;
1912 if (i % nperbucket < nperbucket-1)
1913 raparml[i].p_next = raparml + i + 1;
1916 nfsdstats.ra_size = cache_size;
1917 return 0;
1920 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1921 struct posix_acl *
1922 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1924 struct inode *inode = fhp->fh_dentry->d_inode;
1925 char *name;
1926 void *value = NULL;
1927 ssize_t size;
1928 struct posix_acl *acl;
1930 if (!IS_POSIXACL(inode))
1931 return ERR_PTR(-EOPNOTSUPP);
1933 switch (type) {
1934 case ACL_TYPE_ACCESS:
1935 name = POSIX_ACL_XATTR_ACCESS;
1936 break;
1937 case ACL_TYPE_DEFAULT:
1938 name = POSIX_ACL_XATTR_DEFAULT;
1939 break;
1940 default:
1941 return ERR_PTR(-EOPNOTSUPP);
1944 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1945 if (size < 0)
1946 return ERR_PTR(size);
1948 acl = posix_acl_from_xattr(value, size);
1949 kfree(value);
1950 return acl;
1954 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1956 struct inode *inode = fhp->fh_dentry->d_inode;
1957 char *name;
1958 void *value = NULL;
1959 size_t size;
1960 int error;
1962 if (!IS_POSIXACL(inode) || !inode->i_op ||
1963 !inode->i_op->setxattr || !inode->i_op->removexattr)
1964 return -EOPNOTSUPP;
1965 switch(type) {
1966 case ACL_TYPE_ACCESS:
1967 name = POSIX_ACL_XATTR_ACCESS;
1968 break;
1969 case ACL_TYPE_DEFAULT:
1970 name = POSIX_ACL_XATTR_DEFAULT;
1971 break;
1972 default:
1973 return -EOPNOTSUPP;
1976 if (acl && acl->a_count) {
1977 size = posix_acl_xattr_size(acl->a_count);
1978 value = kmalloc(size, GFP_KERNEL);
1979 if (!value)
1980 return -ENOMEM;
1981 error = posix_acl_to_xattr(acl, value, size);
1982 if (error < 0)
1983 goto getout;
1984 size = error;
1985 } else
1986 size = 0;
1988 if (size)
1989 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1990 else {
1991 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1992 error = 0;
1993 else {
1994 error = vfs_removexattr(fhp->fh_dentry, name);
1995 if (error == -ENODATA)
1996 error = 0;
2000 getout:
2001 kfree(value);
2002 return error;
2004 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */