1 #define MSNFS /* HACK HACK */
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>
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>
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>
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>
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
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
72 struct raparms *p_next;
77 struct file_ra_state p_ra;
78 unsigned int p_hindex;
81 struct raparm_hbucket {
82 struct raparms *pb_head;
84 } ____cacheline_aligned_in_smp;
86 static struct raparms * raparml;
87 #define RAPARM_HASH_BITS 4
88 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
93 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
95 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
96 * or nfs_ok having possibly changed *dpp and *expp
99 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 struct svc_export **expp)
102 struct svc_export *exp = *expp, *exp2 = NULL;
103 struct dentry *dentry = *dpp;
104 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
105 struct dentry *mounts = dget(dentry);
108 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
110 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
112 if (PTR_ERR(exp2) != -ENOENT)
118 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
119 /* successfully crossed mount point */
134 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
135 const char *name, unsigned int len,
136 struct svc_export **exp_ret, struct dentry **dentry_ret)
138 struct svc_export *exp;
139 struct dentry *dparent;
140 struct dentry *dentry;
144 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
146 /* Obtain dentry and export. */
147 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
151 dparent = fhp->fh_dentry;
152 exp = fhp->fh_export;
155 /* Lookup the name, but don't follow links */
156 if (isdotent(name, len)) {
158 dentry = dget(dparent);
159 else if (dparent != exp->ex_path.dentry)
160 dentry = dget_parent(dparent);
161 else if (!EX_NOHIDE(exp))
162 dentry = dget(dparent); /* .. == . just like at / */
164 /* checking mountpoint crossing is very different when stepping up */
165 struct svc_export *exp2 = NULL;
167 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
168 dentry = dget(dparent);
169 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
171 dp = dget_parent(dentry);
175 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
176 if (PTR_ERR(exp2) == -ENOENT) {
178 dentry = dget(dparent);
179 } else if (IS_ERR(exp2)) {
180 host_err = PTR_ERR(exp2);
192 dentry = lookup_one_len(name, dparent, len);
193 host_err = PTR_ERR(dentry);
197 * check if we have crossed a mount point ...
199 if (d_mountpoint(dentry)) {
200 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
206 *dentry_ret = dentry;
212 return nfserrno(host_err);
216 * Look up one component of a pathname.
217 * N.B. After this call _both_ fhp and resfh need an fh_put
219 * If the lookup would cross a mountpoint, and the mounted filesystem
220 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
221 * accepted as it stands and the mounted directory is
222 * returned. Otherwise the covered directory is returned.
223 * NOTE: this mountpoint crossing is not supported properly by all
224 * clients and is explicitly disallowed for NFSv3
225 * NeilBrown <neilb@cse.unsw.edu.au>
228 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
229 unsigned int len, struct svc_fh *resfh)
231 struct svc_export *exp;
232 struct dentry *dentry;
235 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
238 err = check_nfsd_access(exp, rqstp);
242 * Note: we compose the file handle now, but as the
243 * dentry may be negative, it may need to be updated.
245 err = fh_compose(resfh, exp, dentry, fhp);
246 if (!err && !dentry->d_inode)
256 * Set various file attributes.
257 * N.B. After this call fhp needs an fh_put
260 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
261 int check_guard, time_t guardtime)
263 struct dentry *dentry;
265 int accmode = MAY_SATTR;
272 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
273 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
274 if (iap->ia_valid & ATTR_SIZE)
278 err = fh_verify(rqstp, fhp, ftype, accmode);
282 dentry = fhp->fh_dentry;
283 inode = dentry->d_inode;
285 /* Ignore any mode updates on symlinks */
286 if (S_ISLNK(inode->i_mode))
287 iap->ia_valid &= ~ATTR_MODE;
293 * NFSv2 does not differentiate between "set-[ac]time-to-now"
294 * which only requires access, and "set-[ac]time-to-X" which
295 * requires ownership.
296 * So if it looks like it might be "set both to the same time which
297 * is close to now", and if inode_change_ok fails, then we
298 * convert to "set to now" instead of "set to explicit time"
300 * We only call inode_change_ok as the last test as technically
301 * it is not an interface that we should be using. It is only
302 * valid if the filesystem does not define it's own i_op->setattr.
304 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
305 #define MAX_TOUCH_TIME_ERROR (30*60)
306 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
307 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
311 * Now just make sure time is in the right ballpark.
312 * Solaris, at least, doesn't seem to care what the time
313 * request is. We require it be within 30 minutes of now.
315 time_t delta = iap->ia_atime.tv_sec - get_seconds();
318 if (delta < MAX_TOUCH_TIME_ERROR &&
319 inode_change_ok(inode, iap) != 0) {
321 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
322 * This will cause notify_change to set these times
325 iap->ia_valid &= ~BOTH_TIME_SET;
330 * The size case is special.
331 * It changes the file as well as the attributes.
333 if (iap->ia_valid & ATTR_SIZE) {
334 if (iap->ia_size < inode->i_size) {
335 err = nfsd_permission(rqstp, fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
341 * If we are changing the size of the file, then
342 * we need to break all leases.
344 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
345 if (host_err == -EWOULDBLOCK)
346 host_err = -ETIMEDOUT;
347 if (host_err) /* ENOMEM or EWOULDBLOCK */
350 host_err = get_write_access(inode);
355 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
357 put_write_access(inode);
363 imode = inode->i_mode;
364 if (iap->ia_valid & ATTR_MODE) {
365 iap->ia_mode &= S_IALLUGO;
366 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
367 /* if changing uid/gid revoke setuid/setgid in mode */
368 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) {
369 iap->ia_valid |= ATTR_KILL_PRIV;
370 iap->ia_mode &= ~S_ISUID;
372 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
373 iap->ia_mode &= ~S_ISGID;
376 * Revoke setuid/setgid bit on chown/chgrp
378 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
379 iap->ia_valid |= ATTR_KILL_SUID | ATTR_KILL_PRIV;
380 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
381 iap->ia_valid |= ATTR_KILL_SGID;
384 /* Change the attributes. */
386 iap->ia_valid |= ATTR_CTIME;
388 err = nfserr_notsync;
389 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
391 host_err = notify_change(dentry, iap);
392 err = nfserrno(host_err);
396 put_write_access(inode);
398 if (EX_ISSYNC(fhp->fh_export))
399 write_inode_now(inode, 1);
404 err = nfserrno(host_err);
408 #if defined(CONFIG_NFSD_V2_ACL) || \
409 defined(CONFIG_NFSD_V3_ACL) || \
410 defined(CONFIG_NFSD_V4)
411 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
415 buflen = vfs_getxattr(dentry, key, NULL, 0);
419 *buf = kmalloc(buflen, GFP_KERNEL);
423 return vfs_getxattr(dentry, key, *buf, buflen);
427 #if defined(CONFIG_NFSD_V4)
429 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
436 buflen = posix_acl_xattr_size(pacl->a_count);
437 buf = kmalloc(buflen, GFP_KERNEL);
442 len = posix_acl_to_xattr(pacl, buf, buflen);
448 error = vfs_setxattr(dentry, key, buf, len, 0);
455 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
456 struct nfs4_acl *acl)
460 struct dentry *dentry;
462 struct posix_acl *pacl = NULL, *dpacl = NULL;
463 unsigned int flags = 0;
466 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
470 dentry = fhp->fh_dentry;
471 inode = dentry->d_inode;
472 if (S_ISDIR(inode->i_mode))
473 flags = NFS4_ACL_DIR;
475 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
476 if (host_error == -EINVAL) {
477 return nfserr_attrnotsupp;
478 } else if (host_error < 0)
481 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
485 if (S_ISDIR(inode->i_mode))
486 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
489 posix_acl_release(pacl);
490 posix_acl_release(dpacl);
492 if (host_error == -EOPNOTSUPP)
493 return nfserr_attrnotsupp;
495 return nfserrno(host_error);
498 static struct posix_acl *
499 _get_posix_acl(struct dentry *dentry, char *key)
502 struct posix_acl *pacl = NULL;
505 buflen = nfsd_getxattr(dentry, key, &buf);
509 return ERR_PTR(buflen);
511 pacl = posix_acl_from_xattr(buf, buflen);
517 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
519 struct inode *inode = dentry->d_inode;
521 struct posix_acl *pacl = NULL, *dpacl = NULL;
522 unsigned int flags = 0;
524 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
525 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
526 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
528 error = PTR_ERR(pacl);
533 if (S_ISDIR(inode->i_mode)) {
534 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
535 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
537 else if (IS_ERR(dpacl)) {
538 error = PTR_ERR(dpacl);
542 flags = NFS4_ACL_DIR;
545 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
547 error = PTR_ERR(*acl);
551 posix_acl_release(pacl);
552 posix_acl_release(dpacl);
556 #endif /* defined(CONFIG_NFS_V4) */
558 #ifdef CONFIG_NFSD_V3
560 * Check server access rights to a file system object
566 static struct accessmap nfs3_regaccess[] = {
567 { NFS3_ACCESS_READ, MAY_READ },
568 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
569 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
570 { NFS3_ACCESS_EXTEND, MAY_WRITE },
575 static struct accessmap nfs3_diraccess[] = {
576 { NFS3_ACCESS_READ, MAY_READ },
577 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
578 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
579 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
580 { NFS3_ACCESS_DELETE, MAY_REMOVE },
585 static struct accessmap nfs3_anyaccess[] = {
586 /* Some clients - Solaris 2.6 at least, make an access call
587 * to the server to check for access for things like /dev/null
588 * (which really, the server doesn't care about). So
589 * We provide simple access checking for them, looking
590 * mainly at mode bits, and we make sure to ignore read-only
593 { NFS3_ACCESS_READ, MAY_READ },
594 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
595 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
596 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
602 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
604 struct accessmap *map;
605 struct svc_export *export;
606 struct dentry *dentry;
607 u32 query, result = 0, sresult = 0;
610 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
614 export = fhp->fh_export;
615 dentry = fhp->fh_dentry;
617 if (S_ISREG(dentry->d_inode->i_mode))
618 map = nfs3_regaccess;
619 else if (S_ISDIR(dentry->d_inode->i_mode))
620 map = nfs3_diraccess;
622 map = nfs3_anyaccess;
626 for (; map->access; map++) {
627 if (map->access & query) {
630 sresult |= map->access;
632 err2 = nfsd_permission(rqstp, export, dentry, map->how);
635 result |= map->access;
638 /* the following error codes just mean the access was not allowed,
639 * rather than an error occurred */
643 /* simply don't "or" in the access bit. */
653 *supported = sresult;
658 #endif /* CONFIG_NFSD_V3 */
663 * Open an existing file or directory.
664 * The access argument indicates the type of open (read/write/lock)
665 * N.B. After this call fhp needs an fh_put
668 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
669 int access, struct file **filp)
671 struct dentry *dentry;
673 int flags = O_RDONLY|O_LARGEFILE;
678 * If we get here, then the client has already done an "open",
679 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
680 * in case a chmod has now revoked permission.
682 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
686 dentry = fhp->fh_dentry;
687 inode = dentry->d_inode;
689 /* Disallow write access to files with the append-only bit set
690 * or any access when mandatory locking enabled
693 if (IS_APPEND(inode) && (access & MAY_WRITE))
696 * We must ignore files (but only files) which might have mandatory
697 * locks on them because there is no way to know if the accesser has
700 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
707 * Check to see if there are any leases on this file.
708 * This may block while leases are broken.
710 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
711 if (host_err == -EWOULDBLOCK)
712 host_err = -ETIMEDOUT;
713 if (host_err) /* NOMEM or WOULDBLOCK */
716 if (access & MAY_WRITE) {
717 if (access & MAY_READ)
718 flags = O_RDWR|O_LARGEFILE;
720 flags = O_WRONLY|O_LARGEFILE;
724 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
727 host_err = PTR_ERR(*filp);
729 err = nfserrno(host_err);
738 nfsd_close(struct file *filp)
745 * As this calls fsync (not fdatasync) there is no need for a write_inode
748 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
749 const struct file_operations *fop)
751 struct inode *inode = dp->d_inode;
752 int (*fsync) (struct file *, struct dentry *, int);
755 err = filemap_fdatawrite(inode->i_mapping);
756 if (err == 0 && fop && (fsync = fop->fsync))
757 err = fsync(filp, dp, 0);
759 err = filemap_fdatawait(inode->i_mapping);
766 nfsd_sync(struct file *filp)
769 struct inode *inode = filp->f_path.dentry->d_inode;
770 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
771 mutex_lock(&inode->i_mutex);
772 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
773 mutex_unlock(&inode->i_mutex);
779 nfsd_sync_dir(struct dentry *dp)
781 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
785 * Obtain the readahead parameters for the file
786 * specified by (dev, ino).
789 static inline struct raparms *
790 nfsd_get_raparms(dev_t dev, ino_t ino)
792 struct raparms *ra, **rap, **frap = NULL;
795 struct raparm_hbucket *rab;
797 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
798 rab = &raparm_hash[hash];
800 spin_lock(&rab->pb_lock);
801 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
802 if (ra->p_ino == ino && ra->p_dev == dev)
805 if (ra->p_count == 0)
808 depth = nfsdstats.ra_size*11/10;
810 spin_unlock(&rab->pb_lock);
820 if (rap != &rab->pb_head) {
822 ra->p_next = rab->pb_head;
826 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
827 spin_unlock(&rab->pb_lock);
832 * Grab and keep cached pages associated with a file in the svc_rqst
833 * so that they can be passed to the network sendmsg/sendpage routines
834 * directly. They will be released after the sending has completed.
837 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
838 struct splice_desc *sd)
840 struct svc_rqst *rqstp = sd->u.data;
841 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
842 struct page *page = buf->page;
846 ret = buf->ops->confirm(pipe, buf);
852 if (rqstp->rq_res.page_len == 0) {
857 rqstp->rq_res.page_base = buf->offset;
858 rqstp->rq_res.page_len = size;
859 } else if (page != pp[-1]) {
865 rqstp->rq_res.page_len += size;
867 rqstp->rq_res.page_len += size;
872 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
873 struct splice_desc *sd)
875 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
878 static inline int svc_msnfs(struct svc_fh *ffhp)
881 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
888 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
889 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
898 inode = file->f_path.dentry->d_inode;
900 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
903 /* Get readahead parameters */
904 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
907 file->f_ra = ra->p_ra;
909 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
910 struct splice_desc sd = {
917 rqstp->rq_resused = 1;
918 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
922 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
926 /* Write back readahead params */
928 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
929 spin_lock(&rab->pb_lock);
930 ra->p_ra = file->f_ra;
933 spin_unlock(&rab->pb_lock);
937 nfsdstats.io_read += host_err;
940 fsnotify_access(file->f_path.dentry);
942 err = nfserrno(host_err);
947 static void kill_suid(struct dentry *dentry)
950 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
952 mutex_lock(&dentry->d_inode->i_mutex);
953 notify_change(dentry, &ia);
954 mutex_unlock(&dentry->d_inode->i_mutex);
958 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
959 loff_t offset, struct kvec *vec, int vlen,
960 unsigned long cnt, int *stablep)
962 struct svc_export *exp;
963 struct dentry *dentry;
968 int stable = *stablep;
973 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
974 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
978 dentry = file->f_path.dentry;
979 inode = dentry->d_inode;
980 exp = fhp->fh_export;
983 * Request sync writes if
984 * - the sync export option has been set, or
985 * - the client requested O_SYNC behavior (NFSv3 feature).
986 * - The file system doesn't support fsync().
987 * When gathered writes have been configured for this volume,
988 * flushing the data to disk is handled separately below.
991 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
993 *stablep = 2; /* FILE_SYNC */
998 if (stable && !EX_WGATHER(exp))
999 file->f_flags |= O_SYNC;
1001 /* Write the data. */
1002 oldfs = get_fs(); set_fs(KERNEL_DS);
1003 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1005 if (host_err >= 0) {
1006 nfsdstats.io_write += cnt;
1007 fsnotify_modify(file->f_path.dentry);
1010 /* clear setuid/setgid flag after write */
1011 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1014 if (host_err >= 0 && stable) {
1015 static ino_t last_ino;
1016 static dev_t last_dev;
1019 * Gathered writes: If another process is currently
1020 * writing to the file, there's a high chance
1021 * this is another nfsd (triggered by a bulk write
1022 * from a client's biod). Rather than syncing the
1023 * file with each write request, we sleep for 10 msec.
1025 * I don't know if this roughly approximates
1026 * C. Juszak's idea of gathered writes, but it's a
1027 * nice and simple solution (IMHO), and it seems to
1030 if (EX_WGATHER(exp)) {
1031 if (atomic_read(&inode->i_writecount) > 1
1032 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1033 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1035 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1038 if (inode->i_state & I_DIRTY) {
1039 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1040 host_err=nfsd_sync(file);
1043 wake_up(&inode->i_wait);
1046 last_ino = inode->i_ino;
1047 last_dev = inode->i_sb->s_dev;
1050 dprintk("nfsd: write complete host_err=%d\n", host_err);
1054 err = nfserrno(host_err);
1060 * Read data from a file. count must contain the requested read count
1061 * on entry. On return, *count contains the number of bytes actually read.
1062 * N.B. After this call fhp needs an fh_put
1065 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1066 loff_t offset, struct kvec *vec, int vlen,
1067 unsigned long *count)
1072 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1073 MAY_READ|MAY_OWNER_OVERRIDE);
1076 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1078 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1081 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1089 * Write data to a file.
1090 * The stable flag requests synchronous writes.
1091 * N.B. After this call fhp needs an fh_put
1094 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1095 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1101 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1102 MAY_WRITE|MAY_OWNER_OVERRIDE);
1105 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1108 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1113 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1121 #ifdef CONFIG_NFSD_V3
1123 * Commit all pending writes to stable storage.
1124 * Strictly speaking, we could sync just the indicated file region here,
1125 * but there's currently no way we can ask the VFS to do so.
1127 * Unfortunately we cannot lock the file to make sure we return full WCC
1128 * data to the client, as locking happens lower down in the filesystem.
1131 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1132 loff_t offset, unsigned long count)
1137 if ((u64)count > ~(u64)offset)
1138 return nfserr_inval;
1140 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1142 if (EX_ISSYNC(fhp->fh_export)) {
1143 if (file->f_op && file->f_op->fsync) {
1144 err = nfserrno(nfsd_sync(file));
1146 err = nfserr_notsupp;
1153 #endif /* CONFIG_NFSD_V3 */
1156 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1160 * Mode has already been set earlier in create:
1162 iap->ia_valid &= ~ATTR_MODE;
1164 * Setting uid/gid works only for root. Irix appears to
1165 * send along the gid on create when it tries to implement
1166 * setgid directories via NFS:
1168 if (current->fsuid != 0)
1169 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1171 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1176 * Create a file (regular, directory, device, fifo); UNIX sockets
1177 * not yet implemented.
1178 * If the response fh has been verified, the parent directory should
1179 * already be locked. Note that the parent directory is left locked.
1181 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1184 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1185 char *fname, int flen, struct iattr *iap,
1186 int type, dev_t rdev, struct svc_fh *resfhp)
1188 struct dentry *dentry, *dchild = NULL;
1198 if (isdotent(fname, flen))
1201 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1205 dentry = fhp->fh_dentry;
1206 dirp = dentry->d_inode;
1208 err = nfserr_notdir;
1209 if(!dirp->i_op || !dirp->i_op->lookup)
1212 * Check whether the response file handle has been verified yet.
1213 * If it has, the parent directory should already be locked.
1215 if (!resfhp->fh_dentry) {
1216 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1217 fh_lock_nested(fhp, I_MUTEX_PARENT);
1218 dchild = lookup_one_len(fname, dentry, flen);
1219 host_err = PTR_ERR(dchild);
1222 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1226 /* called from nfsd_proc_create */
1227 dchild = dget(resfhp->fh_dentry);
1228 if (!fhp->fh_locked) {
1229 /* not actually possible */
1231 "nfsd_create: parent %s/%s not locked!\n",
1232 dentry->d_parent->d_name.name,
1233 dentry->d_name.name);
1239 * Make sure the child dentry is still negative ...
1242 if (dchild->d_inode) {
1243 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1244 dentry->d_name.name, dchild->d_name.name);
1248 if (!(iap->ia_valid & ATTR_MODE))
1250 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1253 * Get the dir op function pointer.
1258 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1261 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1264 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1267 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1273 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1276 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1279 printk("nfsd: bad file type %o in nfsd_create\n", type);
1284 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1288 if (EX_ISSYNC(fhp->fh_export)) {
1289 err = nfserrno(nfsd_sync_dir(dentry));
1290 write_inode_now(dchild->d_inode, 1);
1294 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1297 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1299 * Update the file handle to get the new inode info.
1302 err = fh_update(resfhp);
1304 if (dchild && !IS_ERR(dchild))
1309 err = nfserrno(host_err);
1313 #ifdef CONFIG_NFSD_V3
1315 * NFSv3 version of nfsd_create
1318 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1319 char *fname, int flen, struct iattr *iap,
1320 struct svc_fh *resfhp, int createmode, u32 *verifier,
1321 int *truncp, int *created)
1323 struct dentry *dentry, *dchild = NULL;
1328 __u32 v_mtime=0, v_atime=0;
1334 if (isdotent(fname, flen))
1336 if (!(iap->ia_valid & ATTR_MODE))
1338 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1342 dentry = fhp->fh_dentry;
1343 dirp = dentry->d_inode;
1345 /* Get all the sanity checks out of the way before
1346 * we lock the parent. */
1347 err = nfserr_notdir;
1348 if(!dirp->i_op || !dirp->i_op->lookup)
1350 fh_lock_nested(fhp, I_MUTEX_PARENT);
1353 * Compose the response file handle.
1355 dchild = lookup_one_len(fname, dentry, flen);
1356 host_err = PTR_ERR(dchild);
1360 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1364 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1365 /* solaris7 gets confused (bugid 4218508) if these have
1366 * the high bit set, so just clear the high bits. If this is
1367 * ever changed to use different attrs for storing the
1368 * verifier, then do_open_lookup() will also need to be fixed
1371 v_mtime = verifier[0]&0x7fffffff;
1372 v_atime = verifier[1]&0x7fffffff;
1375 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1378 if (dchild->d_inode) {
1381 switch (createmode) {
1382 case NFS3_CREATE_UNCHECKED:
1383 if (! S_ISREG(dchild->d_inode->i_mode))
1386 /* in nfsv4, we need to treat this case a little
1387 * differently. we don't want to truncate the
1388 * file now; this would be wrong if the OPEN
1389 * fails for some other reason. furthermore,
1390 * if the size is nonzero, we should ignore it
1391 * according to spec!
1393 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1396 iap->ia_valid &= ATTR_SIZE;
1400 case NFS3_CREATE_EXCLUSIVE:
1401 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1402 && dchild->d_inode->i_atime.tv_sec == v_atime
1403 && dchild->d_inode->i_size == 0 )
1406 case NFS3_CREATE_GUARDED:
1409 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1413 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1415 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1421 if (EX_ISSYNC(fhp->fh_export)) {
1422 err = nfserrno(nfsd_sync_dir(dentry));
1423 /* setattr will sync the child (or not) */
1426 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1427 /* Cram the verifier into atime/mtime */
1428 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1429 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1430 /* XXX someone who knows this better please fix it for nsec */
1431 iap->ia_mtime.tv_sec = v_mtime;
1432 iap->ia_atime.tv_sec = v_atime;
1433 iap->ia_mtime.tv_nsec = 0;
1434 iap->ia_atime.tv_nsec = 0;
1438 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1442 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1444 * Update the filehandle to get the new inode info.
1447 err = fh_update(resfhp);
1451 if (dchild && !IS_ERR(dchild))
1456 err = nfserrno(host_err);
1459 #endif /* CONFIG_NFSD_V3 */
1462 * Read a symlink. On entry, *lenp must contain the maximum path length that
1463 * fits into the buffer. On return, it contains the true length.
1464 * N.B. After this call fhp needs an fh_put
1467 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1469 struct dentry *dentry;
1470 struct inode *inode;
1475 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1479 dentry = fhp->fh_dentry;
1480 inode = dentry->d_inode;
1483 if (!inode->i_op || !inode->i_op->readlink)
1486 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1487 /* N.B. Why does this call need a get_fs()??
1488 * Remove the set_fs and watch the fireworks:-) --okir
1491 oldfs = get_fs(); set_fs(KERNEL_DS);
1492 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1503 err = nfserrno(host_err);
1508 * Create a symlink and look up its inode
1509 * N.B. After this call _both_ fhp and resfhp need an fh_put
1512 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1513 char *fname, int flen,
1514 char *path, int plen,
1515 struct svc_fh *resfhp,
1518 struct dentry *dentry, *dnew;
1527 if (isdotent(fname, flen))
1530 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1534 dentry = fhp->fh_dentry;
1535 dnew = lookup_one_len(fname, dentry, flen);
1536 host_err = PTR_ERR(dnew);
1541 /* Only the MODE ATTRibute is even vaguely meaningful */
1542 if (iap && (iap->ia_valid & ATTR_MODE))
1543 mode = iap->ia_mode & S_IALLUGO;
1545 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1549 if (unlikely(path[plen] != 0)) {
1550 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1551 if (path_alloced == NULL)
1554 strncpy(path_alloced, path, plen);
1555 path_alloced[plen] = 0;
1556 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1557 kfree(path_alloced);
1560 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1563 if (EX_ISSYNC(fhp->fh_export))
1564 host_err = nfsd_sync_dir(dentry);
1566 err = nfserrno(host_err);
1569 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1571 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1573 if (err==0) err = cerr;
1578 err = nfserrno(host_err);
1584 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1587 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1588 char *name, int len, struct svc_fh *tfhp)
1590 struct dentry *ddir, *dnew, *dold;
1591 struct inode *dirp, *dest;
1595 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1598 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1606 if (isdotent(name, len))
1609 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1610 ddir = ffhp->fh_dentry;
1611 dirp = ddir->d_inode;
1613 dnew = lookup_one_len(name, ddir, len);
1614 host_err = PTR_ERR(dnew);
1618 dold = tfhp->fh_dentry;
1619 dest = dold->d_inode;
1621 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1623 err = nfserrno(host_err);
1626 host_err = vfs_link(dold, dirp, dnew);
1628 if (EX_ISSYNC(ffhp->fh_export)) {
1629 err = nfserrno(nfsd_sync_dir(ddir));
1630 write_inode_now(dest, 1);
1634 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1637 err = nfserrno(host_err);
1639 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1648 err = nfserrno(host_err);
1654 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1657 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1658 struct svc_fh *tfhp, char *tname, int tlen)
1660 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1661 struct inode *fdir, *tdir;
1665 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1668 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1672 fdentry = ffhp->fh_dentry;
1673 fdir = fdentry->d_inode;
1675 tdentry = tfhp->fh_dentry;
1676 tdir = tdentry->d_inode;
1678 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1679 if (ffhp->fh_export != tfhp->fh_export)
1683 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1686 /* cannot use fh_lock as we need deadlock protective ordering
1687 * so do it by hand */
1688 trap = lock_rename(tdentry, fdentry);
1689 ffhp->fh_locked = tfhp->fh_locked = 1;
1693 odentry = lookup_one_len(fname, fdentry, flen);
1694 host_err = PTR_ERR(odentry);
1695 if (IS_ERR(odentry))
1699 if (!odentry->d_inode)
1702 if (odentry == trap)
1705 ndentry = lookup_one_len(tname, tdentry, tlen);
1706 host_err = PTR_ERR(ndentry);
1707 if (IS_ERR(ndentry))
1709 host_err = -ENOTEMPTY;
1710 if (ndentry == trap)
1713 if (svc_msnfs(ffhp) &&
1714 ((atomic_read(&odentry->d_count) > 1)
1715 || (atomic_read(&ndentry->d_count) > 1))) {
1721 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1723 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1727 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1728 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1729 host_err = nfsd_sync_dir(tdentry);
1731 host_err = nfsd_sync_dir(fdentry);
1734 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1741 err = nfserrno(host_err);
1743 /* we cannot reply on fh_unlock on the two filehandles,
1744 * as that would do the wrong thing if the two directories
1745 * were the same, so again we do it by hand
1747 fill_post_wcc(ffhp);
1748 fill_post_wcc(tfhp);
1749 unlock_rename(tdentry, fdentry);
1750 ffhp->fh_locked = tfhp->fh_locked = 0;
1757 * Unlink a file or directory
1758 * N.B. After this call fhp needs an fh_put
1761 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1762 char *fname, int flen)
1764 struct dentry *dentry, *rdentry;
1770 if (!flen || isdotent(fname, flen))
1772 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1776 fh_lock_nested(fhp, I_MUTEX_PARENT);
1777 dentry = fhp->fh_dentry;
1778 dirp = dentry->d_inode;
1780 rdentry = lookup_one_len(fname, dentry, flen);
1781 host_err = PTR_ERR(rdentry);
1782 if (IS_ERR(rdentry))
1785 if (!rdentry->d_inode) {
1792 type = rdentry->d_inode->i_mode & S_IFMT;
1794 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1798 if (type != S_IFDIR) { /* It's UNLINK */
1800 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1801 (atomic_read(&rdentry->d_count) > 1)) {
1805 host_err = vfs_unlink(dirp, rdentry);
1806 } else { /* It's RMDIR */
1807 host_err = vfs_rmdir(dirp, rdentry);
1814 if (EX_ISSYNC(fhp->fh_export))
1815 host_err = nfsd_sync_dir(dentry);
1818 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1820 err = nfserrno(host_err);
1826 * Read entries from a directory.
1827 * The NFSv3/4 verifier we ignore for now.
1830 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1831 struct readdir_cd *cdp, filldir_t func)
1836 loff_t offset = *offsetp;
1838 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1842 offset = vfs_llseek(file, offset, 0);
1844 err = nfserrno((int)offset);
1849 * Read the directory entries. This silly loop is necessary because
1850 * readdir() is not guaranteed to fill up the entire buffer, but
1851 * may choose to do less.
1855 cdp->err = nfserr_eof; /* will be cleared on successful read */
1856 host_err = vfs_readdir(file, func, cdp);
1857 } while (host_err >=0 && cdp->err == nfs_ok);
1859 err = nfserrno(host_err);
1862 *offsetp = vfs_llseek(file, 0, 1);
1864 if (err == nfserr_eof || err == nfserr_toosmall)
1865 err = nfs_ok; /* can still be found in ->err */
1873 * Get file system stats
1874 * N.B. After this call fhp needs an fh_put
1877 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1879 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1880 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1885 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1887 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1891 * Check for a user's access permissions to this inode.
1894 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1895 struct dentry *dentry, int acc)
1897 struct inode *inode = dentry->d_inode;
1903 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1905 (acc & MAY_READ)? " read" : "",
1906 (acc & MAY_WRITE)? " write" : "",
1907 (acc & MAY_EXEC)? " exec" : "",
1908 (acc & MAY_SATTR)? " sattr" : "",
1909 (acc & MAY_TRUNC)? " trunc" : "",
1910 (acc & MAY_LOCK)? " lock" : "",
1911 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1913 IS_IMMUTABLE(inode)? " immut" : "",
1914 IS_APPEND(inode)? " append" : "",
1915 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1916 dprintk(" owner %d/%d user %d/%d\n",
1917 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1920 /* Normally we reject any write/sattr etc access on a read-only file
1921 * system. But if it is IRIX doing check on write-access for a
1922 * device special file, we ignore rofs.
1924 if (!(acc & MAY_LOCAL_ACCESS))
1925 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1926 if (exp_rdonly(rqstp, exp) ||
1927 __mnt_is_readonly(exp->ex_path.mnt))
1929 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1932 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1935 if (acc & MAY_LOCK) {
1936 /* If we cannot rely on authentication in NLM requests,
1937 * just allow locks, otherwise require read permission, or
1940 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1943 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1946 * The file owner always gets access permission for accesses that
1947 * would normally be checked at open time. This is to make
1948 * file access work even when the client has done a fchmod(fd, 0).
1950 * However, `cp foo bar' should fail nevertheless when bar is
1951 * readonly. A sensible way to do this might be to reject all
1952 * attempts to truncate a read-only file, because a creat() call
1953 * always implies file truncation.
1954 * ... but this isn't really fair. A process may reasonably call
1955 * ftruncate on an open file descriptor on a file with perm 000.
1956 * We must trust the client to do permission checking - using "ACCESS"
1959 if ((acc & MAY_OWNER_OVERRIDE) &&
1960 inode->i_uid == current->fsuid)
1963 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1965 /* Allow read access to binaries even when mode 111 */
1966 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1967 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1968 err = permission(inode, MAY_EXEC, NULL);
1970 return err? nfserrno(err) : 0;
1974 nfsd_racache_shutdown(void)
1978 dprintk("nfsd: freeing readahead buffers.\n");
1983 * Initialize readahead param cache
1986 nfsd_racache_init(int cache_size)
1995 if (cache_size < 2*RAPARM_HASH_SIZE)
1996 cache_size = 2*RAPARM_HASH_SIZE;
1997 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
2001 "nfsd: Could not allocate memory read-ahead cache.\n");
2005 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2006 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
2007 raparm_hash[i].pb_head = NULL;
2008 spin_lock_init(&raparm_hash[i].pb_lock);
2010 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2011 for (i = 0; i < cache_size - 1; i++) {
2012 if (i % nperbucket == 0)
2013 raparm_hash[j++].pb_head = raparml + i;
2014 if (i % nperbucket < nperbucket-1)
2015 raparml[i].p_next = raparml + i + 1;
2018 nfsdstats.ra_size = cache_size;
2022 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2024 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2026 struct inode *inode = fhp->fh_dentry->d_inode;
2030 struct posix_acl *acl;
2032 if (!IS_POSIXACL(inode))
2033 return ERR_PTR(-EOPNOTSUPP);
2036 case ACL_TYPE_ACCESS:
2037 name = POSIX_ACL_XATTR_ACCESS;
2039 case ACL_TYPE_DEFAULT:
2040 name = POSIX_ACL_XATTR_DEFAULT;
2043 return ERR_PTR(-EOPNOTSUPP);
2046 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2048 return ERR_PTR(size);
2050 acl = posix_acl_from_xattr(value, size);
2056 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2058 struct inode *inode = fhp->fh_dentry->d_inode;
2064 if (!IS_POSIXACL(inode) || !inode->i_op ||
2065 !inode->i_op->setxattr || !inode->i_op->removexattr)
2068 case ACL_TYPE_ACCESS:
2069 name = POSIX_ACL_XATTR_ACCESS;
2071 case ACL_TYPE_DEFAULT:
2072 name = POSIX_ACL_XATTR_DEFAULT;
2078 if (acl && acl->a_count) {
2079 size = posix_acl_xattr_size(acl->a_count);
2080 value = kmalloc(size, GFP_KERNEL);
2083 error = posix_acl_to_xattr(acl, value, size);
2090 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2094 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2096 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2099 error = vfs_removexattr(fhp->fh_dentry, name);
2100 if (error == -ENODATA)
2104 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2110 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */