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
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
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.
78 struct raparms *p_next;
83 struct file_ra_state p_ra;
84 unsigned int p_hindex;
87 struct raparm_hbucket {
88 struct raparms *pb_head;
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];
99 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
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);
114 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
116 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
123 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
124 /* successfully crossed mount point */
130 if (exp2) exp_put(exp2);
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>
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;
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);
167 dparent = fhp->fh_dentry;
168 exp = fhp->fh_export;
171 /* Lookup the name, but don't follow links */
172 if (isdotent(name, len)) {
174 dentry = dget(dparent);
175 else if (dparent != exp->ex_dentry) {
176 dentry = dget_parent(dparent);
177 } else if (!EX_NOHIDE(exp))
178 dentry = dget(dparent); /* .. == . just like at / */
180 /* checking mountpoint crossing is very different when stepping up */
181 struct svc_export *exp2 = NULL;
183 struct vfsmount *mnt = mntget(exp->ex_mnt);
184 dentry = dget(dparent);
185 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
187 dp = dget_parent(dentry);
191 exp2 = exp_parent(exp->ex_client, mnt, dentry,
193 if (PTR_ERR(exp2) == -ENOENT) {
195 dentry = dget(dparent);
196 } else if (IS_ERR(exp2)) {
197 host_err = PTR_ERR(exp2);
209 dentry = lookup_one_len(name, dparent, len);
210 host_err = PTR_ERR(dentry);
214 * check if we have crossed a mount point ...
216 if (d_mountpoint(dentry)) {
217 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
224 * Note: we compose the file handle now, but as the
225 * dentry may be negative, it may need to be updated.
227 err = fh_compose(resfh, exp, dentry, fhp);
228 if (!err && !dentry->d_inode)
236 err = nfserrno(host_err);
241 * Set various file attributes.
242 * N.B. After this call fhp needs an fh_put
245 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
246 int check_guard, time_t guardtime)
248 struct dentry *dentry;
250 int accmode = MAY_SATTR;
257 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
258 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
259 if (iap->ia_valid & ATTR_SIZE)
263 err = fh_verify(rqstp, fhp, ftype, accmode);
267 dentry = fhp->fh_dentry;
268 inode = dentry->d_inode;
270 /* Ignore any mode updates on symlinks */
271 if (S_ISLNK(inode->i_mode))
272 iap->ia_valid &= ~ATTR_MODE;
277 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
278 * which only requires access, and "set-[ac]time-to-X" which
279 * requires ownership.
280 * So if it looks like it might be "set both to the same time which
281 * is close to now", and if inode_change_ok fails, then we
282 * convert to "set to now" instead of "set to explicit time"
284 * We only call inode_change_ok as the last test as technically
285 * it is not an interface that we should be using. It is only
286 * valid if the filesystem does not define it's own i_op->setattr.
288 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
289 #define MAX_TOUCH_TIME_ERROR (30*60)
290 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
291 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
293 /* Looks probable. Now just make sure time is in the right ballpark.
294 * Solaris, at least, doesn't seem to care what the time request is.
295 * We require it be within 30 minutes of now.
297 time_t delta = iap->ia_atime.tv_sec - get_seconds();
298 if (delta<0) delta = -delta;
299 if (delta < MAX_TOUCH_TIME_ERROR &&
300 inode_change_ok(inode, iap) != 0) {
301 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
302 * this will cause notify_change to set these times to "now"
304 iap->ia_valid &= ~BOTH_TIME_SET;
308 /* The size case is special. It changes the file as well as the attributes. */
309 if (iap->ia_valid & ATTR_SIZE) {
310 if (iap->ia_size < inode->i_size) {
311 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
317 * If we are changing the size of the file, then
318 * we need to break all leases.
320 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
321 if (host_err == -EWOULDBLOCK)
322 host_err = -ETIMEDOUT;
323 if (host_err) /* ENOMEM or EWOULDBLOCK */
326 host_err = get_write_access(inode);
331 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
333 put_write_access(inode);
339 imode = inode->i_mode;
340 if (iap->ia_valid & ATTR_MODE) {
341 iap->ia_mode &= S_IALLUGO;
342 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
345 /* Revoke setuid/setgid bit on chown/chgrp */
346 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
347 iap->ia_valid |= ATTR_KILL_SUID;
348 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
349 iap->ia_valid |= ATTR_KILL_SGID;
351 /* Change the attributes. */
353 iap->ia_valid |= ATTR_CTIME;
355 err = nfserr_notsync;
356 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
358 host_err = notify_change(dentry, iap);
359 err = nfserrno(host_err);
363 put_write_access(inode);
365 if (EX_ISSYNC(fhp->fh_export))
366 write_inode_now(inode, 1);
371 err = nfserrno(host_err);
375 #if defined(CONFIG_NFSD_V2_ACL) || \
376 defined(CONFIG_NFSD_V3_ACL) || \
377 defined(CONFIG_NFSD_V4)
378 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
382 buflen = vfs_getxattr(dentry, key, NULL, 0);
386 *buf = kmalloc(buflen, GFP_KERNEL);
390 return vfs_getxattr(dentry, key, *buf, buflen);
394 #if defined(CONFIG_NFSD_V4)
396 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
403 buflen = posix_acl_xattr_size(pacl->a_count);
404 buf = kmalloc(buflen, GFP_KERNEL);
409 len = posix_acl_to_xattr(pacl, buf, buflen);
415 error = vfs_setxattr(dentry, key, buf, len, 0);
422 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
423 struct nfs4_acl *acl)
427 struct dentry *dentry;
429 struct posix_acl *pacl = NULL, *dpacl = NULL;
430 unsigned int flags = 0;
433 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
437 dentry = fhp->fh_dentry;
438 inode = dentry->d_inode;
439 if (S_ISDIR(inode->i_mode))
440 flags = NFS4_ACL_DIR;
442 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
443 if (host_error == -EINVAL) {
444 return nfserr_attrnotsupp;
445 } else if (host_error < 0)
448 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
452 if (S_ISDIR(inode->i_mode))
453 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
456 posix_acl_release(pacl);
457 posix_acl_release(dpacl);
459 if (host_error == -EOPNOTSUPP)
460 return nfserr_attrnotsupp;
462 return nfserrno(host_error);
465 static struct posix_acl *
466 _get_posix_acl(struct dentry *dentry, char *key)
469 struct posix_acl *pacl = NULL;
472 buflen = nfsd_getxattr(dentry, key, &buf);
476 return ERR_PTR(buflen);
478 pacl = posix_acl_from_xattr(buf, buflen);
484 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
486 struct inode *inode = dentry->d_inode;
488 struct posix_acl *pacl = NULL, *dpacl = NULL;
489 unsigned int flags = 0;
491 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
492 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
493 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
495 error = PTR_ERR(pacl);
500 if (S_ISDIR(inode->i_mode)) {
501 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
502 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
504 else if (IS_ERR(dpacl)) {
505 error = PTR_ERR(dpacl);
509 flags = NFS4_ACL_DIR;
512 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
514 error = PTR_ERR(*acl);
518 posix_acl_release(pacl);
519 posix_acl_release(dpacl);
523 #endif /* defined(CONFIG_NFS_V4) */
525 #ifdef CONFIG_NFSD_V3
527 * Check server access rights to a file system object
533 static struct accessmap nfs3_regaccess[] = {
534 { NFS3_ACCESS_READ, MAY_READ },
535 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
536 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
537 { NFS3_ACCESS_EXTEND, MAY_WRITE },
542 static struct accessmap nfs3_diraccess[] = {
543 { NFS3_ACCESS_READ, MAY_READ },
544 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
545 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
546 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
547 { NFS3_ACCESS_DELETE, MAY_REMOVE },
552 static struct accessmap nfs3_anyaccess[] = {
553 /* Some clients - Solaris 2.6 at least, make an access call
554 * to the server to check for access for things like /dev/null
555 * (which really, the server doesn't care about). So
556 * We provide simple access checking for them, looking
557 * mainly at mode bits, and we make sure to ignore read-only
560 { NFS3_ACCESS_READ, MAY_READ },
561 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
562 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
563 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
569 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
571 struct accessmap *map;
572 struct svc_export *export;
573 struct dentry *dentry;
574 u32 query, result = 0, sresult = 0;
577 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
581 export = fhp->fh_export;
582 dentry = fhp->fh_dentry;
584 if (S_ISREG(dentry->d_inode->i_mode))
585 map = nfs3_regaccess;
586 else if (S_ISDIR(dentry->d_inode->i_mode))
587 map = nfs3_diraccess;
589 map = nfs3_anyaccess;
593 for (; map->access; map++) {
594 if (map->access & query) {
597 sresult |= map->access;
599 err2 = nfsd_permission(export, dentry, map->how);
602 result |= map->access;
605 /* the following error codes just mean the access was not allowed,
606 * rather than an error occurred */
610 /* simply don't "or" in the access bit. */
620 *supported = sresult;
625 #endif /* CONFIG_NFSD_V3 */
630 * Open an existing file or directory.
631 * The access argument indicates the type of open (read/write/lock)
632 * N.B. After this call fhp needs an fh_put
635 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
636 int access, struct file **filp)
638 struct dentry *dentry;
640 int flags = O_RDONLY|O_LARGEFILE;
645 * If we get here, then the client has already done an "open",
646 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
647 * in case a chmod has now revoked permission.
649 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
653 dentry = fhp->fh_dentry;
654 inode = dentry->d_inode;
656 /* Disallow write access to files with the append-only bit set
657 * or any access when mandatory locking enabled
660 if (IS_APPEND(inode) && (access & MAY_WRITE))
662 if (IS_ISMNDLK(inode))
669 * Check to see if there are any leases on this file.
670 * This may block while leases are broken.
672 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
673 if (host_err == -EWOULDBLOCK)
674 host_err = -ETIMEDOUT;
675 if (host_err) /* NOMEM or WOULDBLOCK */
678 if (access & MAY_WRITE) {
679 if (access & MAY_READ)
680 flags = O_RDWR|O_LARGEFILE;
682 flags = O_WRONLY|O_LARGEFILE;
686 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
688 host_err = PTR_ERR(*filp);
690 err = nfserrno(host_err);
699 nfsd_close(struct file *filp)
706 * As this calls fsync (not fdatasync) there is no need for a write_inode
709 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
710 const struct file_operations *fop)
712 struct inode *inode = dp->d_inode;
713 int (*fsync) (struct file *, struct dentry *, int);
716 err = filemap_fdatawrite(inode->i_mapping);
717 if (err == 0 && fop && (fsync = fop->fsync))
718 err = fsync(filp, dp, 0);
720 err = filemap_fdatawait(inode->i_mapping);
727 nfsd_sync(struct file *filp)
730 struct inode *inode = filp->f_path.dentry->d_inode;
731 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
732 mutex_lock(&inode->i_mutex);
733 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
734 mutex_unlock(&inode->i_mutex);
740 nfsd_sync_dir(struct dentry *dp)
742 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
746 * Obtain the readahead parameters for the file
747 * specified by (dev, ino).
750 static inline struct raparms *
751 nfsd_get_raparms(dev_t dev, ino_t ino)
753 struct raparms *ra, **rap, **frap = NULL;
756 struct raparm_hbucket *rab;
758 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
759 rab = &raparm_hash[hash];
761 spin_lock(&rab->pb_lock);
762 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
763 if (ra->p_ino == ino && ra->p_dev == dev)
766 if (ra->p_count == 0)
769 depth = nfsdstats.ra_size*11/10;
771 spin_unlock(&rab->pb_lock);
781 if (rap != &rab->pb_head) {
783 ra->p_next = rab->pb_head;
787 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
788 spin_unlock(&rab->pb_lock);
793 * Grab and keep cached pages associated with a file in the svc_rqst
794 * so that they can be passed to the network sendmsg/sendpage routines
795 * directly. They will be released after the sending has completed.
798 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
799 struct splice_desc *sd)
801 struct svc_rqst *rqstp = sd->u.data;
802 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
803 struct page *page = buf->page;
807 ret = buf->ops->confirm(pipe, buf);
813 if (rqstp->rq_res.page_len == 0) {
818 rqstp->rq_res.page_base = buf->offset;
819 rqstp->rq_res.page_len = size;
820 } else if (page != pp[-1]) {
826 rqstp->rq_res.page_len += size;
828 rqstp->rq_res.page_len += size;
833 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
834 struct splice_desc *sd)
836 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
840 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
841 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
850 inode = file->f_path.dentry->d_inode;
852 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
853 (!lock_may_read(inode, offset, *count)))
857 /* Get readahead parameters */
858 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
861 file->f_ra = ra->p_ra;
863 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
864 struct splice_desc sd = {
871 rqstp->rq_resused = 1;
872 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
876 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
880 /* Write back readahead params */
882 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
883 spin_lock(&rab->pb_lock);
884 ra->p_ra = file->f_ra;
887 spin_unlock(&rab->pb_lock);
891 nfsdstats.io_read += host_err;
894 fsnotify_access(file->f_path.dentry);
896 err = nfserrno(host_err);
901 static void kill_suid(struct dentry *dentry)
904 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
906 mutex_lock(&dentry->d_inode->i_mutex);
907 notify_change(dentry, &ia);
908 mutex_unlock(&dentry->d_inode->i_mutex);
912 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
913 loff_t offset, struct kvec *vec, int vlen,
914 unsigned long cnt, int *stablep)
916 struct svc_export *exp;
917 struct dentry *dentry;
922 int stable = *stablep;
927 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
928 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
932 dentry = file->f_path.dentry;
933 inode = dentry->d_inode;
934 exp = fhp->fh_export;
937 * Request sync writes if
938 * - the sync export option has been set, or
939 * - the client requested O_SYNC behavior (NFSv3 feature).
940 * - The file system doesn't support fsync().
941 * When gathered writes have been configured for this volume,
942 * flushing the data to disk is handled separately below.
945 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
947 *stablep = 2; /* FILE_SYNC */
952 if (stable && !EX_WGATHER(exp))
953 file->f_flags |= O_SYNC;
955 /* Write the data. */
956 oldfs = get_fs(); set_fs(KERNEL_DS);
957 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
960 nfsdstats.io_write += cnt;
961 fsnotify_modify(file->f_path.dentry);
964 /* clear setuid/setgid flag after write */
965 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
968 if (host_err >= 0 && stable) {
969 static ino_t last_ino;
970 static dev_t last_dev;
973 * Gathered writes: If another process is currently
974 * writing to the file, there's a high chance
975 * this is another nfsd (triggered by a bulk write
976 * from a client's biod). Rather than syncing the
977 * file with each write request, we sleep for 10 msec.
979 * I don't know if this roughly approximates
980 * C. Juszak's idea of gathered writes, but it's a
981 * nice and simple solution (IMHO), and it seems to
984 if (EX_WGATHER(exp)) {
985 if (atomic_read(&inode->i_writecount) > 1
986 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
987 dprintk("nfsd: write defer %d\n", current->pid);
989 dprintk("nfsd: write resume %d\n", current->pid);
992 if (inode->i_state & I_DIRTY) {
993 dprintk("nfsd: write sync %d\n", current->pid);
994 host_err=nfsd_sync(file);
997 wake_up(&inode->i_wait);
1000 last_ino = inode->i_ino;
1001 last_dev = inode->i_sb->s_dev;
1004 dprintk("nfsd: write complete host_err=%d\n", host_err);
1008 err = nfserrno(host_err);
1014 * Read data from a file. count must contain the requested read count
1015 * on entry. On return, *count contains the number of bytes actually read.
1016 * N.B. After this call fhp needs an fh_put
1019 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1020 loff_t offset, struct kvec *vec, int vlen,
1021 unsigned long *count)
1026 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1027 MAY_READ|MAY_OWNER_OVERRIDE);
1030 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1032 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1035 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1043 * Write data to a file.
1044 * The stable flag requests synchronous writes.
1045 * N.B. After this call fhp needs an fh_put
1048 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1049 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1055 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1056 MAY_WRITE|MAY_OWNER_OVERRIDE);
1059 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1062 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1067 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1075 #ifdef CONFIG_NFSD_V3
1077 * Commit all pending writes to stable storage.
1078 * Strictly speaking, we could sync just the indicated file region here,
1079 * but there's currently no way we can ask the VFS to do so.
1081 * Unfortunately we cannot lock the file to make sure we return full WCC
1082 * data to the client, as locking happens lower down in the filesystem.
1085 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1086 loff_t offset, unsigned long count)
1091 if ((u64)count > ~(u64)offset)
1092 return nfserr_inval;
1094 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1096 if (EX_ISSYNC(fhp->fh_export)) {
1097 if (file->f_op && file->f_op->fsync) {
1098 err = nfserrno(nfsd_sync(file));
1100 err = nfserr_notsupp;
1107 #endif /* CONFIG_NFSD_V3 */
1110 * Create a file (regular, directory, device, fifo); UNIX sockets
1111 * not yet implemented.
1112 * If the response fh has been verified, the parent directory should
1113 * already be locked. Note that the parent directory is left locked.
1115 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1118 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1119 char *fname, int flen, struct iattr *iap,
1120 int type, dev_t rdev, struct svc_fh *resfhp)
1122 struct dentry *dentry, *dchild = NULL;
1131 if (isdotent(fname, flen))
1134 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1138 dentry = fhp->fh_dentry;
1139 dirp = dentry->d_inode;
1141 err = nfserr_notdir;
1142 if(!dirp->i_op || !dirp->i_op->lookup)
1145 * Check whether the response file handle has been verified yet.
1146 * If it has, the parent directory should already be locked.
1148 if (!resfhp->fh_dentry) {
1149 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1150 fh_lock_nested(fhp, I_MUTEX_PARENT);
1151 dchild = lookup_one_len(fname, dentry, flen);
1152 host_err = PTR_ERR(dchild);
1155 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1159 /* called from nfsd_proc_create */
1160 dchild = dget(resfhp->fh_dentry);
1161 if (!fhp->fh_locked) {
1162 /* not actually possible */
1164 "nfsd_create: parent %s/%s not locked!\n",
1165 dentry->d_parent->d_name.name,
1166 dentry->d_name.name);
1172 * Make sure the child dentry is still negative ...
1175 if (dchild->d_inode) {
1176 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1177 dentry->d_name.name, dchild->d_name.name);
1181 if (!(iap->ia_valid & ATTR_MODE))
1183 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1186 * Get the dir op function pointer.
1191 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1194 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1200 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1203 printk("nfsd: bad file type %o in nfsd_create\n", type);
1209 if (EX_ISSYNC(fhp->fh_export)) {
1210 err = nfserrno(nfsd_sync_dir(dentry));
1211 write_inode_now(dchild->d_inode, 1);
1215 /* Set file attributes. Mode has already been set and
1216 * setting uid/gid works only for root. Irix appears to
1217 * send along the gid when it tries to implement setgid
1218 * directories via NFS.
1220 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1221 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1226 * Update the file handle to get the new inode info.
1229 err = fh_update(resfhp);
1231 if (dchild && !IS_ERR(dchild))
1236 err = nfserrno(host_err);
1240 #ifdef CONFIG_NFSD_V3
1242 * NFSv3 version of nfsd_create
1245 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1246 char *fname, int flen, struct iattr *iap,
1247 struct svc_fh *resfhp, int createmode, u32 *verifier,
1248 int *truncp, int *created)
1250 struct dentry *dentry, *dchild = NULL;
1254 __u32 v_mtime=0, v_atime=0;
1260 if (isdotent(fname, flen))
1262 if (!(iap->ia_valid & ATTR_MODE))
1264 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1268 dentry = fhp->fh_dentry;
1269 dirp = dentry->d_inode;
1271 /* Get all the sanity checks out of the way before
1272 * we lock the parent. */
1273 err = nfserr_notdir;
1274 if(!dirp->i_op || !dirp->i_op->lookup)
1276 fh_lock_nested(fhp, I_MUTEX_PARENT);
1279 * Compose the response file handle.
1281 dchild = lookup_one_len(fname, dentry, flen);
1282 host_err = PTR_ERR(dchild);
1286 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1290 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1291 /* solaris7 gets confused (bugid 4218508) if these have
1292 * the high bit set, so just clear the high bits.
1294 v_mtime = verifier[0]&0x7fffffff;
1295 v_atime = verifier[1]&0x7fffffff;
1298 if (dchild->d_inode) {
1301 switch (createmode) {
1302 case NFS3_CREATE_UNCHECKED:
1303 if (! S_ISREG(dchild->d_inode->i_mode))
1306 /* in nfsv4, we need to treat this case a little
1307 * differently. we don't want to truncate the
1308 * file now; this would be wrong if the OPEN
1309 * fails for some other reason. furthermore,
1310 * if the size is nonzero, we should ignore it
1311 * according to spec!
1313 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1316 iap->ia_valid &= ATTR_SIZE;
1320 case NFS3_CREATE_EXCLUSIVE:
1321 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1322 && dchild->d_inode->i_atime.tv_sec == v_atime
1323 && dchild->d_inode->i_size == 0 )
1326 case NFS3_CREATE_GUARDED:
1332 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1338 if (EX_ISSYNC(fhp->fh_export)) {
1339 err = nfserrno(nfsd_sync_dir(dentry));
1340 /* setattr will sync the child (or not) */
1343 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1344 /* Cram the verifier into atime/mtime */
1345 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1346 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1347 /* XXX someone who knows this better please fix it for nsec */
1348 iap->ia_mtime.tv_sec = v_mtime;
1349 iap->ia_atime.tv_sec = v_atime;
1350 iap->ia_mtime.tv_nsec = 0;
1351 iap->ia_atime.tv_nsec = 0;
1354 /* Set file attributes.
1355 * Irix appears to send along the gid when it tries to
1356 * implement setgid directories via NFS. Clear out all that cruft.
1359 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1360 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1366 * Update the filehandle to get the new inode info.
1369 err = fh_update(resfhp);
1373 if (dchild && !IS_ERR(dchild))
1378 err = nfserrno(host_err);
1381 #endif /* CONFIG_NFSD_V3 */
1384 * Read a symlink. On entry, *lenp must contain the maximum path length that
1385 * fits into the buffer. On return, it contains the true length.
1386 * N.B. After this call fhp needs an fh_put
1389 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1391 struct dentry *dentry;
1392 struct inode *inode;
1397 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1401 dentry = fhp->fh_dentry;
1402 inode = dentry->d_inode;
1405 if (!inode->i_op || !inode->i_op->readlink)
1408 touch_atime(fhp->fh_export->ex_mnt, dentry);
1409 /* N.B. Why does this call need a get_fs()??
1410 * Remove the set_fs and watch the fireworks:-) --okir
1413 oldfs = get_fs(); set_fs(KERNEL_DS);
1414 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1425 err = nfserrno(host_err);
1430 * Create a symlink and look up its inode
1431 * N.B. After this call _both_ fhp and resfhp need an fh_put
1434 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1435 char *fname, int flen,
1436 char *path, int plen,
1437 struct svc_fh *resfhp,
1440 struct dentry *dentry, *dnew;
1449 if (isdotent(fname, flen))
1452 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1456 dentry = fhp->fh_dentry;
1457 dnew = lookup_one_len(fname, dentry, flen);
1458 host_err = PTR_ERR(dnew);
1463 /* Only the MODE ATTRibute is even vaguely meaningful */
1464 if (iap && (iap->ia_valid & ATTR_MODE))
1465 mode = iap->ia_mode & S_IALLUGO;
1467 if (unlikely(path[plen] != 0)) {
1468 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1469 if (path_alloced == NULL)
1472 strncpy(path_alloced, path, plen);
1473 path_alloced[plen] = 0;
1474 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1475 kfree(path_alloced);
1478 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1481 if (EX_ISSYNC(fhp->fh_export))
1482 host_err = nfsd_sync_dir(dentry);
1484 err = nfserrno(host_err);
1487 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1489 if (err==0) err = cerr;
1494 err = nfserrno(host_err);
1500 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1503 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1504 char *name, int len, struct svc_fh *tfhp)
1506 struct dentry *ddir, *dnew, *dold;
1507 struct inode *dirp, *dest;
1511 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1514 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1522 if (isdotent(name, len))
1525 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1526 ddir = ffhp->fh_dentry;
1527 dirp = ddir->d_inode;
1529 dnew = lookup_one_len(name, ddir, len);
1530 host_err = PTR_ERR(dnew);
1534 dold = tfhp->fh_dentry;
1535 dest = dold->d_inode;
1537 host_err = vfs_link(dold, dirp, dnew);
1539 if (EX_ISSYNC(ffhp->fh_export)) {
1540 err = nfserrno(nfsd_sync_dir(ddir));
1541 write_inode_now(dest, 1);
1545 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1548 err = nfserrno(host_err);
1558 err = nfserrno(host_err);
1564 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1567 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1568 struct svc_fh *tfhp, char *tname, int tlen)
1570 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1571 struct inode *fdir, *tdir;
1575 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1578 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1582 fdentry = ffhp->fh_dentry;
1583 fdir = fdentry->d_inode;
1585 tdentry = tfhp->fh_dentry;
1586 tdir = tdentry->d_inode;
1588 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1589 if (ffhp->fh_export != tfhp->fh_export)
1593 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1596 /* cannot use fh_lock as we need deadlock protective ordering
1597 * so do it by hand */
1598 trap = lock_rename(tdentry, fdentry);
1599 ffhp->fh_locked = tfhp->fh_locked = 1;
1603 odentry = lookup_one_len(fname, fdentry, flen);
1604 host_err = PTR_ERR(odentry);
1605 if (IS_ERR(odentry))
1609 if (!odentry->d_inode)
1612 if (odentry == trap)
1615 ndentry = lookup_one_len(tname, tdentry, tlen);
1616 host_err = PTR_ERR(ndentry);
1617 if (IS_ERR(ndentry))
1619 host_err = -ENOTEMPTY;
1620 if (ndentry == trap)
1624 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1625 ((atomic_read(&odentry->d_count) > 1)
1626 || (atomic_read(&ndentry->d_count) > 1))) {
1630 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1631 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1632 host_err = nfsd_sync_dir(tdentry);
1634 host_err = nfsd_sync_dir(fdentry);
1642 err = nfserrno(host_err);
1644 /* we cannot reply on fh_unlock on the two filehandles,
1645 * as that would do the wrong thing if the two directories
1646 * were the same, so again we do it by hand
1648 fill_post_wcc(ffhp);
1649 fill_post_wcc(tfhp);
1650 unlock_rename(tdentry, fdentry);
1651 ffhp->fh_locked = tfhp->fh_locked = 0;
1658 * Unlink a file or directory
1659 * N.B. After this call fhp needs an fh_put
1662 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1663 char *fname, int flen)
1665 struct dentry *dentry, *rdentry;
1671 if (!flen || isdotent(fname, flen))
1673 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1677 fh_lock_nested(fhp, I_MUTEX_PARENT);
1678 dentry = fhp->fh_dentry;
1679 dirp = dentry->d_inode;
1681 rdentry = lookup_one_len(fname, dentry, flen);
1682 host_err = PTR_ERR(rdentry);
1683 if (IS_ERR(rdentry))
1686 if (!rdentry->d_inode) {
1693 type = rdentry->d_inode->i_mode & S_IFMT;
1695 if (type != S_IFDIR) { /* It's UNLINK */
1697 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1698 (atomic_read(&rdentry->d_count) > 1)) {
1702 host_err = vfs_unlink(dirp, rdentry);
1703 } else { /* It's RMDIR */
1704 host_err = vfs_rmdir(dirp, rdentry);
1711 if (EX_ISSYNC(fhp->fh_export))
1712 host_err = nfsd_sync_dir(dentry);
1715 err = nfserrno(host_err);
1721 * Read entries from a directory.
1722 * The NFSv3/4 verifier we ignore for now.
1725 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1726 struct readdir_cd *cdp, filldir_t func)
1731 loff_t offset = *offsetp;
1733 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1737 offset = vfs_llseek(file, offset, 0);
1739 err = nfserrno((int)offset);
1744 * Read the directory entries. This silly loop is necessary because
1745 * readdir() is not guaranteed to fill up the entire buffer, but
1746 * may choose to do less.
1750 cdp->err = nfserr_eof; /* will be cleared on successful read */
1751 host_err = vfs_readdir(file, func, cdp);
1752 } while (host_err >=0 && cdp->err == nfs_ok);
1754 err = nfserrno(host_err);
1757 *offsetp = vfs_llseek(file, 0, 1);
1759 if (err == nfserr_eof || err == nfserr_toosmall)
1760 err = nfs_ok; /* can still be found in ->err */
1768 * Get file system stats
1769 * N.B. After this call fhp needs an fh_put
1772 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1774 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1775 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1781 * Check for a user's access permissions to this inode.
1784 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1786 struct inode *inode = dentry->d_inode;
1792 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1794 (acc & MAY_READ)? " read" : "",
1795 (acc & MAY_WRITE)? " write" : "",
1796 (acc & MAY_EXEC)? " exec" : "",
1797 (acc & MAY_SATTR)? " sattr" : "",
1798 (acc & MAY_TRUNC)? " trunc" : "",
1799 (acc & MAY_LOCK)? " lock" : "",
1800 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1802 IS_IMMUTABLE(inode)? " immut" : "",
1803 IS_APPEND(inode)? " append" : "",
1804 IS_RDONLY(inode)? " ro" : "");
1805 dprintk(" owner %d/%d user %d/%d\n",
1806 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1809 /* Normally we reject any write/sattr etc access on a read-only file
1810 * system. But if it is IRIX doing check on write-access for a
1811 * device special file, we ignore rofs.
1813 if (!(acc & MAY_LOCAL_ACCESS))
1814 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1815 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1817 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1820 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1823 if (acc & MAY_LOCK) {
1824 /* If we cannot rely on authentication in NLM requests,
1825 * just allow locks, otherwise require read permission, or
1828 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1831 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1834 * The file owner always gets access permission for accesses that
1835 * would normally be checked at open time. This is to make
1836 * file access work even when the client has done a fchmod(fd, 0).
1838 * However, `cp foo bar' should fail nevertheless when bar is
1839 * readonly. A sensible way to do this might be to reject all
1840 * attempts to truncate a read-only file, because a creat() call
1841 * always implies file truncation.
1842 * ... but this isn't really fair. A process may reasonably call
1843 * ftruncate on an open file descriptor on a file with perm 000.
1844 * We must trust the client to do permission checking - using "ACCESS"
1847 if ((acc & MAY_OWNER_OVERRIDE) &&
1848 inode->i_uid == current->fsuid)
1851 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1853 /* Allow read access to binaries even when mode 111 */
1854 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1855 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1856 err = permission(inode, MAY_EXEC, NULL);
1858 return err? nfserrno(err) : 0;
1862 nfsd_racache_shutdown(void)
1866 dprintk("nfsd: freeing readahead buffers.\n");
1871 * Initialize readahead param cache
1874 nfsd_racache_init(int cache_size)
1883 if (cache_size < 2*RAPARM_HASH_SIZE)
1884 cache_size = 2*RAPARM_HASH_SIZE;
1885 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1889 "nfsd: Could not allocate memory read-ahead cache.\n");
1893 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1894 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1895 raparm_hash[i].pb_head = NULL;
1896 spin_lock_init(&raparm_hash[i].pb_lock);
1898 nperbucket = cache_size >> RAPARM_HASH_BITS;
1899 for (i = 0; i < cache_size - 1; i++) {
1900 if (i % nperbucket == 0)
1901 raparm_hash[j++].pb_head = raparml + i;
1902 if (i % nperbucket < nperbucket-1)
1903 raparml[i].p_next = raparml + i + 1;
1906 nfsdstats.ra_size = cache_size;
1910 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1912 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1914 struct inode *inode = fhp->fh_dentry->d_inode;
1918 struct posix_acl *acl;
1920 if (!IS_POSIXACL(inode))
1921 return ERR_PTR(-EOPNOTSUPP);
1924 case ACL_TYPE_ACCESS:
1925 name = POSIX_ACL_XATTR_ACCESS;
1927 case ACL_TYPE_DEFAULT:
1928 name = POSIX_ACL_XATTR_DEFAULT;
1931 return ERR_PTR(-EOPNOTSUPP);
1934 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1936 return ERR_PTR(size);
1938 acl = posix_acl_from_xattr(value, size);
1944 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1946 struct inode *inode = fhp->fh_dentry->d_inode;
1952 if (!IS_POSIXACL(inode) || !inode->i_op ||
1953 !inode->i_op->setxattr || !inode->i_op->removexattr)
1956 case ACL_TYPE_ACCESS:
1957 name = POSIX_ACL_XATTR_ACCESS;
1959 case ACL_TYPE_DEFAULT:
1960 name = POSIX_ACL_XATTR_DEFAULT;
1966 if (acl && acl->a_count) {
1967 size = posix_acl_xattr_size(acl->a_count);
1968 value = kmalloc(size, GFP_KERNEL);
1971 error = posix_acl_to_xattr(acl, value, size);
1979 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1981 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1984 error = vfs_removexattr(fhp->fh_dentry, name);
1985 if (error == -ENODATA)
1994 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */