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 = rqst_exp_get_by_name(rqstp, mnt, mounts);
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 = rqst_exp_parent(rqstp, mnt, dentry);
192 if (PTR_ERR(exp2) == -ENOENT) {
194 dentry = dget(dparent);
195 } else if (IS_ERR(exp2)) {
196 host_err = PTR_ERR(exp2);
208 dentry = lookup_one_len(name, dparent, len);
209 host_err = PTR_ERR(dentry);
213 * check if we have crossed a mount point ...
215 if (d_mountpoint(dentry)) {
216 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
223 * Note: we compose the file handle now, but as the
224 * dentry may be negative, it may need to be updated.
226 err = fh_compose(resfh, exp, dentry, fhp);
227 if (!err && !dentry->d_inode)
235 err = nfserrno(host_err);
240 * Set various file attributes.
241 * N.B. After this call fhp needs an fh_put
244 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
245 int check_guard, time_t guardtime)
247 struct dentry *dentry;
249 int accmode = MAY_SATTR;
256 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
257 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
258 if (iap->ia_valid & ATTR_SIZE)
262 err = fh_verify(rqstp, fhp, ftype, accmode);
266 dentry = fhp->fh_dentry;
267 inode = dentry->d_inode;
269 /* Ignore any mode updates on symlinks */
270 if (S_ISLNK(inode->i_mode))
271 iap->ia_valid &= ~ATTR_MODE;
276 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
277 * which only requires access, and "set-[ac]time-to-X" which
278 * requires ownership.
279 * So if it looks like it might be "set both to the same time which
280 * is close to now", and if inode_change_ok fails, then we
281 * convert to "set to now" instead of "set to explicit time"
283 * We only call inode_change_ok as the last test as technically
284 * it is not an interface that we should be using. It is only
285 * valid if the filesystem does not define it's own i_op->setattr.
287 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
288 #define MAX_TOUCH_TIME_ERROR (30*60)
289 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
290 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
292 /* Looks probable. Now just make sure time is in the right ballpark.
293 * Solaris, at least, doesn't seem to care what the time request is.
294 * We require it be within 30 minutes of now.
296 time_t delta = iap->ia_atime.tv_sec - get_seconds();
297 if (delta<0) delta = -delta;
298 if (delta < MAX_TOUCH_TIME_ERROR &&
299 inode_change_ok(inode, iap) != 0) {
300 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
301 * this will cause notify_change to set these times to "now"
303 iap->ia_valid &= ~BOTH_TIME_SET;
307 /* The size case is special. It changes the file as well as the attributes. */
308 if (iap->ia_valid & ATTR_SIZE) {
309 if (iap->ia_size < inode->i_size) {
310 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
316 * If we are changing the size of the file, then
317 * we need to break all leases.
319 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
320 if (host_err == -EWOULDBLOCK)
321 host_err = -ETIMEDOUT;
322 if (host_err) /* ENOMEM or EWOULDBLOCK */
325 host_err = get_write_access(inode);
330 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
332 put_write_access(inode);
338 imode = inode->i_mode;
339 if (iap->ia_valid & ATTR_MODE) {
340 iap->ia_mode &= S_IALLUGO;
341 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
344 /* Revoke setuid/setgid bit on chown/chgrp */
345 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
346 iap->ia_valid |= ATTR_KILL_SUID;
347 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
348 iap->ia_valid |= ATTR_KILL_SGID;
350 /* Change the attributes. */
352 iap->ia_valid |= ATTR_CTIME;
354 err = nfserr_notsync;
355 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
357 host_err = notify_change(dentry, iap);
358 err = nfserrno(host_err);
362 put_write_access(inode);
364 if (EX_ISSYNC(fhp->fh_export))
365 write_inode_now(inode, 1);
370 err = nfserrno(host_err);
374 #if defined(CONFIG_NFSD_V2_ACL) || \
375 defined(CONFIG_NFSD_V3_ACL) || \
376 defined(CONFIG_NFSD_V4)
377 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
381 buflen = vfs_getxattr(dentry, key, NULL, 0);
385 *buf = kmalloc(buflen, GFP_KERNEL);
389 return vfs_getxattr(dentry, key, *buf, buflen);
393 #if defined(CONFIG_NFSD_V4)
395 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
402 buflen = posix_acl_xattr_size(pacl->a_count);
403 buf = kmalloc(buflen, GFP_KERNEL);
408 len = posix_acl_to_xattr(pacl, buf, buflen);
414 error = vfs_setxattr(dentry, key, buf, len, 0);
421 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
422 struct nfs4_acl *acl)
426 struct dentry *dentry;
428 struct posix_acl *pacl = NULL, *dpacl = NULL;
429 unsigned int flags = 0;
432 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
436 dentry = fhp->fh_dentry;
437 inode = dentry->d_inode;
438 if (S_ISDIR(inode->i_mode))
439 flags = NFS4_ACL_DIR;
441 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
442 if (host_error == -EINVAL) {
443 return nfserr_attrnotsupp;
444 } else if (host_error < 0)
447 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
451 if (S_ISDIR(inode->i_mode))
452 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
455 posix_acl_release(pacl);
456 posix_acl_release(dpacl);
458 if (host_error == -EOPNOTSUPP)
459 return nfserr_attrnotsupp;
461 return nfserrno(host_error);
464 static struct posix_acl *
465 _get_posix_acl(struct dentry *dentry, char *key)
468 struct posix_acl *pacl = NULL;
471 buflen = nfsd_getxattr(dentry, key, &buf);
475 return ERR_PTR(buflen);
477 pacl = posix_acl_from_xattr(buf, buflen);
483 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
485 struct inode *inode = dentry->d_inode;
487 struct posix_acl *pacl = NULL, *dpacl = NULL;
488 unsigned int flags = 0;
490 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
491 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
492 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
494 error = PTR_ERR(pacl);
499 if (S_ISDIR(inode->i_mode)) {
500 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
501 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
503 else if (IS_ERR(dpacl)) {
504 error = PTR_ERR(dpacl);
508 flags = NFS4_ACL_DIR;
511 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
513 error = PTR_ERR(*acl);
517 posix_acl_release(pacl);
518 posix_acl_release(dpacl);
522 #endif /* defined(CONFIG_NFS_V4) */
524 #ifdef CONFIG_NFSD_V3
526 * Check server access rights to a file system object
532 static struct accessmap nfs3_regaccess[] = {
533 { NFS3_ACCESS_READ, MAY_READ },
534 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
535 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
536 { NFS3_ACCESS_EXTEND, MAY_WRITE },
541 static struct accessmap nfs3_diraccess[] = {
542 { NFS3_ACCESS_READ, MAY_READ },
543 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
544 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
545 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
546 { NFS3_ACCESS_DELETE, MAY_REMOVE },
551 static struct accessmap nfs3_anyaccess[] = {
552 /* Some clients - Solaris 2.6 at least, make an access call
553 * to the server to check for access for things like /dev/null
554 * (which really, the server doesn't care about). So
555 * We provide simple access checking for them, looking
556 * mainly at mode bits, and we make sure to ignore read-only
559 { NFS3_ACCESS_READ, MAY_READ },
560 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
561 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
562 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
568 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
570 struct accessmap *map;
571 struct svc_export *export;
572 struct dentry *dentry;
573 u32 query, result = 0, sresult = 0;
576 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
580 export = fhp->fh_export;
581 dentry = fhp->fh_dentry;
583 if (S_ISREG(dentry->d_inode->i_mode))
584 map = nfs3_regaccess;
585 else if (S_ISDIR(dentry->d_inode->i_mode))
586 map = nfs3_diraccess;
588 map = nfs3_anyaccess;
592 for (; map->access; map++) {
593 if (map->access & query) {
596 sresult |= map->access;
598 err2 = nfsd_permission(export, dentry, map->how);
601 result |= map->access;
604 /* the following error codes just mean the access was not allowed,
605 * rather than an error occurred */
609 /* simply don't "or" in the access bit. */
619 *supported = sresult;
624 #endif /* CONFIG_NFSD_V3 */
629 * Open an existing file or directory.
630 * The access argument indicates the type of open (read/write/lock)
631 * N.B. After this call fhp needs an fh_put
634 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
635 int access, struct file **filp)
637 struct dentry *dentry;
639 int flags = O_RDONLY|O_LARGEFILE;
644 * If we get here, then the client has already done an "open",
645 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
646 * in case a chmod has now revoked permission.
648 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
652 dentry = fhp->fh_dentry;
653 inode = dentry->d_inode;
655 /* Disallow write access to files with the append-only bit set
656 * or any access when mandatory locking enabled
659 if (IS_APPEND(inode) && (access & MAY_WRITE))
661 if (IS_ISMNDLK(inode))
668 * Check to see if there are any leases on this file.
669 * This may block while leases are broken.
671 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
672 if (host_err == -EWOULDBLOCK)
673 host_err = -ETIMEDOUT;
674 if (host_err) /* NOMEM or WOULDBLOCK */
677 if (access & MAY_WRITE) {
678 if (access & MAY_READ)
679 flags = O_RDWR|O_LARGEFILE;
681 flags = O_WRONLY|O_LARGEFILE;
685 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
687 host_err = PTR_ERR(*filp);
689 err = nfserrno(host_err);
698 nfsd_close(struct file *filp)
705 * As this calls fsync (not fdatasync) there is no need for a write_inode
708 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
709 const struct file_operations *fop)
711 struct inode *inode = dp->d_inode;
712 int (*fsync) (struct file *, struct dentry *, int);
715 err = filemap_fdatawrite(inode->i_mapping);
716 if (err == 0 && fop && (fsync = fop->fsync))
717 err = fsync(filp, dp, 0);
719 err = filemap_fdatawait(inode->i_mapping);
726 nfsd_sync(struct file *filp)
729 struct inode *inode = filp->f_path.dentry->d_inode;
730 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
731 mutex_lock(&inode->i_mutex);
732 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
733 mutex_unlock(&inode->i_mutex);
739 nfsd_sync_dir(struct dentry *dp)
741 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
745 * Obtain the readahead parameters for the file
746 * specified by (dev, ino).
749 static inline struct raparms *
750 nfsd_get_raparms(dev_t dev, ino_t ino)
752 struct raparms *ra, **rap, **frap = NULL;
755 struct raparm_hbucket *rab;
757 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
758 rab = &raparm_hash[hash];
760 spin_lock(&rab->pb_lock);
761 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
762 if (ra->p_ino == ino && ra->p_dev == dev)
765 if (ra->p_count == 0)
768 depth = nfsdstats.ra_size*11/10;
770 spin_unlock(&rab->pb_lock);
780 if (rap != &rab->pb_head) {
782 ra->p_next = rab->pb_head;
786 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
787 spin_unlock(&rab->pb_lock);
792 * Grab and keep cached pages associated with a file in the svc_rqst
793 * so that they can be passed to the network sendmsg/sendpage routines
794 * directly. They will be released after the sending has completed.
797 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
798 struct splice_desc *sd)
800 struct svc_rqst *rqstp = sd->u.data;
801 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
802 struct page *page = buf->page;
806 ret = buf->ops->confirm(pipe, buf);
812 if (rqstp->rq_res.page_len == 0) {
817 rqstp->rq_res.page_base = buf->offset;
818 rqstp->rq_res.page_len = size;
819 } else if (page != pp[-1]) {
825 rqstp->rq_res.page_len += size;
827 rqstp->rq_res.page_len += size;
832 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
833 struct splice_desc *sd)
835 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
839 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
840 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
849 inode = file->f_path.dentry->d_inode;
851 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
852 (!lock_may_read(inode, offset, *count)))
856 /* Get readahead parameters */
857 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
860 file->f_ra = ra->p_ra;
862 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
863 struct splice_desc sd = {
870 rqstp->rq_resused = 1;
871 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
875 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
879 /* Write back readahead params */
881 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
882 spin_lock(&rab->pb_lock);
883 ra->p_ra = file->f_ra;
886 spin_unlock(&rab->pb_lock);
890 nfsdstats.io_read += host_err;
893 fsnotify_access(file->f_path.dentry);
895 err = nfserrno(host_err);
900 static void kill_suid(struct dentry *dentry)
903 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
905 mutex_lock(&dentry->d_inode->i_mutex);
906 notify_change(dentry, &ia);
907 mutex_unlock(&dentry->d_inode->i_mutex);
911 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
912 loff_t offset, struct kvec *vec, int vlen,
913 unsigned long cnt, int *stablep)
915 struct svc_export *exp;
916 struct dentry *dentry;
921 int stable = *stablep;
926 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
927 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
931 dentry = file->f_path.dentry;
932 inode = dentry->d_inode;
933 exp = fhp->fh_export;
936 * Request sync writes if
937 * - the sync export option has been set, or
938 * - the client requested O_SYNC behavior (NFSv3 feature).
939 * - The file system doesn't support fsync().
940 * When gathered writes have been configured for this volume,
941 * flushing the data to disk is handled separately below.
944 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
946 *stablep = 2; /* FILE_SYNC */
951 if (stable && !EX_WGATHER(exp))
952 file->f_flags |= O_SYNC;
954 /* Write the data. */
955 oldfs = get_fs(); set_fs(KERNEL_DS);
956 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
959 nfsdstats.io_write += cnt;
960 fsnotify_modify(file->f_path.dentry);
963 /* clear setuid/setgid flag after write */
964 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
967 if (host_err >= 0 && stable) {
968 static ino_t last_ino;
969 static dev_t last_dev;
972 * Gathered writes: If another process is currently
973 * writing to the file, there's a high chance
974 * this is another nfsd (triggered by a bulk write
975 * from a client's biod). Rather than syncing the
976 * file with each write request, we sleep for 10 msec.
978 * I don't know if this roughly approximates
979 * C. Juszak's idea of gathered writes, but it's a
980 * nice and simple solution (IMHO), and it seems to
983 if (EX_WGATHER(exp)) {
984 if (atomic_read(&inode->i_writecount) > 1
985 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
986 dprintk("nfsd: write defer %d\n", current->pid);
988 dprintk("nfsd: write resume %d\n", current->pid);
991 if (inode->i_state & I_DIRTY) {
992 dprintk("nfsd: write sync %d\n", current->pid);
993 host_err=nfsd_sync(file);
996 wake_up(&inode->i_wait);
999 last_ino = inode->i_ino;
1000 last_dev = inode->i_sb->s_dev;
1003 dprintk("nfsd: write complete host_err=%d\n", host_err);
1007 err = nfserrno(host_err);
1013 * Read data from a file. count must contain the requested read count
1014 * on entry. On return, *count contains the number of bytes actually read.
1015 * N.B. After this call fhp needs an fh_put
1018 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1019 loff_t offset, struct kvec *vec, int vlen,
1020 unsigned long *count)
1025 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1026 MAY_READ|MAY_OWNER_OVERRIDE);
1029 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1031 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1034 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1042 * Write data to a file.
1043 * The stable flag requests synchronous writes.
1044 * N.B. After this call fhp needs an fh_put
1047 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1048 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1054 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1055 MAY_WRITE|MAY_OWNER_OVERRIDE);
1058 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1061 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1066 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1074 #ifdef CONFIG_NFSD_V3
1076 * Commit all pending writes to stable storage.
1077 * Strictly speaking, we could sync just the indicated file region here,
1078 * but there's currently no way we can ask the VFS to do so.
1080 * Unfortunately we cannot lock the file to make sure we return full WCC
1081 * data to the client, as locking happens lower down in the filesystem.
1084 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1085 loff_t offset, unsigned long count)
1090 if ((u64)count > ~(u64)offset)
1091 return nfserr_inval;
1093 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1095 if (EX_ISSYNC(fhp->fh_export)) {
1096 if (file->f_op && file->f_op->fsync) {
1097 err = nfserrno(nfsd_sync(file));
1099 err = nfserr_notsupp;
1106 #endif /* CONFIG_NFSD_V3 */
1109 * Create a file (regular, directory, device, fifo); UNIX sockets
1110 * not yet implemented.
1111 * If the response fh has been verified, the parent directory should
1112 * already be locked. Note that the parent directory is left locked.
1114 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1117 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1118 char *fname, int flen, struct iattr *iap,
1119 int type, dev_t rdev, struct svc_fh *resfhp)
1121 struct dentry *dentry, *dchild = NULL;
1130 if (isdotent(fname, flen))
1133 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1137 dentry = fhp->fh_dentry;
1138 dirp = dentry->d_inode;
1140 err = nfserr_notdir;
1141 if(!dirp->i_op || !dirp->i_op->lookup)
1144 * Check whether the response file handle has been verified yet.
1145 * If it has, the parent directory should already be locked.
1147 if (!resfhp->fh_dentry) {
1148 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1149 fh_lock_nested(fhp, I_MUTEX_PARENT);
1150 dchild = lookup_one_len(fname, dentry, flen);
1151 host_err = PTR_ERR(dchild);
1154 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1158 /* called from nfsd_proc_create */
1159 dchild = dget(resfhp->fh_dentry);
1160 if (!fhp->fh_locked) {
1161 /* not actually possible */
1163 "nfsd_create: parent %s/%s not locked!\n",
1164 dentry->d_parent->d_name.name,
1165 dentry->d_name.name);
1171 * Make sure the child dentry is still negative ...
1174 if (dchild->d_inode) {
1175 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1176 dentry->d_name.name, dchild->d_name.name);
1180 if (!(iap->ia_valid & ATTR_MODE))
1182 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1185 * Get the dir op function pointer.
1190 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1193 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1199 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1202 printk("nfsd: bad file type %o in nfsd_create\n", type);
1208 if (EX_ISSYNC(fhp->fh_export)) {
1209 err = nfserrno(nfsd_sync_dir(dentry));
1210 write_inode_now(dchild->d_inode, 1);
1214 /* Set file attributes. Mode has already been set and
1215 * setting uid/gid works only for root. Irix appears to
1216 * send along the gid when it tries to implement setgid
1217 * directories via NFS.
1219 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1220 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1225 * Update the file handle to get the new inode info.
1228 err = fh_update(resfhp);
1230 if (dchild && !IS_ERR(dchild))
1235 err = nfserrno(host_err);
1239 #ifdef CONFIG_NFSD_V3
1241 * NFSv3 version of nfsd_create
1244 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1245 char *fname, int flen, struct iattr *iap,
1246 struct svc_fh *resfhp, int createmode, u32 *verifier,
1247 int *truncp, int *created)
1249 struct dentry *dentry, *dchild = NULL;
1253 __u32 v_mtime=0, v_atime=0;
1259 if (isdotent(fname, flen))
1261 if (!(iap->ia_valid & ATTR_MODE))
1263 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1267 dentry = fhp->fh_dentry;
1268 dirp = dentry->d_inode;
1270 /* Get all the sanity checks out of the way before
1271 * we lock the parent. */
1272 err = nfserr_notdir;
1273 if(!dirp->i_op || !dirp->i_op->lookup)
1275 fh_lock_nested(fhp, I_MUTEX_PARENT);
1278 * Compose the response file handle.
1280 dchild = lookup_one_len(fname, dentry, flen);
1281 host_err = PTR_ERR(dchild);
1285 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1289 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1290 /* solaris7 gets confused (bugid 4218508) if these have
1291 * the high bit set, so just clear the high bits.
1293 v_mtime = verifier[0]&0x7fffffff;
1294 v_atime = verifier[1]&0x7fffffff;
1297 if (dchild->d_inode) {
1300 switch (createmode) {
1301 case NFS3_CREATE_UNCHECKED:
1302 if (! S_ISREG(dchild->d_inode->i_mode))
1305 /* in nfsv4, we need to treat this case a little
1306 * differently. we don't want to truncate the
1307 * file now; this would be wrong if the OPEN
1308 * fails for some other reason. furthermore,
1309 * if the size is nonzero, we should ignore it
1310 * according to spec!
1312 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1315 iap->ia_valid &= ATTR_SIZE;
1319 case NFS3_CREATE_EXCLUSIVE:
1320 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1321 && dchild->d_inode->i_atime.tv_sec == v_atime
1322 && dchild->d_inode->i_size == 0 )
1325 case NFS3_CREATE_GUARDED:
1331 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1337 if (EX_ISSYNC(fhp->fh_export)) {
1338 err = nfserrno(nfsd_sync_dir(dentry));
1339 /* setattr will sync the child (or not) */
1342 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1343 /* Cram the verifier into atime/mtime */
1344 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1345 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1346 /* XXX someone who knows this better please fix it for nsec */
1347 iap->ia_mtime.tv_sec = v_mtime;
1348 iap->ia_atime.tv_sec = v_atime;
1349 iap->ia_mtime.tv_nsec = 0;
1350 iap->ia_atime.tv_nsec = 0;
1353 /* Set file attributes.
1354 * Irix appears to send along the gid when it tries to
1355 * implement setgid directories via NFS. Clear out all that cruft.
1358 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1359 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1365 * Update the filehandle to get the new inode info.
1368 err = fh_update(resfhp);
1372 if (dchild && !IS_ERR(dchild))
1377 err = nfserrno(host_err);
1380 #endif /* CONFIG_NFSD_V3 */
1383 * Read a symlink. On entry, *lenp must contain the maximum path length that
1384 * fits into the buffer. On return, it contains the true length.
1385 * N.B. After this call fhp needs an fh_put
1388 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1390 struct dentry *dentry;
1391 struct inode *inode;
1396 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1400 dentry = fhp->fh_dentry;
1401 inode = dentry->d_inode;
1404 if (!inode->i_op || !inode->i_op->readlink)
1407 touch_atime(fhp->fh_export->ex_mnt, dentry);
1408 /* N.B. Why does this call need a get_fs()??
1409 * Remove the set_fs and watch the fireworks:-) --okir
1412 oldfs = get_fs(); set_fs(KERNEL_DS);
1413 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1424 err = nfserrno(host_err);
1429 * Create a symlink and look up its inode
1430 * N.B. After this call _both_ fhp and resfhp need an fh_put
1433 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1434 char *fname, int flen,
1435 char *path, int plen,
1436 struct svc_fh *resfhp,
1439 struct dentry *dentry, *dnew;
1448 if (isdotent(fname, flen))
1451 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1455 dentry = fhp->fh_dentry;
1456 dnew = lookup_one_len(fname, dentry, flen);
1457 host_err = PTR_ERR(dnew);
1462 /* Only the MODE ATTRibute is even vaguely meaningful */
1463 if (iap && (iap->ia_valid & ATTR_MODE))
1464 mode = iap->ia_mode & S_IALLUGO;
1466 if (unlikely(path[plen] != 0)) {
1467 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1468 if (path_alloced == NULL)
1471 strncpy(path_alloced, path, plen);
1472 path_alloced[plen] = 0;
1473 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1474 kfree(path_alloced);
1477 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1480 if (EX_ISSYNC(fhp->fh_export))
1481 host_err = nfsd_sync_dir(dentry);
1483 err = nfserrno(host_err);
1486 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1488 if (err==0) err = cerr;
1493 err = nfserrno(host_err);
1499 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1502 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1503 char *name, int len, struct svc_fh *tfhp)
1505 struct dentry *ddir, *dnew, *dold;
1506 struct inode *dirp, *dest;
1510 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1513 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1521 if (isdotent(name, len))
1524 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1525 ddir = ffhp->fh_dentry;
1526 dirp = ddir->d_inode;
1528 dnew = lookup_one_len(name, ddir, len);
1529 host_err = PTR_ERR(dnew);
1533 dold = tfhp->fh_dentry;
1534 dest = dold->d_inode;
1536 host_err = vfs_link(dold, dirp, dnew);
1538 if (EX_ISSYNC(ffhp->fh_export)) {
1539 err = nfserrno(nfsd_sync_dir(ddir));
1540 write_inode_now(dest, 1);
1544 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1547 err = nfserrno(host_err);
1557 err = nfserrno(host_err);
1563 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1566 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1567 struct svc_fh *tfhp, char *tname, int tlen)
1569 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1570 struct inode *fdir, *tdir;
1574 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1577 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1581 fdentry = ffhp->fh_dentry;
1582 fdir = fdentry->d_inode;
1584 tdentry = tfhp->fh_dentry;
1585 tdir = tdentry->d_inode;
1587 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1588 if (ffhp->fh_export != tfhp->fh_export)
1592 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1595 /* cannot use fh_lock as we need deadlock protective ordering
1596 * so do it by hand */
1597 trap = lock_rename(tdentry, fdentry);
1598 ffhp->fh_locked = tfhp->fh_locked = 1;
1602 odentry = lookup_one_len(fname, fdentry, flen);
1603 host_err = PTR_ERR(odentry);
1604 if (IS_ERR(odentry))
1608 if (!odentry->d_inode)
1611 if (odentry == trap)
1614 ndentry = lookup_one_len(tname, tdentry, tlen);
1615 host_err = PTR_ERR(ndentry);
1616 if (IS_ERR(ndentry))
1618 host_err = -ENOTEMPTY;
1619 if (ndentry == trap)
1623 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1624 ((atomic_read(&odentry->d_count) > 1)
1625 || (atomic_read(&ndentry->d_count) > 1))) {
1629 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1630 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1631 host_err = nfsd_sync_dir(tdentry);
1633 host_err = nfsd_sync_dir(fdentry);
1641 err = nfserrno(host_err);
1643 /* we cannot reply on fh_unlock on the two filehandles,
1644 * as that would do the wrong thing if the two directories
1645 * were the same, so again we do it by hand
1647 fill_post_wcc(ffhp);
1648 fill_post_wcc(tfhp);
1649 unlock_rename(tdentry, fdentry);
1650 ffhp->fh_locked = tfhp->fh_locked = 0;
1657 * Unlink a file or directory
1658 * N.B. After this call fhp needs an fh_put
1661 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1662 char *fname, int flen)
1664 struct dentry *dentry, *rdentry;
1670 if (!flen || isdotent(fname, flen))
1672 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1676 fh_lock_nested(fhp, I_MUTEX_PARENT);
1677 dentry = fhp->fh_dentry;
1678 dirp = dentry->d_inode;
1680 rdentry = lookup_one_len(fname, dentry, flen);
1681 host_err = PTR_ERR(rdentry);
1682 if (IS_ERR(rdentry))
1685 if (!rdentry->d_inode) {
1692 type = rdentry->d_inode->i_mode & S_IFMT;
1694 if (type != S_IFDIR) { /* It's UNLINK */
1696 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1697 (atomic_read(&rdentry->d_count) > 1)) {
1701 host_err = vfs_unlink(dirp, rdentry);
1702 } else { /* It's RMDIR */
1703 host_err = vfs_rmdir(dirp, rdentry);
1710 if (EX_ISSYNC(fhp->fh_export))
1711 host_err = nfsd_sync_dir(dentry);
1714 err = nfserrno(host_err);
1720 * Read entries from a directory.
1721 * The NFSv3/4 verifier we ignore for now.
1724 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1725 struct readdir_cd *cdp, filldir_t func)
1730 loff_t offset = *offsetp;
1732 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1736 offset = vfs_llseek(file, offset, 0);
1738 err = nfserrno((int)offset);
1743 * Read the directory entries. This silly loop is necessary because
1744 * readdir() is not guaranteed to fill up the entire buffer, but
1745 * may choose to do less.
1749 cdp->err = nfserr_eof; /* will be cleared on successful read */
1750 host_err = vfs_readdir(file, func, cdp);
1751 } while (host_err >=0 && cdp->err == nfs_ok);
1753 err = nfserrno(host_err);
1756 *offsetp = vfs_llseek(file, 0, 1);
1758 if (err == nfserr_eof || err == nfserr_toosmall)
1759 err = nfs_ok; /* can still be found in ->err */
1767 * Get file system stats
1768 * N.B. After this call fhp needs an fh_put
1771 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1773 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1774 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1780 * Check for a user's access permissions to this inode.
1783 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1785 struct inode *inode = dentry->d_inode;
1791 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1793 (acc & MAY_READ)? " read" : "",
1794 (acc & MAY_WRITE)? " write" : "",
1795 (acc & MAY_EXEC)? " exec" : "",
1796 (acc & MAY_SATTR)? " sattr" : "",
1797 (acc & MAY_TRUNC)? " trunc" : "",
1798 (acc & MAY_LOCK)? " lock" : "",
1799 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1801 IS_IMMUTABLE(inode)? " immut" : "",
1802 IS_APPEND(inode)? " append" : "",
1803 IS_RDONLY(inode)? " ro" : "");
1804 dprintk(" owner %d/%d user %d/%d\n",
1805 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1808 /* Normally we reject any write/sattr etc access on a read-only file
1809 * system. But if it is IRIX doing check on write-access for a
1810 * device special file, we ignore rofs.
1812 if (!(acc & MAY_LOCAL_ACCESS))
1813 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1814 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1816 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1819 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1822 if (acc & MAY_LOCK) {
1823 /* If we cannot rely on authentication in NLM requests,
1824 * just allow locks, otherwise require read permission, or
1827 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1830 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1833 * The file owner always gets access permission for accesses that
1834 * would normally be checked at open time. This is to make
1835 * file access work even when the client has done a fchmod(fd, 0).
1837 * However, `cp foo bar' should fail nevertheless when bar is
1838 * readonly. A sensible way to do this might be to reject all
1839 * attempts to truncate a read-only file, because a creat() call
1840 * always implies file truncation.
1841 * ... but this isn't really fair. A process may reasonably call
1842 * ftruncate on an open file descriptor on a file with perm 000.
1843 * We must trust the client to do permission checking - using "ACCESS"
1846 if ((acc & MAY_OWNER_OVERRIDE) &&
1847 inode->i_uid == current->fsuid)
1850 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1852 /* Allow read access to binaries even when mode 111 */
1853 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1854 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1855 err = permission(inode, MAY_EXEC, NULL);
1857 return err? nfserrno(err) : 0;
1861 nfsd_racache_shutdown(void)
1865 dprintk("nfsd: freeing readahead buffers.\n");
1870 * Initialize readahead param cache
1873 nfsd_racache_init(int cache_size)
1882 if (cache_size < 2*RAPARM_HASH_SIZE)
1883 cache_size = 2*RAPARM_HASH_SIZE;
1884 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1888 "nfsd: Could not allocate memory read-ahead cache.\n");
1892 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1893 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1894 raparm_hash[i].pb_head = NULL;
1895 spin_lock_init(&raparm_hash[i].pb_lock);
1897 nperbucket = cache_size >> RAPARM_HASH_BITS;
1898 for (i = 0; i < cache_size - 1; i++) {
1899 if (i % nperbucket == 0)
1900 raparm_hash[j++].pb_head = raparml + i;
1901 if (i % nperbucket < nperbucket-1)
1902 raparml[i].p_next = raparml + i + 1;
1905 nfsdstats.ra_size = cache_size;
1909 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1911 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1913 struct inode *inode = fhp->fh_dentry->d_inode;
1917 struct posix_acl *acl;
1919 if (!IS_POSIXACL(inode))
1920 return ERR_PTR(-EOPNOTSUPP);
1923 case ACL_TYPE_ACCESS:
1924 name = POSIX_ACL_XATTR_ACCESS;
1926 case ACL_TYPE_DEFAULT:
1927 name = POSIX_ACL_XATTR_DEFAULT;
1930 return ERR_PTR(-EOPNOTSUPP);
1933 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1935 return ERR_PTR(size);
1937 acl = posix_acl_from_xattr(value, size);
1943 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1945 struct inode *inode = fhp->fh_dentry->d_inode;
1951 if (!IS_POSIXACL(inode) || !inode->i_op ||
1952 !inode->i_op->setxattr || !inode->i_op->removexattr)
1955 case ACL_TYPE_ACCESS:
1956 name = POSIX_ACL_XATTR_ACCESS;
1958 case ACL_TYPE_DEFAULT:
1959 name = POSIX_ACL_XATTR_DEFAULT;
1965 if (acl && acl->a_count) {
1966 size = posix_acl_xattr_size(acl->a_count);
1967 value = kmalloc(size, GFP_KERNEL);
1970 error = posix_acl_to_xattr(acl, value, size);
1978 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1980 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1983 error = vfs_removexattr(fhp->fh_dentry, name);
1984 if (error == -ENODATA)
1993 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */