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;
173 /* Lookup the name, but don't follow links */
174 if (isdotent(name, len)) {
176 dentry = dget(dparent);
177 else if (dparent != exp->ex_dentry) {
178 dentry = dget_parent(dparent);
179 } else if (!EX_NOHIDE(exp))
180 dentry = dget(dparent); /* .. == . just like at / */
182 /* checking mountpoint crossing is very different when stepping up */
183 struct svc_export *exp2 = NULL;
185 struct vfsmount *mnt = mntget(exp->ex_mnt);
186 dentry = dget(dparent);
187 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
189 dp = dget_parent(dentry);
193 exp2 = exp_parent(exp->ex_client, mnt, dentry,
195 if (PTR_ERR(exp2) == -ENOENT) {
197 dentry = dget(dparent);
198 } else if (IS_ERR(exp2)) {
199 host_err = PTR_ERR(exp2);
211 dentry = lookup_one_len(name, dparent, len);
212 host_err = PTR_ERR(dentry);
216 * check if we have crossed a mount point ...
218 if (d_mountpoint(dentry)) {
219 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
226 * Note: we compose the file handle now, but as the
227 * dentry may be negative, it may need to be updated.
229 err = fh_compose(resfh, exp, dentry, fhp);
230 if (!err && !dentry->d_inode)
238 err = nfserrno(host_err);
243 * Set various file attributes.
244 * N.B. After this call fhp needs an fh_put
247 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
248 int check_guard, time_t guardtime)
250 struct dentry *dentry;
252 int accmode = MAY_SATTR;
259 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
260 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
261 if (iap->ia_valid & ATTR_SIZE)
265 err = fh_verify(rqstp, fhp, ftype, accmode);
269 dentry = fhp->fh_dentry;
270 inode = dentry->d_inode;
272 /* Ignore any mode updates on symlinks */
273 if (S_ISLNK(inode->i_mode))
274 iap->ia_valid &= ~ATTR_MODE;
279 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
280 * which only requires access, and "set-[ac]time-to-X" which
281 * requires ownership.
282 * So if it looks like it might be "set both to the same time which
283 * is close to now", and if inode_change_ok fails, then we
284 * convert to "set to now" instead of "set to explicit time"
286 * We only call inode_change_ok as the last test as technically
287 * it is not an interface that we should be using. It is only
288 * valid if the filesystem does not define it's own i_op->setattr.
290 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
291 #define MAX_TOUCH_TIME_ERROR (30*60)
292 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
293 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
295 /* Looks probable. Now just make sure time is in the right ballpark.
296 * Solaris, at least, doesn't seem to care what the time request is.
297 * We require it be within 30 minutes of now.
299 time_t delta = iap->ia_atime.tv_sec - get_seconds();
300 if (delta<0) delta = -delta;
301 if (delta < MAX_TOUCH_TIME_ERROR &&
302 inode_change_ok(inode, iap) != 0) {
303 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
304 * this will cause notify_change to set these times to "now"
306 iap->ia_valid &= ~BOTH_TIME_SET;
310 /* The size case is special. It changes the file as well as the attributes. */
311 if (iap->ia_valid & ATTR_SIZE) {
312 if (iap->ia_size < inode->i_size) {
313 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
319 * If we are changing the size of the file, then
320 * we need to break all leases.
322 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
323 if (host_err == -EWOULDBLOCK)
324 host_err = -ETIMEDOUT;
325 if (host_err) /* ENOMEM or EWOULDBLOCK */
328 host_err = get_write_access(inode);
333 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
335 put_write_access(inode);
341 imode = inode->i_mode;
342 if (iap->ia_valid & ATTR_MODE) {
343 iap->ia_mode &= S_IALLUGO;
344 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
347 /* Revoke setuid/setgid bit on chown/chgrp */
348 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
349 iap->ia_valid |= ATTR_KILL_SUID;
350 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
351 iap->ia_valid |= ATTR_KILL_SGID;
353 /* Change the attributes. */
355 iap->ia_valid |= ATTR_CTIME;
357 err = nfserr_notsync;
358 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
360 host_err = notify_change(dentry, iap);
361 err = nfserrno(host_err);
365 put_write_access(inode);
367 if (EX_ISSYNC(fhp->fh_export))
368 write_inode_now(inode, 1);
373 err = nfserrno(host_err);
377 #if defined(CONFIG_NFSD_V2_ACL) || \
378 defined(CONFIG_NFSD_V3_ACL) || \
379 defined(CONFIG_NFSD_V4)
380 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
384 buflen = vfs_getxattr(dentry, key, NULL, 0);
388 *buf = kmalloc(buflen, GFP_KERNEL);
392 return vfs_getxattr(dentry, key, *buf, buflen);
396 #if defined(CONFIG_NFSD_V4)
398 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
405 buflen = posix_acl_xattr_size(pacl->a_count);
406 buf = kmalloc(buflen, GFP_KERNEL);
411 len = posix_acl_to_xattr(pacl, buf, buflen);
417 error = vfs_setxattr(dentry, key, buf, len, 0);
424 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
425 struct nfs4_acl *acl)
429 struct dentry *dentry;
431 struct posix_acl *pacl = NULL, *dpacl = NULL;
432 unsigned int flags = 0;
435 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
439 dentry = fhp->fh_dentry;
440 inode = dentry->d_inode;
441 if (S_ISDIR(inode->i_mode))
442 flags = NFS4_ACL_DIR;
444 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
445 if (host_error == -EINVAL) {
446 return nfserr_attrnotsupp;
447 } else if (host_error < 0)
450 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
454 if (S_ISDIR(inode->i_mode))
455 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
458 posix_acl_release(pacl);
459 posix_acl_release(dpacl);
461 if (host_error == -EOPNOTSUPP)
462 return nfserr_attrnotsupp;
464 return nfserrno(host_error);
467 static struct posix_acl *
468 _get_posix_acl(struct dentry *dentry, char *key)
471 struct posix_acl *pacl = NULL;
474 buflen = nfsd_getxattr(dentry, key, &buf);
478 return ERR_PTR(buflen);
480 pacl = posix_acl_from_xattr(buf, buflen);
486 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
488 struct inode *inode = dentry->d_inode;
490 struct posix_acl *pacl = NULL, *dpacl = NULL;
491 unsigned int flags = 0;
493 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
494 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
495 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
497 error = PTR_ERR(pacl);
502 if (S_ISDIR(inode->i_mode)) {
503 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
504 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
506 else if (IS_ERR(dpacl)) {
507 error = PTR_ERR(dpacl);
511 flags = NFS4_ACL_DIR;
514 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
516 error = PTR_ERR(*acl);
520 posix_acl_release(pacl);
521 posix_acl_release(dpacl);
525 #endif /* defined(CONFIG_NFS_V4) */
527 #ifdef CONFIG_NFSD_V3
529 * Check server access rights to a file system object
535 static struct accessmap nfs3_regaccess[] = {
536 { NFS3_ACCESS_READ, MAY_READ },
537 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
538 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
539 { NFS3_ACCESS_EXTEND, MAY_WRITE },
544 static struct accessmap nfs3_diraccess[] = {
545 { NFS3_ACCESS_READ, MAY_READ },
546 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
547 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
548 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
549 { NFS3_ACCESS_DELETE, MAY_REMOVE },
554 static struct accessmap nfs3_anyaccess[] = {
555 /* Some clients - Solaris 2.6 at least, make an access call
556 * to the server to check for access for things like /dev/null
557 * (which really, the server doesn't care about). So
558 * We provide simple access checking for them, looking
559 * mainly at mode bits, and we make sure to ignore read-only
562 { NFS3_ACCESS_READ, MAY_READ },
563 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
564 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
565 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
571 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
573 struct accessmap *map;
574 struct svc_export *export;
575 struct dentry *dentry;
576 u32 query, result = 0, sresult = 0;
579 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
583 export = fhp->fh_export;
584 dentry = fhp->fh_dentry;
586 if (S_ISREG(dentry->d_inode->i_mode))
587 map = nfs3_regaccess;
588 else if (S_ISDIR(dentry->d_inode->i_mode))
589 map = nfs3_diraccess;
591 map = nfs3_anyaccess;
595 for (; map->access; map++) {
596 if (map->access & query) {
599 sresult |= map->access;
601 err2 = nfsd_permission(export, dentry, map->how);
604 result |= map->access;
607 /* the following error codes just mean the access was not allowed,
608 * rather than an error occurred */
612 /* simply don't "or" in the access bit. */
622 *supported = sresult;
627 #endif /* CONFIG_NFSD_V3 */
632 * Open an existing file or directory.
633 * The access argument indicates the type of open (read/write/lock)
634 * N.B. After this call fhp needs an fh_put
637 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
638 int access, struct file **filp)
640 struct dentry *dentry;
642 int flags = O_RDONLY|O_LARGEFILE;
647 * If we get here, then the client has already done an "open",
648 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
649 * in case a chmod has now revoked permission.
651 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
655 dentry = fhp->fh_dentry;
656 inode = dentry->d_inode;
658 /* Disallow write access to files with the append-only bit set
659 * or any access when mandatory locking enabled
662 if (IS_APPEND(inode) && (access & MAY_WRITE))
664 if (IS_ISMNDLK(inode))
671 * Check to see if there are any leases on this file.
672 * This may block while leases are broken.
674 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
675 if (host_err == -EWOULDBLOCK)
676 host_err = -ETIMEDOUT;
677 if (host_err) /* NOMEM or WOULDBLOCK */
680 if (access & MAY_WRITE) {
681 if (access & MAY_READ)
682 flags = O_RDWR|O_LARGEFILE;
684 flags = O_WRONLY|O_LARGEFILE;
688 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
690 host_err = PTR_ERR(*filp);
692 err = nfserrno(host_err);
701 nfsd_close(struct file *filp)
708 * As this calls fsync (not fdatasync) there is no need for a write_inode
711 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
712 const struct file_operations *fop)
714 struct inode *inode = dp->d_inode;
715 int (*fsync) (struct file *, struct dentry *, int);
718 err = filemap_fdatawrite(inode->i_mapping);
719 if (err == 0 && fop && (fsync = fop->fsync))
720 err = fsync(filp, dp, 0);
722 err = filemap_fdatawait(inode->i_mapping);
729 nfsd_sync(struct file *filp)
732 struct inode *inode = filp->f_path.dentry->d_inode;
733 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
734 mutex_lock(&inode->i_mutex);
735 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
736 mutex_unlock(&inode->i_mutex);
742 nfsd_sync_dir(struct dentry *dp)
744 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
748 * Obtain the readahead parameters for the file
749 * specified by (dev, ino).
752 static inline struct raparms *
753 nfsd_get_raparms(dev_t dev, ino_t ino)
755 struct raparms *ra, **rap, **frap = NULL;
758 struct raparm_hbucket *rab;
760 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
761 rab = &raparm_hash[hash];
763 spin_lock(&rab->pb_lock);
764 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
765 if (ra->p_ino == ino && ra->p_dev == dev)
768 if (ra->p_count == 0)
771 depth = nfsdstats.ra_size*11/10;
773 spin_unlock(&rab->pb_lock);
783 if (rap != &rab->pb_head) {
785 ra->p_next = rab->pb_head;
789 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
790 spin_unlock(&rab->pb_lock);
795 * Grab and keep cached pages associated with a file in the svc_rqst
796 * so that they can be passed to the network sendmsg/sendpage routines
797 * directly. They will be released after the sending has completed.
800 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
801 struct splice_desc *sd)
803 struct svc_rqst *rqstp = sd->u.data;
804 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
805 struct page *page = buf->page;
809 ret = buf->ops->confirm(pipe, buf);
815 if (rqstp->rq_res.page_len == 0) {
820 rqstp->rq_res.page_base = buf->offset;
821 rqstp->rq_res.page_len = size;
822 } else if (page != pp[-1]) {
828 rqstp->rq_res.page_len += size;
830 rqstp->rq_res.page_len += size;
835 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
836 struct splice_desc *sd)
838 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
842 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
843 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
852 inode = file->f_path.dentry->d_inode;
854 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
855 (!lock_may_read(inode, offset, *count)))
859 /* Get readahead parameters */
860 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
863 file->f_ra = ra->p_ra;
865 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
866 struct splice_desc sd = {
873 rqstp->rq_resused = 1;
874 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
878 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
882 /* Write back readahead params */
884 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
885 spin_lock(&rab->pb_lock);
886 ra->p_ra = file->f_ra;
889 spin_unlock(&rab->pb_lock);
893 nfsdstats.io_read += host_err;
896 fsnotify_access(file->f_path.dentry);
898 err = nfserrno(host_err);
903 static void kill_suid(struct dentry *dentry)
906 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
908 mutex_lock(&dentry->d_inode->i_mutex);
909 notify_change(dentry, &ia);
910 mutex_unlock(&dentry->d_inode->i_mutex);
914 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
915 loff_t offset, struct kvec *vec, int vlen,
916 unsigned long cnt, int *stablep)
918 struct svc_export *exp;
919 struct dentry *dentry;
924 int stable = *stablep;
929 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
930 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
934 dentry = file->f_path.dentry;
935 inode = dentry->d_inode;
936 exp = fhp->fh_export;
939 * Request sync writes if
940 * - the sync export option has been set, or
941 * - the client requested O_SYNC behavior (NFSv3 feature).
942 * - The file system doesn't support fsync().
943 * When gathered writes have been configured for this volume,
944 * flushing the data to disk is handled separately below.
947 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
949 *stablep = 2; /* FILE_SYNC */
954 if (stable && !EX_WGATHER(exp))
955 file->f_flags |= O_SYNC;
957 /* Write the data. */
958 oldfs = get_fs(); set_fs(KERNEL_DS);
959 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
962 nfsdstats.io_write += cnt;
963 fsnotify_modify(file->f_path.dentry);
966 /* clear setuid/setgid flag after write */
967 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
970 if (host_err >= 0 && stable) {
971 static ino_t last_ino;
972 static dev_t last_dev;
975 * Gathered writes: If another process is currently
976 * writing to the file, there's a high chance
977 * this is another nfsd (triggered by a bulk write
978 * from a client's biod). Rather than syncing the
979 * file with each write request, we sleep for 10 msec.
981 * I don't know if this roughly approximates
982 * C. Juszak's idea of gathered writes, but it's a
983 * nice and simple solution (IMHO), and it seems to
986 if (EX_WGATHER(exp)) {
987 if (atomic_read(&inode->i_writecount) > 1
988 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
989 dprintk("nfsd: write defer %d\n", current->pid);
991 dprintk("nfsd: write resume %d\n", current->pid);
994 if (inode->i_state & I_DIRTY) {
995 dprintk("nfsd: write sync %d\n", current->pid);
996 host_err=nfsd_sync(file);
999 wake_up(&inode->i_wait);
1002 last_ino = inode->i_ino;
1003 last_dev = inode->i_sb->s_dev;
1006 dprintk("nfsd: write complete host_err=%d\n", host_err);
1010 err = nfserrno(host_err);
1016 * Read data from a file. count must contain the requested read count
1017 * on entry. On return, *count contains the number of bytes actually read.
1018 * N.B. After this call fhp needs an fh_put
1021 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1022 loff_t offset, struct kvec *vec, int vlen,
1023 unsigned long *count)
1028 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1029 MAY_READ|MAY_OWNER_OVERRIDE);
1032 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1034 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1037 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1045 * Write data to a file.
1046 * The stable flag requests synchronous writes.
1047 * N.B. After this call fhp needs an fh_put
1050 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1051 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1057 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1058 MAY_WRITE|MAY_OWNER_OVERRIDE);
1061 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1064 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1069 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1077 #ifdef CONFIG_NFSD_V3
1079 * Commit all pending writes to stable storage.
1080 * Strictly speaking, we could sync just the indicated file region here,
1081 * but there's currently no way we can ask the VFS to do so.
1083 * Unfortunately we cannot lock the file to make sure we return full WCC
1084 * data to the client, as locking happens lower down in the filesystem.
1087 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1088 loff_t offset, unsigned long count)
1093 if ((u64)count > ~(u64)offset)
1094 return nfserr_inval;
1096 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1098 if (EX_ISSYNC(fhp->fh_export)) {
1099 if (file->f_op && file->f_op->fsync) {
1100 err = nfserrno(nfsd_sync(file));
1102 err = nfserr_notsupp;
1109 #endif /* CONFIG_NFSD_V3 */
1112 * Create a file (regular, directory, device, fifo); UNIX sockets
1113 * not yet implemented.
1114 * If the response fh has been verified, the parent directory should
1115 * already be locked. Note that the parent directory is left locked.
1117 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1120 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1121 char *fname, int flen, struct iattr *iap,
1122 int type, dev_t rdev, struct svc_fh *resfhp)
1124 struct dentry *dentry, *dchild = NULL;
1133 if (isdotent(fname, flen))
1136 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1140 dentry = fhp->fh_dentry;
1141 dirp = dentry->d_inode;
1143 err = nfserr_notdir;
1144 if(!dirp->i_op || !dirp->i_op->lookup)
1147 * Check whether the response file handle has been verified yet.
1148 * If it has, the parent directory should already be locked.
1150 if (!resfhp->fh_dentry) {
1151 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1152 fh_lock_nested(fhp, I_MUTEX_PARENT);
1153 dchild = lookup_one_len(fname, dentry, flen);
1154 host_err = PTR_ERR(dchild);
1157 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1161 /* called from nfsd_proc_create */
1162 dchild = dget(resfhp->fh_dentry);
1163 if (!fhp->fh_locked) {
1164 /* not actually possible */
1166 "nfsd_create: parent %s/%s not locked!\n",
1167 dentry->d_parent->d_name.name,
1168 dentry->d_name.name);
1174 * Make sure the child dentry is still negative ...
1177 if (dchild->d_inode) {
1178 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1179 dentry->d_name.name, dchild->d_name.name);
1183 if (!(iap->ia_valid & ATTR_MODE))
1185 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1188 * Get the dir op function pointer.
1193 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1196 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1202 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1205 printk("nfsd: bad file type %o in nfsd_create\n", type);
1211 if (EX_ISSYNC(fhp->fh_export)) {
1212 err = nfserrno(nfsd_sync_dir(dentry));
1213 write_inode_now(dchild->d_inode, 1);
1217 /* Set file attributes. Mode has already been set and
1218 * setting uid/gid works only for root. Irix appears to
1219 * send along the gid when it tries to implement setgid
1220 * directories via NFS.
1222 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1223 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1228 * Update the file handle to get the new inode info.
1231 err = fh_update(resfhp);
1233 if (dchild && !IS_ERR(dchild))
1238 err = nfserrno(host_err);
1242 #ifdef CONFIG_NFSD_V3
1244 * NFSv3 version of nfsd_create
1247 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1248 char *fname, int flen, struct iattr *iap,
1249 struct svc_fh *resfhp, int createmode, u32 *verifier,
1250 int *truncp, int *created)
1252 struct dentry *dentry, *dchild = NULL;
1256 __u32 v_mtime=0, v_atime=0;
1262 if (isdotent(fname, flen))
1264 if (!(iap->ia_valid & ATTR_MODE))
1266 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1270 dentry = fhp->fh_dentry;
1271 dirp = dentry->d_inode;
1273 /* Get all the sanity checks out of the way before
1274 * we lock the parent. */
1275 err = nfserr_notdir;
1276 if(!dirp->i_op || !dirp->i_op->lookup)
1278 fh_lock_nested(fhp, I_MUTEX_PARENT);
1281 * Compose the response file handle.
1283 dchild = lookup_one_len(fname, dentry, flen);
1284 host_err = PTR_ERR(dchild);
1288 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1292 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1293 /* solaris7 gets confused (bugid 4218508) if these have
1294 * the high bit set, so just clear the high bits.
1296 v_mtime = verifier[0]&0x7fffffff;
1297 v_atime = verifier[1]&0x7fffffff;
1300 if (dchild->d_inode) {
1303 switch (createmode) {
1304 case NFS3_CREATE_UNCHECKED:
1305 if (! S_ISREG(dchild->d_inode->i_mode))
1308 /* in nfsv4, we need to treat this case a little
1309 * differently. we don't want to truncate the
1310 * file now; this would be wrong if the OPEN
1311 * fails for some other reason. furthermore,
1312 * if the size is nonzero, we should ignore it
1313 * according to spec!
1315 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1318 iap->ia_valid &= ATTR_SIZE;
1322 case NFS3_CREATE_EXCLUSIVE:
1323 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1324 && dchild->d_inode->i_atime.tv_sec == v_atime
1325 && dchild->d_inode->i_size == 0 )
1328 case NFS3_CREATE_GUARDED:
1334 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1340 if (EX_ISSYNC(fhp->fh_export)) {
1341 err = nfserrno(nfsd_sync_dir(dentry));
1342 /* setattr will sync the child (or not) */
1345 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1346 /* Cram the verifier into atime/mtime */
1347 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1348 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1349 /* XXX someone who knows this better please fix it for nsec */
1350 iap->ia_mtime.tv_sec = v_mtime;
1351 iap->ia_atime.tv_sec = v_atime;
1352 iap->ia_mtime.tv_nsec = 0;
1353 iap->ia_atime.tv_nsec = 0;
1356 /* Set file attributes.
1357 * Irix appears to send along the gid when it tries to
1358 * implement setgid directories via NFS. Clear out all that cruft.
1361 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1362 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1368 * Update the filehandle to get the new inode info.
1371 err = fh_update(resfhp);
1375 if (dchild && !IS_ERR(dchild))
1380 err = nfserrno(host_err);
1383 #endif /* CONFIG_NFSD_V3 */
1386 * Read a symlink. On entry, *lenp must contain the maximum path length that
1387 * fits into the buffer. On return, it contains the true length.
1388 * N.B. After this call fhp needs an fh_put
1391 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1393 struct dentry *dentry;
1394 struct inode *inode;
1399 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1403 dentry = fhp->fh_dentry;
1404 inode = dentry->d_inode;
1407 if (!inode->i_op || !inode->i_op->readlink)
1410 touch_atime(fhp->fh_export->ex_mnt, dentry);
1411 /* N.B. Why does this call need a get_fs()??
1412 * Remove the set_fs and watch the fireworks:-) --okir
1415 oldfs = get_fs(); set_fs(KERNEL_DS);
1416 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1427 err = nfserrno(host_err);
1432 * Create a symlink and look up its inode
1433 * N.B. After this call _both_ fhp and resfhp need an fh_put
1436 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1437 char *fname, int flen,
1438 char *path, int plen,
1439 struct svc_fh *resfhp,
1442 struct dentry *dentry, *dnew;
1451 if (isdotent(fname, flen))
1454 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1458 dentry = fhp->fh_dentry;
1459 dnew = lookup_one_len(fname, dentry, flen);
1460 host_err = PTR_ERR(dnew);
1465 /* Only the MODE ATTRibute is even vaguely meaningful */
1466 if (iap && (iap->ia_valid & ATTR_MODE))
1467 mode = iap->ia_mode & S_IALLUGO;
1469 if (unlikely(path[plen] != 0)) {
1470 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1471 if (path_alloced == NULL)
1474 strncpy(path_alloced, path, plen);
1475 path_alloced[plen] = 0;
1476 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1477 kfree(path_alloced);
1480 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1483 if (EX_ISSYNC(fhp->fh_export))
1484 host_err = nfsd_sync_dir(dentry);
1486 err = nfserrno(host_err);
1489 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1491 if (err==0) err = cerr;
1496 err = nfserrno(host_err);
1502 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1505 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1506 char *name, int len, struct svc_fh *tfhp)
1508 struct dentry *ddir, *dnew, *dold;
1509 struct inode *dirp, *dest;
1513 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1516 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1524 if (isdotent(name, len))
1527 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1528 ddir = ffhp->fh_dentry;
1529 dirp = ddir->d_inode;
1531 dnew = lookup_one_len(name, ddir, len);
1532 host_err = PTR_ERR(dnew);
1536 dold = tfhp->fh_dentry;
1537 dest = dold->d_inode;
1539 host_err = vfs_link(dold, dirp, dnew);
1541 if (EX_ISSYNC(ffhp->fh_export)) {
1542 err = nfserrno(nfsd_sync_dir(ddir));
1543 write_inode_now(dest, 1);
1547 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1550 err = nfserrno(host_err);
1560 err = nfserrno(host_err);
1566 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1569 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1570 struct svc_fh *tfhp, char *tname, int tlen)
1572 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1573 struct inode *fdir, *tdir;
1577 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1580 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1584 fdentry = ffhp->fh_dentry;
1585 fdir = fdentry->d_inode;
1587 tdentry = tfhp->fh_dentry;
1588 tdir = tdentry->d_inode;
1590 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1591 if (ffhp->fh_export != tfhp->fh_export)
1595 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1598 /* cannot use fh_lock as we need deadlock protective ordering
1599 * so do it by hand */
1600 trap = lock_rename(tdentry, fdentry);
1601 ffhp->fh_locked = tfhp->fh_locked = 1;
1605 odentry = lookup_one_len(fname, fdentry, flen);
1606 host_err = PTR_ERR(odentry);
1607 if (IS_ERR(odentry))
1611 if (!odentry->d_inode)
1614 if (odentry == trap)
1617 ndentry = lookup_one_len(tname, tdentry, tlen);
1618 host_err = PTR_ERR(ndentry);
1619 if (IS_ERR(ndentry))
1621 host_err = -ENOTEMPTY;
1622 if (ndentry == trap)
1626 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1627 ((atomic_read(&odentry->d_count) > 1)
1628 || (atomic_read(&ndentry->d_count) > 1))) {
1632 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1633 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1634 host_err = nfsd_sync_dir(tdentry);
1636 host_err = nfsd_sync_dir(fdentry);
1644 err = nfserrno(host_err);
1646 /* we cannot reply on fh_unlock on the two filehandles,
1647 * as that would do the wrong thing if the two directories
1648 * were the same, so again we do it by hand
1650 fill_post_wcc(ffhp);
1651 fill_post_wcc(tfhp);
1652 unlock_rename(tdentry, fdentry);
1653 ffhp->fh_locked = tfhp->fh_locked = 0;
1660 * Unlink a file or directory
1661 * N.B. After this call fhp needs an fh_put
1664 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1665 char *fname, int flen)
1667 struct dentry *dentry, *rdentry;
1673 if (!flen || isdotent(fname, flen))
1675 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1679 fh_lock_nested(fhp, I_MUTEX_PARENT);
1680 dentry = fhp->fh_dentry;
1681 dirp = dentry->d_inode;
1683 rdentry = lookup_one_len(fname, dentry, flen);
1684 host_err = PTR_ERR(rdentry);
1685 if (IS_ERR(rdentry))
1688 if (!rdentry->d_inode) {
1695 type = rdentry->d_inode->i_mode & S_IFMT;
1697 if (type != S_IFDIR) { /* It's UNLINK */
1699 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1700 (atomic_read(&rdentry->d_count) > 1)) {
1704 host_err = vfs_unlink(dirp, rdentry);
1705 } else { /* It's RMDIR */
1706 host_err = vfs_rmdir(dirp, rdentry);
1713 if (EX_ISSYNC(fhp->fh_export))
1714 host_err = nfsd_sync_dir(dentry);
1717 err = nfserrno(host_err);
1723 * Read entries from a directory.
1724 * The NFSv3/4 verifier we ignore for now.
1727 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1728 struct readdir_cd *cdp, filldir_t func)
1733 loff_t offset = *offsetp;
1735 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1739 offset = vfs_llseek(file, offset, 0);
1741 err = nfserrno((int)offset);
1746 * Read the directory entries. This silly loop is necessary because
1747 * readdir() is not guaranteed to fill up the entire buffer, but
1748 * may choose to do less.
1752 cdp->err = nfserr_eof; /* will be cleared on successful read */
1753 host_err = vfs_readdir(file, func, cdp);
1754 } while (host_err >=0 && cdp->err == nfs_ok);
1756 err = nfserrno(host_err);
1759 *offsetp = vfs_llseek(file, 0, 1);
1761 if (err == nfserr_eof || err == nfserr_toosmall)
1762 err = nfs_ok; /* can still be found in ->err */
1770 * Get file system stats
1771 * N.B. After this call fhp needs an fh_put
1774 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1776 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1777 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1783 * Check for a user's access permissions to this inode.
1786 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1788 struct inode *inode = dentry->d_inode;
1794 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1796 (acc & MAY_READ)? " read" : "",
1797 (acc & MAY_WRITE)? " write" : "",
1798 (acc & MAY_EXEC)? " exec" : "",
1799 (acc & MAY_SATTR)? " sattr" : "",
1800 (acc & MAY_TRUNC)? " trunc" : "",
1801 (acc & MAY_LOCK)? " lock" : "",
1802 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1804 IS_IMMUTABLE(inode)? " immut" : "",
1805 IS_APPEND(inode)? " append" : "",
1806 IS_RDONLY(inode)? " ro" : "");
1807 dprintk(" owner %d/%d user %d/%d\n",
1808 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1811 /* Normally we reject any write/sattr etc access on a read-only file
1812 * system. But if it is IRIX doing check on write-access for a
1813 * device special file, we ignore rofs.
1815 if (!(acc & MAY_LOCAL_ACCESS))
1816 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1817 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1819 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1822 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1825 if (acc & MAY_LOCK) {
1826 /* If we cannot rely on authentication in NLM requests,
1827 * just allow locks, otherwise require read permission, or
1830 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1833 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1836 * The file owner always gets access permission for accesses that
1837 * would normally be checked at open time. This is to make
1838 * file access work even when the client has done a fchmod(fd, 0).
1840 * However, `cp foo bar' should fail nevertheless when bar is
1841 * readonly. A sensible way to do this might be to reject all
1842 * attempts to truncate a read-only file, because a creat() call
1843 * always implies file truncation.
1844 * ... but this isn't really fair. A process may reasonably call
1845 * ftruncate on an open file descriptor on a file with perm 000.
1846 * We must trust the client to do permission checking - using "ACCESS"
1849 if ((acc & MAY_OWNER_OVERRIDE) &&
1850 inode->i_uid == current->fsuid)
1853 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1855 /* Allow read access to binaries even when mode 111 */
1856 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1857 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1858 err = permission(inode, MAY_EXEC, NULL);
1860 return err? nfserrno(err) : 0;
1864 nfsd_racache_shutdown(void)
1868 dprintk("nfsd: freeing readahead buffers.\n");
1873 * Initialize readahead param cache
1876 nfsd_racache_init(int cache_size)
1885 if (cache_size < 2*RAPARM_HASH_SIZE)
1886 cache_size = 2*RAPARM_HASH_SIZE;
1887 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1891 "nfsd: Could not allocate memory read-ahead cache.\n");
1895 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1896 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1897 raparm_hash[i].pb_head = NULL;
1898 spin_lock_init(&raparm_hash[i].pb_lock);
1900 nperbucket = cache_size >> RAPARM_HASH_BITS;
1901 for (i = 0; i < cache_size - 1; i++) {
1902 if (i % nperbucket == 0)
1903 raparm_hash[j++].pb_head = raparml + i;
1904 if (i % nperbucket < nperbucket-1)
1905 raparml[i].p_next = raparml + i + 1;
1908 nfsdstats.ra_size = cache_size;
1912 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1914 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1916 struct inode *inode = fhp->fh_dentry->d_inode;
1920 struct posix_acl *acl;
1922 if (!IS_POSIXACL(inode))
1923 return ERR_PTR(-EOPNOTSUPP);
1926 case ACL_TYPE_ACCESS:
1927 name = POSIX_ACL_XATTR_ACCESS;
1929 case ACL_TYPE_DEFAULT:
1930 name = POSIX_ACL_XATTR_DEFAULT;
1933 return ERR_PTR(-EOPNOTSUPP);
1936 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1938 return ERR_PTR(size);
1940 acl = posix_acl_from_xattr(value, size);
1946 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1948 struct inode *inode = fhp->fh_dentry->d_inode;
1954 if (!IS_POSIXACL(inode) || !inode->i_op ||
1955 !inode->i_op->setxattr || !inode->i_op->removexattr)
1958 case ACL_TYPE_ACCESS:
1959 name = POSIX_ACL_XATTR_ACCESS;
1961 case ACL_TYPE_DEFAULT:
1962 name = POSIX_ACL_XATTR_DEFAULT;
1968 if (acl && acl->a_count) {
1969 size = posix_acl_xattr_size(acl->a_count);
1970 value = kmalloc(size, GFP_KERNEL);
1973 error = posix_acl_to_xattr(acl, value, size);
1981 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1983 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1986 error = vfs_removexattr(fhp->fh_dentry, name);
1987 if (error == -ENODATA)
1996 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */