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 ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
124 /* successfully crossed mount point */
139 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
140 const char *name, int len,
141 struct svc_export **exp_ret, struct dentry **dentry_ret)
143 struct svc_export *exp;
144 struct dentry *dparent;
145 struct dentry *dentry;
149 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
151 /* Obtain dentry and export. */
152 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
156 dparent = fhp->fh_dentry;
157 exp = fhp->fh_export;
160 /* Lookup the name, but don't follow links */
161 if (isdotent(name, len)) {
163 dentry = dget(dparent);
164 else if (dparent != exp->ex_dentry) {
165 dentry = dget_parent(dparent);
166 } else if (!EX_NOHIDE(exp))
167 dentry = dget(dparent); /* .. == . just like at / */
169 /* checking mountpoint crossing is very different when stepping up */
170 struct svc_export *exp2 = NULL;
172 struct vfsmount *mnt = mntget(exp->ex_mnt);
173 dentry = dget(dparent);
174 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
176 dp = dget_parent(dentry);
180 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
181 if (PTR_ERR(exp2) == -ENOENT) {
183 dentry = dget(dparent);
184 } else if (IS_ERR(exp2)) {
185 host_err = PTR_ERR(exp2);
197 dentry = lookup_one_len(name, dparent, len);
198 host_err = PTR_ERR(dentry);
202 * check if we have crossed a mount point ...
204 if (d_mountpoint(dentry)) {
205 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
211 *dentry_ret = dentry;
217 return nfserrno(host_err);
221 * Look up one component of a pathname.
222 * N.B. After this call _both_ fhp and resfh need an fh_put
224 * If the lookup would cross a mountpoint, and the mounted filesystem
225 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
226 * accepted as it stands and the mounted directory is
227 * returned. Otherwise the covered directory is returned.
228 * NOTE: this mountpoint crossing is not supported properly by all
229 * clients and is explicitly disallowed for NFSv3
230 * NeilBrown <neilb@cse.unsw.edu.au>
233 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
234 int len, struct svc_fh *resfh)
236 struct svc_export *exp;
237 struct dentry *dentry;
240 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
243 err = check_nfsd_access(exp, rqstp);
247 * Note: we compose the file handle now, but as the
248 * dentry may be negative, it may need to be updated.
250 err = fh_compose(resfh, exp, dentry, fhp);
251 if (!err && !dentry->d_inode)
261 * Set various file attributes.
262 * N.B. After this call fhp needs an fh_put
265 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
266 int check_guard, time_t guardtime)
268 struct dentry *dentry;
270 int accmode = MAY_SATTR;
277 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
278 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
279 if (iap->ia_valid & ATTR_SIZE)
283 err = fh_verify(rqstp, fhp, ftype, accmode);
287 dentry = fhp->fh_dentry;
288 inode = dentry->d_inode;
290 /* Ignore any mode updates on symlinks */
291 if (S_ISLNK(inode->i_mode))
292 iap->ia_valid &= ~ATTR_MODE;
297 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
298 * which only requires access, and "set-[ac]time-to-X" which
299 * requires ownership.
300 * So if it looks like it might be "set both to the same time which
301 * is close to now", and if inode_change_ok fails, then we
302 * convert to "set to now" instead of "set to explicit time"
304 * We only call inode_change_ok as the last test as technically
305 * it is not an interface that we should be using. It is only
306 * valid if the filesystem does not define it's own i_op->setattr.
308 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
309 #define MAX_TOUCH_TIME_ERROR (30*60)
310 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
311 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
313 /* Looks probable. Now just make sure time is in the right ballpark.
314 * Solaris, at least, doesn't seem to care what the time request is.
315 * We require it be within 30 minutes of now.
317 time_t delta = iap->ia_atime.tv_sec - get_seconds();
318 if (delta<0) delta = -delta;
319 if (delta < MAX_TOUCH_TIME_ERROR &&
320 inode_change_ok(inode, iap) != 0) {
321 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
322 * this will cause notify_change to set these times to "now"
324 iap->ia_valid &= ~BOTH_TIME_SET;
328 /* The size case is special. It changes the file as well as the attributes. */
329 if (iap->ia_valid & ATTR_SIZE) {
330 if (iap->ia_size < inode->i_size) {
331 err = nfsd_permission(rqstp, fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
337 * If we are changing the size of the file, then
338 * we need to break all leases.
340 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
341 if (host_err == -EWOULDBLOCK)
342 host_err = -ETIMEDOUT;
343 if (host_err) /* ENOMEM or EWOULDBLOCK */
346 host_err = get_write_access(inode);
351 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
353 put_write_access(inode);
359 imode = inode->i_mode;
360 if (iap->ia_valid & ATTR_MODE) {
361 iap->ia_mode &= S_IALLUGO;
362 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
365 /* Revoke setuid/setgid bit on chown/chgrp */
366 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
367 iap->ia_valid |= ATTR_KILL_SUID;
368 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
369 iap->ia_valid |= ATTR_KILL_SGID;
371 /* Change the attributes. */
373 iap->ia_valid |= ATTR_CTIME;
375 err = nfserr_notsync;
376 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
378 host_err = notify_change(dentry, iap);
379 err = nfserrno(host_err);
383 put_write_access(inode);
385 if (EX_ISSYNC(fhp->fh_export))
386 write_inode_now(inode, 1);
391 err = nfserrno(host_err);
395 #if defined(CONFIG_NFSD_V2_ACL) || \
396 defined(CONFIG_NFSD_V3_ACL) || \
397 defined(CONFIG_NFSD_V4)
398 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
402 buflen = vfs_getxattr(dentry, key, NULL, 0);
406 *buf = kmalloc(buflen, GFP_KERNEL);
410 return vfs_getxattr(dentry, key, *buf, buflen);
414 #if defined(CONFIG_NFSD_V4)
416 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
423 buflen = posix_acl_xattr_size(pacl->a_count);
424 buf = kmalloc(buflen, GFP_KERNEL);
429 len = posix_acl_to_xattr(pacl, buf, buflen);
435 error = vfs_setxattr(dentry, key, buf, len, 0);
442 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
443 struct nfs4_acl *acl)
447 struct dentry *dentry;
449 struct posix_acl *pacl = NULL, *dpacl = NULL;
450 unsigned int flags = 0;
453 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
457 dentry = fhp->fh_dentry;
458 inode = dentry->d_inode;
459 if (S_ISDIR(inode->i_mode))
460 flags = NFS4_ACL_DIR;
462 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
463 if (host_error == -EINVAL) {
464 return nfserr_attrnotsupp;
465 } else if (host_error < 0)
468 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
472 if (S_ISDIR(inode->i_mode))
473 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
476 posix_acl_release(pacl);
477 posix_acl_release(dpacl);
479 if (host_error == -EOPNOTSUPP)
480 return nfserr_attrnotsupp;
482 return nfserrno(host_error);
485 static struct posix_acl *
486 _get_posix_acl(struct dentry *dentry, char *key)
489 struct posix_acl *pacl = NULL;
492 buflen = nfsd_getxattr(dentry, key, &buf);
496 return ERR_PTR(buflen);
498 pacl = posix_acl_from_xattr(buf, buflen);
504 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
506 struct inode *inode = dentry->d_inode;
508 struct posix_acl *pacl = NULL, *dpacl = NULL;
509 unsigned int flags = 0;
511 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
512 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
513 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
515 error = PTR_ERR(pacl);
520 if (S_ISDIR(inode->i_mode)) {
521 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
522 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
524 else if (IS_ERR(dpacl)) {
525 error = PTR_ERR(dpacl);
529 flags = NFS4_ACL_DIR;
532 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
534 error = PTR_ERR(*acl);
538 posix_acl_release(pacl);
539 posix_acl_release(dpacl);
543 #endif /* defined(CONFIG_NFS_V4) */
545 #ifdef CONFIG_NFSD_V3
547 * Check server access rights to a file system object
553 static struct accessmap nfs3_regaccess[] = {
554 { NFS3_ACCESS_READ, MAY_READ },
555 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
556 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
557 { NFS3_ACCESS_EXTEND, MAY_WRITE },
562 static struct accessmap nfs3_diraccess[] = {
563 { NFS3_ACCESS_READ, MAY_READ },
564 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
565 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
566 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
567 { NFS3_ACCESS_DELETE, MAY_REMOVE },
572 static struct accessmap nfs3_anyaccess[] = {
573 /* Some clients - Solaris 2.6 at least, make an access call
574 * to the server to check for access for things like /dev/null
575 * (which really, the server doesn't care about). So
576 * We provide simple access checking for them, looking
577 * mainly at mode bits, and we make sure to ignore read-only
580 { NFS3_ACCESS_READ, MAY_READ },
581 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
582 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
583 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
589 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
591 struct accessmap *map;
592 struct svc_export *export;
593 struct dentry *dentry;
594 u32 query, result = 0, sresult = 0;
597 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
601 export = fhp->fh_export;
602 dentry = fhp->fh_dentry;
604 if (S_ISREG(dentry->d_inode->i_mode))
605 map = nfs3_regaccess;
606 else if (S_ISDIR(dentry->d_inode->i_mode))
607 map = nfs3_diraccess;
609 map = nfs3_anyaccess;
613 for (; map->access; map++) {
614 if (map->access & query) {
617 sresult |= map->access;
619 err2 = nfsd_permission(rqstp, export, dentry, map->how);
622 result |= map->access;
625 /* the following error codes just mean the access was not allowed,
626 * rather than an error occurred */
630 /* simply don't "or" in the access bit. */
640 *supported = sresult;
645 #endif /* CONFIG_NFSD_V3 */
650 * Open an existing file or directory.
651 * The access argument indicates the type of open (read/write/lock)
652 * N.B. After this call fhp needs an fh_put
655 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
656 int access, struct file **filp)
658 struct dentry *dentry;
660 int flags = O_RDONLY|O_LARGEFILE;
665 * If we get here, then the client has already done an "open",
666 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
667 * in case a chmod has now revoked permission.
669 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
673 dentry = fhp->fh_dentry;
674 inode = dentry->d_inode;
676 /* Disallow write access to files with the append-only bit set
677 * or any access when mandatory locking enabled
680 if (IS_APPEND(inode) && (access & MAY_WRITE))
682 if (IS_ISMNDLK(inode))
689 * Check to see if there are any leases on this file.
690 * This may block while leases are broken.
692 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
693 if (host_err == -EWOULDBLOCK)
694 host_err = -ETIMEDOUT;
695 if (host_err) /* NOMEM or WOULDBLOCK */
698 if (access & MAY_WRITE) {
699 if (access & MAY_READ)
700 flags = O_RDWR|O_LARGEFILE;
702 flags = O_WRONLY|O_LARGEFILE;
706 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
708 host_err = PTR_ERR(*filp);
710 err = nfserrno(host_err);
719 nfsd_close(struct file *filp)
726 * As this calls fsync (not fdatasync) there is no need for a write_inode
729 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
730 const struct file_operations *fop)
732 struct inode *inode = dp->d_inode;
733 int (*fsync) (struct file *, struct dentry *, int);
736 err = filemap_fdatawrite(inode->i_mapping);
737 if (err == 0 && fop && (fsync = fop->fsync))
738 err = fsync(filp, dp, 0);
740 err = filemap_fdatawait(inode->i_mapping);
747 nfsd_sync(struct file *filp)
750 struct inode *inode = filp->f_path.dentry->d_inode;
751 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
752 mutex_lock(&inode->i_mutex);
753 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
754 mutex_unlock(&inode->i_mutex);
760 nfsd_sync_dir(struct dentry *dp)
762 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
766 * Obtain the readahead parameters for the file
767 * specified by (dev, ino).
770 static inline struct raparms *
771 nfsd_get_raparms(dev_t dev, ino_t ino)
773 struct raparms *ra, **rap, **frap = NULL;
776 struct raparm_hbucket *rab;
778 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
779 rab = &raparm_hash[hash];
781 spin_lock(&rab->pb_lock);
782 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
783 if (ra->p_ino == ino && ra->p_dev == dev)
786 if (ra->p_count == 0)
789 depth = nfsdstats.ra_size*11/10;
791 spin_unlock(&rab->pb_lock);
801 if (rap != &rab->pb_head) {
803 ra->p_next = rab->pb_head;
807 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
808 spin_unlock(&rab->pb_lock);
813 * Grab and keep cached pages associated with a file in the svc_rqst
814 * so that they can be passed to the network sendmsg/sendpage routines
815 * directly. They will be released after the sending has completed.
818 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
819 struct splice_desc *sd)
821 struct svc_rqst *rqstp = sd->u.data;
822 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
823 struct page *page = buf->page;
827 ret = buf->ops->confirm(pipe, buf);
833 if (rqstp->rq_res.page_len == 0) {
838 rqstp->rq_res.page_base = buf->offset;
839 rqstp->rq_res.page_len = size;
840 } else if (page != pp[-1]) {
846 rqstp->rq_res.page_len += size;
848 rqstp->rq_res.page_len += size;
853 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
854 struct splice_desc *sd)
856 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
860 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
861 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
870 inode = file->f_path.dentry->d_inode;
872 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
873 (!lock_may_read(inode, offset, *count)))
877 /* Get readahead parameters */
878 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
881 file->f_ra = ra->p_ra;
883 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
884 struct splice_desc sd = {
891 rqstp->rq_resused = 1;
892 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
896 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
900 /* Write back readahead params */
902 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
903 spin_lock(&rab->pb_lock);
904 ra->p_ra = file->f_ra;
907 spin_unlock(&rab->pb_lock);
911 nfsdstats.io_read += host_err;
914 fsnotify_access(file->f_path.dentry);
916 err = nfserrno(host_err);
921 static void kill_suid(struct dentry *dentry)
924 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
926 mutex_lock(&dentry->d_inode->i_mutex);
927 notify_change(dentry, &ia);
928 mutex_unlock(&dentry->d_inode->i_mutex);
932 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
933 loff_t offset, struct kvec *vec, int vlen,
934 unsigned long cnt, int *stablep)
936 struct svc_export *exp;
937 struct dentry *dentry;
942 int stable = *stablep;
947 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
948 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
952 dentry = file->f_path.dentry;
953 inode = dentry->d_inode;
954 exp = fhp->fh_export;
957 * Request sync writes if
958 * - the sync export option has been set, or
959 * - the client requested O_SYNC behavior (NFSv3 feature).
960 * - The file system doesn't support fsync().
961 * When gathered writes have been configured for this volume,
962 * flushing the data to disk is handled separately below.
965 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
967 *stablep = 2; /* FILE_SYNC */
972 if (stable && !EX_WGATHER(exp))
973 file->f_flags |= O_SYNC;
975 /* Write the data. */
976 oldfs = get_fs(); set_fs(KERNEL_DS);
977 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
980 nfsdstats.io_write += cnt;
981 fsnotify_modify(file->f_path.dentry);
984 /* clear setuid/setgid flag after write */
985 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
988 if (host_err >= 0 && stable) {
989 static ino_t last_ino;
990 static dev_t last_dev;
993 * Gathered writes: If another process is currently
994 * writing to the file, there's a high chance
995 * this is another nfsd (triggered by a bulk write
996 * from a client's biod). Rather than syncing the
997 * file with each write request, we sleep for 10 msec.
999 * I don't know if this roughly approximates
1000 * C. Juszak's idea of gathered writes, but it's a
1001 * nice and simple solution (IMHO), and it seems to
1004 if (EX_WGATHER(exp)) {
1005 if (atomic_read(&inode->i_writecount) > 1
1006 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1007 dprintk("nfsd: write defer %d\n", current->pid);
1009 dprintk("nfsd: write resume %d\n", current->pid);
1012 if (inode->i_state & I_DIRTY) {
1013 dprintk("nfsd: write sync %d\n", current->pid);
1014 host_err=nfsd_sync(file);
1017 wake_up(&inode->i_wait);
1020 last_ino = inode->i_ino;
1021 last_dev = inode->i_sb->s_dev;
1024 dprintk("nfsd: write complete host_err=%d\n", host_err);
1028 err = nfserrno(host_err);
1034 * Read data from a file. count must contain the requested read count
1035 * on entry. On return, *count contains the number of bytes actually read.
1036 * N.B. After this call fhp needs an fh_put
1039 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1040 loff_t offset, struct kvec *vec, int vlen,
1041 unsigned long *count)
1046 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1047 MAY_READ|MAY_OWNER_OVERRIDE);
1050 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1052 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1055 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1063 * Write data to a file.
1064 * The stable flag requests synchronous writes.
1065 * N.B. After this call fhp needs an fh_put
1068 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1069 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1075 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1076 MAY_WRITE|MAY_OWNER_OVERRIDE);
1079 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1082 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1087 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1095 #ifdef CONFIG_NFSD_V3
1097 * Commit all pending writes to stable storage.
1098 * Strictly speaking, we could sync just the indicated file region here,
1099 * but there's currently no way we can ask the VFS to do so.
1101 * Unfortunately we cannot lock the file to make sure we return full WCC
1102 * data to the client, as locking happens lower down in the filesystem.
1105 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1106 loff_t offset, unsigned long count)
1111 if ((u64)count > ~(u64)offset)
1112 return nfserr_inval;
1114 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1116 if (EX_ISSYNC(fhp->fh_export)) {
1117 if (file->f_op && file->f_op->fsync) {
1118 err = nfserrno(nfsd_sync(file));
1120 err = nfserr_notsupp;
1127 #endif /* CONFIG_NFSD_V3 */
1130 * Create a file (regular, directory, device, fifo); UNIX sockets
1131 * not yet implemented.
1132 * If the response fh has been verified, the parent directory should
1133 * already be locked. Note that the parent directory is left locked.
1135 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1138 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1139 char *fname, int flen, struct iattr *iap,
1140 int type, dev_t rdev, struct svc_fh *resfhp)
1142 struct dentry *dentry, *dchild = NULL;
1151 if (isdotent(fname, flen))
1154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1158 dentry = fhp->fh_dentry;
1159 dirp = dentry->d_inode;
1161 err = nfserr_notdir;
1162 if(!dirp->i_op || !dirp->i_op->lookup)
1165 * Check whether the response file handle has been verified yet.
1166 * If it has, the parent directory should already be locked.
1168 if (!resfhp->fh_dentry) {
1169 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1170 fh_lock_nested(fhp, I_MUTEX_PARENT);
1171 dchild = lookup_one_len(fname, dentry, flen);
1172 host_err = PTR_ERR(dchild);
1175 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1179 /* called from nfsd_proc_create */
1180 dchild = dget(resfhp->fh_dentry);
1181 if (!fhp->fh_locked) {
1182 /* not actually possible */
1184 "nfsd_create: parent %s/%s not locked!\n",
1185 dentry->d_parent->d_name.name,
1186 dentry->d_name.name);
1192 * Make sure the child dentry is still negative ...
1195 if (dchild->d_inode) {
1196 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1197 dentry->d_name.name, dchild->d_name.name);
1201 if (!(iap->ia_valid & ATTR_MODE))
1203 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1206 * Get the dir op function pointer.
1211 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1214 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1220 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1223 printk("nfsd: bad file type %o in nfsd_create\n", type);
1229 if (EX_ISSYNC(fhp->fh_export)) {
1230 err = nfserrno(nfsd_sync_dir(dentry));
1231 write_inode_now(dchild->d_inode, 1);
1235 /* Set file attributes. Mode has already been set and
1236 * setting uid/gid works only for root. Irix appears to
1237 * send along the gid when it tries to implement setgid
1238 * directories via NFS.
1240 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1241 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1246 * Update the file handle to get the new inode info.
1249 err = fh_update(resfhp);
1251 if (dchild && !IS_ERR(dchild))
1256 err = nfserrno(host_err);
1260 #ifdef CONFIG_NFSD_V3
1262 * NFSv3 version of nfsd_create
1265 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1266 char *fname, int flen, struct iattr *iap,
1267 struct svc_fh *resfhp, int createmode, u32 *verifier,
1268 int *truncp, int *created)
1270 struct dentry *dentry, *dchild = NULL;
1274 __u32 v_mtime=0, v_atime=0;
1280 if (isdotent(fname, flen))
1282 if (!(iap->ia_valid & ATTR_MODE))
1284 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1288 dentry = fhp->fh_dentry;
1289 dirp = dentry->d_inode;
1291 /* Get all the sanity checks out of the way before
1292 * we lock the parent. */
1293 err = nfserr_notdir;
1294 if(!dirp->i_op || !dirp->i_op->lookup)
1296 fh_lock_nested(fhp, I_MUTEX_PARENT);
1299 * Compose the response file handle.
1301 dchild = lookup_one_len(fname, dentry, flen);
1302 host_err = PTR_ERR(dchild);
1306 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1310 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1311 /* solaris7 gets confused (bugid 4218508) if these have
1312 * the high bit set, so just clear the high bits.
1314 v_mtime = verifier[0]&0x7fffffff;
1315 v_atime = verifier[1]&0x7fffffff;
1318 if (dchild->d_inode) {
1321 switch (createmode) {
1322 case NFS3_CREATE_UNCHECKED:
1323 if (! S_ISREG(dchild->d_inode->i_mode))
1326 /* in nfsv4, we need to treat this case a little
1327 * differently. we don't want to truncate the
1328 * file now; this would be wrong if the OPEN
1329 * fails for some other reason. furthermore,
1330 * if the size is nonzero, we should ignore it
1331 * according to spec!
1333 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1336 iap->ia_valid &= ATTR_SIZE;
1340 case NFS3_CREATE_EXCLUSIVE:
1341 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1342 && dchild->d_inode->i_atime.tv_sec == v_atime
1343 && dchild->d_inode->i_size == 0 )
1346 case NFS3_CREATE_GUARDED:
1352 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1358 if (EX_ISSYNC(fhp->fh_export)) {
1359 err = nfserrno(nfsd_sync_dir(dentry));
1360 /* setattr will sync the child (or not) */
1363 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1364 /* Cram the verifier into atime/mtime */
1365 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1366 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1367 /* XXX someone who knows this better please fix it for nsec */
1368 iap->ia_mtime.tv_sec = v_mtime;
1369 iap->ia_atime.tv_sec = v_atime;
1370 iap->ia_mtime.tv_nsec = 0;
1371 iap->ia_atime.tv_nsec = 0;
1374 /* Set file attributes.
1375 * Irix appears to send along the gid when it tries to
1376 * implement setgid directories via NFS. Clear out all that cruft.
1379 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1380 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1386 * Update the filehandle to get the new inode info.
1389 err = fh_update(resfhp);
1393 if (dchild && !IS_ERR(dchild))
1398 err = nfserrno(host_err);
1401 #endif /* CONFIG_NFSD_V3 */
1404 * Read a symlink. On entry, *lenp must contain the maximum path length that
1405 * fits into the buffer. On return, it contains the true length.
1406 * N.B. After this call fhp needs an fh_put
1409 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1411 struct dentry *dentry;
1412 struct inode *inode;
1417 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1421 dentry = fhp->fh_dentry;
1422 inode = dentry->d_inode;
1425 if (!inode->i_op || !inode->i_op->readlink)
1428 touch_atime(fhp->fh_export->ex_mnt, dentry);
1429 /* N.B. Why does this call need a get_fs()??
1430 * Remove the set_fs and watch the fireworks:-) --okir
1433 oldfs = get_fs(); set_fs(KERNEL_DS);
1434 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1445 err = nfserrno(host_err);
1450 * Create a symlink and look up its inode
1451 * N.B. After this call _both_ fhp and resfhp need an fh_put
1454 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1455 char *fname, int flen,
1456 char *path, int plen,
1457 struct svc_fh *resfhp,
1460 struct dentry *dentry, *dnew;
1469 if (isdotent(fname, flen))
1472 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1476 dentry = fhp->fh_dentry;
1477 dnew = lookup_one_len(fname, dentry, flen);
1478 host_err = PTR_ERR(dnew);
1483 /* Only the MODE ATTRibute is even vaguely meaningful */
1484 if (iap && (iap->ia_valid & ATTR_MODE))
1485 mode = iap->ia_mode & S_IALLUGO;
1487 if (unlikely(path[plen] != 0)) {
1488 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1489 if (path_alloced == NULL)
1492 strncpy(path_alloced, path, plen);
1493 path_alloced[plen] = 0;
1494 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1495 kfree(path_alloced);
1498 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1501 if (EX_ISSYNC(fhp->fh_export))
1502 host_err = nfsd_sync_dir(dentry);
1504 err = nfserrno(host_err);
1507 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1509 if (err==0) err = cerr;
1514 err = nfserrno(host_err);
1520 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1523 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1524 char *name, int len, struct svc_fh *tfhp)
1526 struct dentry *ddir, *dnew, *dold;
1527 struct inode *dirp, *dest;
1531 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1534 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1542 if (isdotent(name, len))
1545 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1546 ddir = ffhp->fh_dentry;
1547 dirp = ddir->d_inode;
1549 dnew = lookup_one_len(name, ddir, len);
1550 host_err = PTR_ERR(dnew);
1554 dold = tfhp->fh_dentry;
1555 dest = dold->d_inode;
1557 host_err = vfs_link(dold, dirp, dnew);
1559 if (EX_ISSYNC(ffhp->fh_export)) {
1560 err = nfserrno(nfsd_sync_dir(ddir));
1561 write_inode_now(dest, 1);
1565 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1568 err = nfserrno(host_err);
1578 err = nfserrno(host_err);
1584 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1587 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1588 struct svc_fh *tfhp, char *tname, int tlen)
1590 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1591 struct inode *fdir, *tdir;
1595 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1598 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1602 fdentry = ffhp->fh_dentry;
1603 fdir = fdentry->d_inode;
1605 tdentry = tfhp->fh_dentry;
1606 tdir = tdentry->d_inode;
1608 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1609 if (ffhp->fh_export != tfhp->fh_export)
1613 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1616 /* cannot use fh_lock as we need deadlock protective ordering
1617 * so do it by hand */
1618 trap = lock_rename(tdentry, fdentry);
1619 ffhp->fh_locked = tfhp->fh_locked = 1;
1623 odentry = lookup_one_len(fname, fdentry, flen);
1624 host_err = PTR_ERR(odentry);
1625 if (IS_ERR(odentry))
1629 if (!odentry->d_inode)
1632 if (odentry == trap)
1635 ndentry = lookup_one_len(tname, tdentry, tlen);
1636 host_err = PTR_ERR(ndentry);
1637 if (IS_ERR(ndentry))
1639 host_err = -ENOTEMPTY;
1640 if (ndentry == trap)
1644 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1645 ((atomic_read(&odentry->d_count) > 1)
1646 || (atomic_read(&ndentry->d_count) > 1))) {
1650 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1651 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1652 host_err = nfsd_sync_dir(tdentry);
1654 host_err = nfsd_sync_dir(fdentry);
1662 err = nfserrno(host_err);
1664 /* we cannot reply on fh_unlock on the two filehandles,
1665 * as that would do the wrong thing if the two directories
1666 * were the same, so again we do it by hand
1668 fill_post_wcc(ffhp);
1669 fill_post_wcc(tfhp);
1670 unlock_rename(tdentry, fdentry);
1671 ffhp->fh_locked = tfhp->fh_locked = 0;
1678 * Unlink a file or directory
1679 * N.B. After this call fhp needs an fh_put
1682 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1683 char *fname, int flen)
1685 struct dentry *dentry, *rdentry;
1691 if (!flen || isdotent(fname, flen))
1693 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1697 fh_lock_nested(fhp, I_MUTEX_PARENT);
1698 dentry = fhp->fh_dentry;
1699 dirp = dentry->d_inode;
1701 rdentry = lookup_one_len(fname, dentry, flen);
1702 host_err = PTR_ERR(rdentry);
1703 if (IS_ERR(rdentry))
1706 if (!rdentry->d_inode) {
1713 type = rdentry->d_inode->i_mode & S_IFMT;
1715 if (type != S_IFDIR) { /* It's UNLINK */
1717 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1718 (atomic_read(&rdentry->d_count) > 1)) {
1722 host_err = vfs_unlink(dirp, rdentry);
1723 } else { /* It's RMDIR */
1724 host_err = vfs_rmdir(dirp, rdentry);
1731 if (EX_ISSYNC(fhp->fh_export))
1732 host_err = nfsd_sync_dir(dentry);
1735 err = nfserrno(host_err);
1741 * Read entries from a directory.
1742 * The NFSv3/4 verifier we ignore for now.
1745 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1746 struct readdir_cd *cdp, filldir_t func)
1751 loff_t offset = *offsetp;
1753 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1757 offset = vfs_llseek(file, offset, 0);
1759 err = nfserrno((int)offset);
1764 * Read the directory entries. This silly loop is necessary because
1765 * readdir() is not guaranteed to fill up the entire buffer, but
1766 * may choose to do less.
1770 cdp->err = nfserr_eof; /* will be cleared on successful read */
1771 host_err = vfs_readdir(file, func, cdp);
1772 } while (host_err >=0 && cdp->err == nfs_ok);
1774 err = nfserrno(host_err);
1777 *offsetp = vfs_llseek(file, 0, 1);
1779 if (err == nfserr_eof || err == nfserr_toosmall)
1780 err = nfs_ok; /* can still be found in ->err */
1788 * Get file system stats
1789 * N.B. After this call fhp needs an fh_put
1792 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1794 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1795 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1800 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1802 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1806 * Check for a user's access permissions to this inode.
1809 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1810 struct dentry *dentry, int acc)
1812 struct inode *inode = dentry->d_inode;
1818 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1820 (acc & MAY_READ)? " read" : "",
1821 (acc & MAY_WRITE)? " write" : "",
1822 (acc & MAY_EXEC)? " exec" : "",
1823 (acc & MAY_SATTR)? " sattr" : "",
1824 (acc & MAY_TRUNC)? " trunc" : "",
1825 (acc & MAY_LOCK)? " lock" : "",
1826 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1828 IS_IMMUTABLE(inode)? " immut" : "",
1829 IS_APPEND(inode)? " append" : "",
1830 IS_RDONLY(inode)? " ro" : "");
1831 dprintk(" owner %d/%d user %d/%d\n",
1832 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1835 /* Normally we reject any write/sattr etc access on a read-only file
1836 * system. But if it is IRIX doing check on write-access for a
1837 * device special file, we ignore rofs.
1839 if (!(acc & MAY_LOCAL_ACCESS))
1840 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1841 if (exp_rdonly(rqstp, exp) || IS_RDONLY(inode))
1843 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1846 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1849 if (acc & MAY_LOCK) {
1850 /* If we cannot rely on authentication in NLM requests,
1851 * just allow locks, otherwise require read permission, or
1854 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1857 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1860 * The file owner always gets access permission for accesses that
1861 * would normally be checked at open time. This is to make
1862 * file access work even when the client has done a fchmod(fd, 0).
1864 * However, `cp foo bar' should fail nevertheless when bar is
1865 * readonly. A sensible way to do this might be to reject all
1866 * attempts to truncate a read-only file, because a creat() call
1867 * always implies file truncation.
1868 * ... but this isn't really fair. A process may reasonably call
1869 * ftruncate on an open file descriptor on a file with perm 000.
1870 * We must trust the client to do permission checking - using "ACCESS"
1873 if ((acc & MAY_OWNER_OVERRIDE) &&
1874 inode->i_uid == current->fsuid)
1877 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1879 /* Allow read access to binaries even when mode 111 */
1880 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1881 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1882 err = permission(inode, MAY_EXEC, NULL);
1884 return err? nfserrno(err) : 0;
1888 nfsd_racache_shutdown(void)
1892 dprintk("nfsd: freeing readahead buffers.\n");
1897 * Initialize readahead param cache
1900 nfsd_racache_init(int cache_size)
1909 if (cache_size < 2*RAPARM_HASH_SIZE)
1910 cache_size = 2*RAPARM_HASH_SIZE;
1911 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1915 "nfsd: Could not allocate memory read-ahead cache.\n");
1919 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1920 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1921 raparm_hash[i].pb_head = NULL;
1922 spin_lock_init(&raparm_hash[i].pb_lock);
1924 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
1925 for (i = 0; i < cache_size - 1; i++) {
1926 if (i % nperbucket == 0)
1927 raparm_hash[j++].pb_head = raparml + i;
1928 if (i % nperbucket < nperbucket-1)
1929 raparml[i].p_next = raparml + i + 1;
1932 nfsdstats.ra_size = cache_size;
1936 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1938 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1940 struct inode *inode = fhp->fh_dentry->d_inode;
1944 struct posix_acl *acl;
1946 if (!IS_POSIXACL(inode))
1947 return ERR_PTR(-EOPNOTSUPP);
1950 case ACL_TYPE_ACCESS:
1951 name = POSIX_ACL_XATTR_ACCESS;
1953 case ACL_TYPE_DEFAULT:
1954 name = POSIX_ACL_XATTR_DEFAULT;
1957 return ERR_PTR(-EOPNOTSUPP);
1960 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1962 return ERR_PTR(size);
1964 acl = posix_acl_from_xattr(value, size);
1970 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1972 struct inode *inode = fhp->fh_dentry->d_inode;
1978 if (!IS_POSIXACL(inode) || !inode->i_op ||
1979 !inode->i_op->setxattr || !inode->i_op->removexattr)
1982 case ACL_TYPE_ACCESS:
1983 name = POSIX_ACL_XATTR_ACCESS;
1985 case ACL_TYPE_DEFAULT:
1986 name = POSIX_ACL_XATTR_DEFAULT;
1992 if (acl && acl->a_count) {
1993 size = posix_acl_xattr_size(acl->a_count);
1994 value = kmalloc(size, GFP_KERNEL);
1997 error = posix_acl_to_xattr(acl, value, size);
2005 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2007 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2010 error = vfs_removexattr(fhp->fh_dentry, name);
2011 if (error == -ENODATA)
2020 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */