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);
118 if (PTR_ERR(exp2) != -ENOENT)
124 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
125 /* successfully crossed mount point */
140 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
141 const char *name, int len,
142 struct svc_export **exp_ret, struct dentry **dentry_ret)
144 struct svc_export *exp;
145 struct dentry *dparent;
146 struct dentry *dentry;
150 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152 /* Obtain dentry and export. */
153 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
157 dparent = fhp->fh_dentry;
158 exp = fhp->fh_export;
161 /* Lookup the name, but don't follow links */
162 if (isdotent(name, len)) {
164 dentry = dget(dparent);
165 else if (dparent != exp->ex_dentry) {
166 dentry = dget_parent(dparent);
167 } else if (!EX_NOHIDE(exp))
168 dentry = dget(dparent); /* .. == . just like at / */
170 /* checking mountpoint crossing is very different when stepping up */
171 struct svc_export *exp2 = NULL;
173 struct vfsmount *mnt = mntget(exp->ex_mnt);
174 dentry = dget(dparent);
175 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
177 dp = dget_parent(dentry);
181 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
182 if (PTR_ERR(exp2) == -ENOENT) {
184 dentry = dget(dparent);
185 } else if (IS_ERR(exp2)) {
186 host_err = PTR_ERR(exp2);
198 dentry = lookup_one_len(name, dparent, len);
199 host_err = PTR_ERR(dentry);
203 * check if we have crossed a mount point ...
205 if (d_mountpoint(dentry)) {
206 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
212 *dentry_ret = dentry;
218 return nfserrno(host_err);
222 * Look up one component of a pathname.
223 * N.B. After this call _both_ fhp and resfh need an fh_put
225 * If the lookup would cross a mountpoint, and the mounted filesystem
226 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
227 * accepted as it stands and the mounted directory is
228 * returned. Otherwise the covered directory is returned.
229 * NOTE: this mountpoint crossing is not supported properly by all
230 * clients and is explicitly disallowed for NFSv3
231 * NeilBrown <neilb@cse.unsw.edu.au>
234 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
235 int len, struct svc_fh *resfh)
237 struct svc_export *exp;
238 struct dentry *dentry;
241 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
244 err = check_nfsd_access(exp, rqstp);
248 * Note: we compose the file handle now, but as the
249 * dentry may be negative, it may need to be updated.
251 err = fh_compose(resfh, exp, dentry, fhp);
252 if (!err && !dentry->d_inode)
262 * Set various file attributes.
263 * N.B. After this call fhp needs an fh_put
266 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
267 int check_guard, time_t guardtime)
269 struct dentry *dentry;
271 int accmode = MAY_SATTR;
278 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
279 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
280 if (iap->ia_valid & ATTR_SIZE)
284 err = fh_verify(rqstp, fhp, ftype, accmode);
288 dentry = fhp->fh_dentry;
289 inode = dentry->d_inode;
291 /* Ignore any mode updates on symlinks */
292 if (S_ISLNK(inode->i_mode))
293 iap->ia_valid &= ~ATTR_MODE;
299 * NFSv2 does not differentiate between "set-[ac]time-to-now"
300 * which only requires access, and "set-[ac]time-to-X" which
301 * requires ownership.
302 * So if it looks like it might be "set both to the same time which
303 * is close to now", and if inode_change_ok fails, then we
304 * convert to "set to now" instead of "set to explicit time"
306 * We only call inode_change_ok as the last test as technically
307 * it is not an interface that we should be using. It is only
308 * valid if the filesystem does not define it's own i_op->setattr.
310 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
311 #define MAX_TOUCH_TIME_ERROR (30*60)
312 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
313 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
317 * Now just make sure time is in the right ballpark.
318 * Solaris, at least, doesn't seem to care what the time
319 * request is. We require it be within 30 minutes of now.
321 time_t delta = iap->ia_atime.tv_sec - get_seconds();
324 if (delta < MAX_TOUCH_TIME_ERROR &&
325 inode_change_ok(inode, iap) != 0) {
327 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
328 * This will cause notify_change to set these times
331 iap->ia_valid &= ~BOTH_TIME_SET;
336 * The size case is special.
337 * It changes the file as well as the attributes.
339 if (iap->ia_valid & ATTR_SIZE) {
340 if (iap->ia_size < inode->i_size) {
341 err = nfsd_permission(rqstp, fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
347 * If we are changing the size of the file, then
348 * we need to break all leases.
350 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
351 if (host_err == -EWOULDBLOCK)
352 host_err = -ETIMEDOUT;
353 if (host_err) /* ENOMEM or EWOULDBLOCK */
356 host_err = get_write_access(inode);
361 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
363 put_write_access(inode);
369 imode = inode->i_mode;
370 if (iap->ia_valid & ATTR_MODE) {
371 iap->ia_mode &= S_IALLUGO;
372 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
375 /* Revoke setuid/setgid bit on chown/chgrp */
376 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
377 iap->ia_valid |= ATTR_KILL_SUID;
378 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
379 iap->ia_valid |= ATTR_KILL_SGID;
381 /* Change the attributes. */
383 iap->ia_valid |= ATTR_CTIME;
385 err = nfserr_notsync;
386 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
388 host_err = notify_change(dentry, iap);
389 err = nfserrno(host_err);
393 put_write_access(inode);
395 if (EX_ISSYNC(fhp->fh_export))
396 write_inode_now(inode, 1);
401 err = nfserrno(host_err);
405 #if defined(CONFIG_NFSD_V2_ACL) || \
406 defined(CONFIG_NFSD_V3_ACL) || \
407 defined(CONFIG_NFSD_V4)
408 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
412 buflen = vfs_getxattr(dentry, key, NULL, 0);
416 *buf = kmalloc(buflen, GFP_KERNEL);
420 return vfs_getxattr(dentry, key, *buf, buflen);
424 #if defined(CONFIG_NFSD_V4)
426 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
433 buflen = posix_acl_xattr_size(pacl->a_count);
434 buf = kmalloc(buflen, GFP_KERNEL);
439 len = posix_acl_to_xattr(pacl, buf, buflen);
445 error = vfs_setxattr(dentry, key, buf, len, 0);
452 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
453 struct nfs4_acl *acl)
457 struct dentry *dentry;
459 struct posix_acl *pacl = NULL, *dpacl = NULL;
460 unsigned int flags = 0;
463 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
467 dentry = fhp->fh_dentry;
468 inode = dentry->d_inode;
469 if (S_ISDIR(inode->i_mode))
470 flags = NFS4_ACL_DIR;
472 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
473 if (host_error == -EINVAL) {
474 return nfserr_attrnotsupp;
475 } else if (host_error < 0)
478 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
482 if (S_ISDIR(inode->i_mode))
483 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
486 posix_acl_release(pacl);
487 posix_acl_release(dpacl);
489 if (host_error == -EOPNOTSUPP)
490 return nfserr_attrnotsupp;
492 return nfserrno(host_error);
495 static struct posix_acl *
496 _get_posix_acl(struct dentry *dentry, char *key)
499 struct posix_acl *pacl = NULL;
502 buflen = nfsd_getxattr(dentry, key, &buf);
506 return ERR_PTR(buflen);
508 pacl = posix_acl_from_xattr(buf, buflen);
514 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
516 struct inode *inode = dentry->d_inode;
518 struct posix_acl *pacl = NULL, *dpacl = NULL;
519 unsigned int flags = 0;
521 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
522 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
523 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
525 error = PTR_ERR(pacl);
530 if (S_ISDIR(inode->i_mode)) {
531 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
532 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
534 else if (IS_ERR(dpacl)) {
535 error = PTR_ERR(dpacl);
539 flags = NFS4_ACL_DIR;
542 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
544 error = PTR_ERR(*acl);
548 posix_acl_release(pacl);
549 posix_acl_release(dpacl);
553 #endif /* defined(CONFIG_NFS_V4) */
555 #ifdef CONFIG_NFSD_V3
557 * Check server access rights to a file system object
563 static struct accessmap nfs3_regaccess[] = {
564 { NFS3_ACCESS_READ, MAY_READ },
565 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
566 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
567 { NFS3_ACCESS_EXTEND, MAY_WRITE },
572 static struct accessmap nfs3_diraccess[] = {
573 { NFS3_ACCESS_READ, MAY_READ },
574 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
575 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
576 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
577 { NFS3_ACCESS_DELETE, MAY_REMOVE },
582 static struct accessmap nfs3_anyaccess[] = {
583 /* Some clients - Solaris 2.6 at least, make an access call
584 * to the server to check for access for things like /dev/null
585 * (which really, the server doesn't care about). So
586 * We provide simple access checking for them, looking
587 * mainly at mode bits, and we make sure to ignore read-only
590 { NFS3_ACCESS_READ, MAY_READ },
591 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
592 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
593 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
599 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
601 struct accessmap *map;
602 struct svc_export *export;
603 struct dentry *dentry;
604 u32 query, result = 0, sresult = 0;
607 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
611 export = fhp->fh_export;
612 dentry = fhp->fh_dentry;
614 if (S_ISREG(dentry->d_inode->i_mode))
615 map = nfs3_regaccess;
616 else if (S_ISDIR(dentry->d_inode->i_mode))
617 map = nfs3_diraccess;
619 map = nfs3_anyaccess;
623 for (; map->access; map++) {
624 if (map->access & query) {
627 sresult |= map->access;
629 err2 = nfsd_permission(rqstp, export, dentry, map->how);
632 result |= map->access;
635 /* the following error codes just mean the access was not allowed,
636 * rather than an error occurred */
640 /* simply don't "or" in the access bit. */
650 *supported = sresult;
655 #endif /* CONFIG_NFSD_V3 */
660 * Open an existing file or directory.
661 * The access argument indicates the type of open (read/write/lock)
662 * N.B. After this call fhp needs an fh_put
665 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
666 int access, struct file **filp)
668 struct dentry *dentry;
670 int flags = O_RDONLY|O_LARGEFILE;
675 * If we get here, then the client has already done an "open",
676 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
677 * in case a chmod has now revoked permission.
679 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
683 dentry = fhp->fh_dentry;
684 inode = dentry->d_inode;
686 /* Disallow write access to files with the append-only bit set
687 * or any access when mandatory locking enabled
690 if (IS_APPEND(inode) && (access & MAY_WRITE))
692 if (IS_ISMNDLK(inode))
699 * Check to see if there are any leases on this file.
700 * This may block while leases are broken.
702 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
703 if (host_err == -EWOULDBLOCK)
704 host_err = -ETIMEDOUT;
705 if (host_err) /* NOMEM or WOULDBLOCK */
708 if (access & MAY_WRITE) {
709 if (access & MAY_READ)
710 flags = O_RDWR|O_LARGEFILE;
712 flags = O_WRONLY|O_LARGEFILE;
716 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
718 host_err = PTR_ERR(*filp);
720 err = nfserrno(host_err);
729 nfsd_close(struct file *filp)
736 * As this calls fsync (not fdatasync) there is no need for a write_inode
739 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
740 const struct file_operations *fop)
742 struct inode *inode = dp->d_inode;
743 int (*fsync) (struct file *, struct dentry *, int);
746 err = filemap_fdatawrite(inode->i_mapping);
747 if (err == 0 && fop && (fsync = fop->fsync))
748 err = fsync(filp, dp, 0);
750 err = filemap_fdatawait(inode->i_mapping);
757 nfsd_sync(struct file *filp)
760 struct inode *inode = filp->f_path.dentry->d_inode;
761 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
762 mutex_lock(&inode->i_mutex);
763 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
764 mutex_unlock(&inode->i_mutex);
770 nfsd_sync_dir(struct dentry *dp)
772 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
776 * Obtain the readahead parameters for the file
777 * specified by (dev, ino).
780 static inline struct raparms *
781 nfsd_get_raparms(dev_t dev, ino_t ino)
783 struct raparms *ra, **rap, **frap = NULL;
786 struct raparm_hbucket *rab;
788 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
789 rab = &raparm_hash[hash];
791 spin_lock(&rab->pb_lock);
792 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
793 if (ra->p_ino == ino && ra->p_dev == dev)
796 if (ra->p_count == 0)
799 depth = nfsdstats.ra_size*11/10;
801 spin_unlock(&rab->pb_lock);
811 if (rap != &rab->pb_head) {
813 ra->p_next = rab->pb_head;
817 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
818 spin_unlock(&rab->pb_lock);
823 * Grab and keep cached pages associated with a file in the svc_rqst
824 * so that they can be passed to the network sendmsg/sendpage routines
825 * directly. They will be released after the sending has completed.
828 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
829 struct splice_desc *sd)
831 struct svc_rqst *rqstp = sd->u.data;
832 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
833 struct page *page = buf->page;
837 ret = buf->ops->confirm(pipe, buf);
843 if (rqstp->rq_res.page_len == 0) {
848 rqstp->rq_res.page_base = buf->offset;
849 rqstp->rq_res.page_len = size;
850 } else if (page != pp[-1]) {
856 rqstp->rq_res.page_len += size;
858 rqstp->rq_res.page_len += size;
863 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
864 struct splice_desc *sd)
866 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
870 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
871 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
880 inode = file->f_path.dentry->d_inode;
882 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
883 (!lock_may_read(inode, offset, *count)))
887 /* Get readahead parameters */
888 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
891 file->f_ra = ra->p_ra;
893 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
894 struct splice_desc sd = {
901 rqstp->rq_resused = 1;
902 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
906 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
910 /* Write back readahead params */
912 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
913 spin_lock(&rab->pb_lock);
914 ra->p_ra = file->f_ra;
917 spin_unlock(&rab->pb_lock);
921 nfsdstats.io_read += host_err;
924 fsnotify_access(file->f_path.dentry);
926 err = nfserrno(host_err);
931 static void kill_suid(struct dentry *dentry)
934 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
936 mutex_lock(&dentry->d_inode->i_mutex);
937 notify_change(dentry, &ia);
938 mutex_unlock(&dentry->d_inode->i_mutex);
942 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
943 loff_t offset, struct kvec *vec, int vlen,
944 unsigned long cnt, int *stablep)
946 struct svc_export *exp;
947 struct dentry *dentry;
952 int stable = *stablep;
957 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
958 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
962 dentry = file->f_path.dentry;
963 inode = dentry->d_inode;
964 exp = fhp->fh_export;
967 * Request sync writes if
968 * - the sync export option has been set, or
969 * - the client requested O_SYNC behavior (NFSv3 feature).
970 * - The file system doesn't support fsync().
971 * When gathered writes have been configured for this volume,
972 * flushing the data to disk is handled separately below.
975 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
977 *stablep = 2; /* FILE_SYNC */
982 if (stable && !EX_WGATHER(exp))
983 file->f_flags |= O_SYNC;
985 /* Write the data. */
986 oldfs = get_fs(); set_fs(KERNEL_DS);
987 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
990 nfsdstats.io_write += cnt;
991 fsnotify_modify(file->f_path.dentry);
994 /* clear setuid/setgid flag after write */
995 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
998 if (host_err >= 0 && stable) {
999 static ino_t last_ino;
1000 static dev_t last_dev;
1003 * Gathered writes: If another process is currently
1004 * writing to the file, there's a high chance
1005 * this is another nfsd (triggered by a bulk write
1006 * from a client's biod). Rather than syncing the
1007 * file with each write request, we sleep for 10 msec.
1009 * I don't know if this roughly approximates
1010 * C. Juszak's idea of gathered writes, but it's a
1011 * nice and simple solution (IMHO), and it seems to
1014 if (EX_WGATHER(exp)) {
1015 if (atomic_read(&inode->i_writecount) > 1
1016 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1017 dprintk("nfsd: write defer %d\n", current->pid);
1019 dprintk("nfsd: write resume %d\n", current->pid);
1022 if (inode->i_state & I_DIRTY) {
1023 dprintk("nfsd: write sync %d\n", current->pid);
1024 host_err=nfsd_sync(file);
1027 wake_up(&inode->i_wait);
1030 last_ino = inode->i_ino;
1031 last_dev = inode->i_sb->s_dev;
1034 dprintk("nfsd: write complete host_err=%d\n", host_err);
1038 err = nfserrno(host_err);
1044 * Read data from a file. count must contain the requested read count
1045 * on entry. On return, *count contains the number of bytes actually read.
1046 * N.B. After this call fhp needs an fh_put
1049 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1050 loff_t offset, struct kvec *vec, int vlen,
1051 unsigned long *count)
1056 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1057 MAY_READ|MAY_OWNER_OVERRIDE);
1060 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1062 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1065 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1073 * Write data to a file.
1074 * The stable flag requests synchronous writes.
1075 * N.B. After this call fhp needs an fh_put
1078 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1079 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1085 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1086 MAY_WRITE|MAY_OWNER_OVERRIDE);
1089 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1092 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1097 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1105 #ifdef CONFIG_NFSD_V3
1107 * Commit all pending writes to stable storage.
1108 * Strictly speaking, we could sync just the indicated file region here,
1109 * but there's currently no way we can ask the VFS to do so.
1111 * Unfortunately we cannot lock the file to make sure we return full WCC
1112 * data to the client, as locking happens lower down in the filesystem.
1115 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1116 loff_t offset, unsigned long count)
1121 if ((u64)count > ~(u64)offset)
1122 return nfserr_inval;
1124 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1126 if (EX_ISSYNC(fhp->fh_export)) {
1127 if (file->f_op && file->f_op->fsync) {
1128 err = nfserrno(nfsd_sync(file));
1130 err = nfserr_notsupp;
1137 #endif /* CONFIG_NFSD_V3 */
1140 * Create a file (regular, directory, device, fifo); UNIX sockets
1141 * not yet implemented.
1142 * If the response fh has been verified, the parent directory should
1143 * already be locked. Note that the parent directory is left locked.
1145 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1148 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1149 char *fname, int flen, struct iattr *iap,
1150 int type, dev_t rdev, struct svc_fh *resfhp)
1152 struct dentry *dentry, *dchild = NULL;
1161 if (isdotent(fname, flen))
1164 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1168 dentry = fhp->fh_dentry;
1169 dirp = dentry->d_inode;
1171 err = nfserr_notdir;
1172 if(!dirp->i_op || !dirp->i_op->lookup)
1175 * Check whether the response file handle has been verified yet.
1176 * If it has, the parent directory should already be locked.
1178 if (!resfhp->fh_dentry) {
1179 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1180 fh_lock_nested(fhp, I_MUTEX_PARENT);
1181 dchild = lookup_one_len(fname, dentry, flen);
1182 host_err = PTR_ERR(dchild);
1185 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1189 /* called from nfsd_proc_create */
1190 dchild = dget(resfhp->fh_dentry);
1191 if (!fhp->fh_locked) {
1192 /* not actually possible */
1194 "nfsd_create: parent %s/%s not locked!\n",
1195 dentry->d_parent->d_name.name,
1196 dentry->d_name.name);
1202 * Make sure the child dentry is still negative ...
1205 if (dchild->d_inode) {
1206 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1207 dentry->d_name.name, dchild->d_name.name);
1211 if (!(iap->ia_valid & ATTR_MODE))
1213 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1216 * Get the dir op function pointer.
1221 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1224 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1230 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1233 printk("nfsd: bad file type %o in nfsd_create\n", type);
1239 if (EX_ISSYNC(fhp->fh_export)) {
1240 err = nfserrno(nfsd_sync_dir(dentry));
1241 write_inode_now(dchild->d_inode, 1);
1245 /* Set file attributes. Mode has already been set and
1246 * setting uid/gid works only for root. Irix appears to
1247 * send along the gid when it tries to implement setgid
1248 * directories via NFS.
1250 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1251 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1256 * Update the file handle to get the new inode info.
1259 err = fh_update(resfhp);
1261 if (dchild && !IS_ERR(dchild))
1266 err = nfserrno(host_err);
1270 #ifdef CONFIG_NFSD_V3
1272 * NFSv3 version of nfsd_create
1275 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1276 char *fname, int flen, struct iattr *iap,
1277 struct svc_fh *resfhp, int createmode, u32 *verifier,
1278 int *truncp, int *created)
1280 struct dentry *dentry, *dchild = NULL;
1284 __u32 v_mtime=0, v_atime=0;
1290 if (isdotent(fname, flen))
1292 if (!(iap->ia_valid & ATTR_MODE))
1294 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1298 dentry = fhp->fh_dentry;
1299 dirp = dentry->d_inode;
1301 /* Get all the sanity checks out of the way before
1302 * we lock the parent. */
1303 err = nfserr_notdir;
1304 if(!dirp->i_op || !dirp->i_op->lookup)
1306 fh_lock_nested(fhp, I_MUTEX_PARENT);
1309 * Compose the response file handle.
1311 dchild = lookup_one_len(fname, dentry, flen);
1312 host_err = PTR_ERR(dchild);
1316 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1320 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1321 /* solaris7 gets confused (bugid 4218508) if these have
1322 * the high bit set, so just clear the high bits. If this is
1323 * ever changed to use different attrs for storing the
1324 * verifier, then do_open_lookup() will also need to be fixed
1327 v_mtime = verifier[0]&0x7fffffff;
1328 v_atime = verifier[1]&0x7fffffff;
1331 if (dchild->d_inode) {
1334 switch (createmode) {
1335 case NFS3_CREATE_UNCHECKED:
1336 if (! S_ISREG(dchild->d_inode->i_mode))
1339 /* in nfsv4, we need to treat this case a little
1340 * differently. we don't want to truncate the
1341 * file now; this would be wrong if the OPEN
1342 * fails for some other reason. furthermore,
1343 * if the size is nonzero, we should ignore it
1344 * according to spec!
1346 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1349 iap->ia_valid &= ATTR_SIZE;
1353 case NFS3_CREATE_EXCLUSIVE:
1354 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1355 && dchild->d_inode->i_atime.tv_sec == v_atime
1356 && dchild->d_inode->i_size == 0 )
1359 case NFS3_CREATE_GUARDED:
1365 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1371 if (EX_ISSYNC(fhp->fh_export)) {
1372 err = nfserrno(nfsd_sync_dir(dentry));
1373 /* setattr will sync the child (or not) */
1376 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1377 /* Cram the verifier into atime/mtime */
1378 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1379 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1380 /* XXX someone who knows this better please fix it for nsec */
1381 iap->ia_mtime.tv_sec = v_mtime;
1382 iap->ia_atime.tv_sec = v_atime;
1383 iap->ia_mtime.tv_nsec = 0;
1384 iap->ia_atime.tv_nsec = 0;
1387 /* Set file attributes.
1388 * Irix appears to send along the gid when it tries to
1389 * implement setgid directories via NFS. Clear out all that cruft.
1392 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1393 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1399 * Update the filehandle to get the new inode info.
1402 err = fh_update(resfhp);
1406 if (dchild && !IS_ERR(dchild))
1411 err = nfserrno(host_err);
1414 #endif /* CONFIG_NFSD_V3 */
1417 * Read a symlink. On entry, *lenp must contain the maximum path length that
1418 * fits into the buffer. On return, it contains the true length.
1419 * N.B. After this call fhp needs an fh_put
1422 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1424 struct dentry *dentry;
1425 struct inode *inode;
1430 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1434 dentry = fhp->fh_dentry;
1435 inode = dentry->d_inode;
1438 if (!inode->i_op || !inode->i_op->readlink)
1441 touch_atime(fhp->fh_export->ex_mnt, dentry);
1442 /* N.B. Why does this call need a get_fs()??
1443 * Remove the set_fs and watch the fireworks:-) --okir
1446 oldfs = get_fs(); set_fs(KERNEL_DS);
1447 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1458 err = nfserrno(host_err);
1463 * Create a symlink and look up its inode
1464 * N.B. After this call _both_ fhp and resfhp need an fh_put
1467 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1468 char *fname, int flen,
1469 char *path, int plen,
1470 struct svc_fh *resfhp,
1473 struct dentry *dentry, *dnew;
1482 if (isdotent(fname, flen))
1485 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1489 dentry = fhp->fh_dentry;
1490 dnew = lookup_one_len(fname, dentry, flen);
1491 host_err = PTR_ERR(dnew);
1496 /* Only the MODE ATTRibute is even vaguely meaningful */
1497 if (iap && (iap->ia_valid & ATTR_MODE))
1498 mode = iap->ia_mode & S_IALLUGO;
1500 if (unlikely(path[plen] != 0)) {
1501 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1502 if (path_alloced == NULL)
1505 strncpy(path_alloced, path, plen);
1506 path_alloced[plen] = 0;
1507 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1508 kfree(path_alloced);
1511 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1514 if (EX_ISSYNC(fhp->fh_export))
1515 host_err = nfsd_sync_dir(dentry);
1517 err = nfserrno(host_err);
1520 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1522 if (err==0) err = cerr;
1527 err = nfserrno(host_err);
1533 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1536 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1537 char *name, int len, struct svc_fh *tfhp)
1539 struct dentry *ddir, *dnew, *dold;
1540 struct inode *dirp, *dest;
1544 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1547 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1555 if (isdotent(name, len))
1558 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1559 ddir = ffhp->fh_dentry;
1560 dirp = ddir->d_inode;
1562 dnew = lookup_one_len(name, ddir, len);
1563 host_err = PTR_ERR(dnew);
1567 dold = tfhp->fh_dentry;
1568 dest = dold->d_inode;
1570 host_err = vfs_link(dold, dirp, dnew);
1572 if (EX_ISSYNC(ffhp->fh_export)) {
1573 err = nfserrno(nfsd_sync_dir(ddir));
1574 write_inode_now(dest, 1);
1578 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1581 err = nfserrno(host_err);
1591 err = nfserrno(host_err);
1597 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1600 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1601 struct svc_fh *tfhp, char *tname, int tlen)
1603 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1604 struct inode *fdir, *tdir;
1608 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1611 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1615 fdentry = ffhp->fh_dentry;
1616 fdir = fdentry->d_inode;
1618 tdentry = tfhp->fh_dentry;
1619 tdir = tdentry->d_inode;
1621 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1622 if (ffhp->fh_export != tfhp->fh_export)
1626 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1629 /* cannot use fh_lock as we need deadlock protective ordering
1630 * so do it by hand */
1631 trap = lock_rename(tdentry, fdentry);
1632 ffhp->fh_locked = tfhp->fh_locked = 1;
1636 odentry = lookup_one_len(fname, fdentry, flen);
1637 host_err = PTR_ERR(odentry);
1638 if (IS_ERR(odentry))
1642 if (!odentry->d_inode)
1645 if (odentry == trap)
1648 ndentry = lookup_one_len(tname, tdentry, tlen);
1649 host_err = PTR_ERR(ndentry);
1650 if (IS_ERR(ndentry))
1652 host_err = -ENOTEMPTY;
1653 if (ndentry == trap)
1657 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1658 ((atomic_read(&odentry->d_count) > 1)
1659 || (atomic_read(&ndentry->d_count) > 1))) {
1663 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1664 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1665 host_err = nfsd_sync_dir(tdentry);
1667 host_err = nfsd_sync_dir(fdentry);
1675 err = nfserrno(host_err);
1677 /* we cannot reply on fh_unlock on the two filehandles,
1678 * as that would do the wrong thing if the two directories
1679 * were the same, so again we do it by hand
1681 fill_post_wcc(ffhp);
1682 fill_post_wcc(tfhp);
1683 unlock_rename(tdentry, fdentry);
1684 ffhp->fh_locked = tfhp->fh_locked = 0;
1691 * Unlink a file or directory
1692 * N.B. After this call fhp needs an fh_put
1695 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1696 char *fname, int flen)
1698 struct dentry *dentry, *rdentry;
1704 if (!flen || isdotent(fname, flen))
1706 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1710 fh_lock_nested(fhp, I_MUTEX_PARENT);
1711 dentry = fhp->fh_dentry;
1712 dirp = dentry->d_inode;
1714 rdentry = lookup_one_len(fname, dentry, flen);
1715 host_err = PTR_ERR(rdentry);
1716 if (IS_ERR(rdentry))
1719 if (!rdentry->d_inode) {
1726 type = rdentry->d_inode->i_mode & S_IFMT;
1728 if (type != S_IFDIR) { /* It's UNLINK */
1730 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1731 (atomic_read(&rdentry->d_count) > 1)) {
1735 host_err = vfs_unlink(dirp, rdentry);
1736 } else { /* It's RMDIR */
1737 host_err = vfs_rmdir(dirp, rdentry);
1744 if (EX_ISSYNC(fhp->fh_export))
1745 host_err = nfsd_sync_dir(dentry);
1748 err = nfserrno(host_err);
1754 * Read entries from a directory.
1755 * The NFSv3/4 verifier we ignore for now.
1758 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1759 struct readdir_cd *cdp, filldir_t func)
1764 loff_t offset = *offsetp;
1766 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1770 offset = vfs_llseek(file, offset, 0);
1772 err = nfserrno((int)offset);
1777 * Read the directory entries. This silly loop is necessary because
1778 * readdir() is not guaranteed to fill up the entire buffer, but
1779 * may choose to do less.
1783 cdp->err = nfserr_eof; /* will be cleared on successful read */
1784 host_err = vfs_readdir(file, func, cdp);
1785 } while (host_err >=0 && cdp->err == nfs_ok);
1787 err = nfserrno(host_err);
1790 *offsetp = vfs_llseek(file, 0, 1);
1792 if (err == nfserr_eof || err == nfserr_toosmall)
1793 err = nfs_ok; /* can still be found in ->err */
1801 * Get file system stats
1802 * N.B. After this call fhp needs an fh_put
1805 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1807 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1808 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1813 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1815 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1819 * Check for a user's access permissions to this inode.
1822 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1823 struct dentry *dentry, int acc)
1825 struct inode *inode = dentry->d_inode;
1831 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1833 (acc & MAY_READ)? " read" : "",
1834 (acc & MAY_WRITE)? " write" : "",
1835 (acc & MAY_EXEC)? " exec" : "",
1836 (acc & MAY_SATTR)? " sattr" : "",
1837 (acc & MAY_TRUNC)? " trunc" : "",
1838 (acc & MAY_LOCK)? " lock" : "",
1839 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1841 IS_IMMUTABLE(inode)? " immut" : "",
1842 IS_APPEND(inode)? " append" : "",
1843 IS_RDONLY(inode)? " ro" : "");
1844 dprintk(" owner %d/%d user %d/%d\n",
1845 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1848 /* Normally we reject any write/sattr etc access on a read-only file
1849 * system. But if it is IRIX doing check on write-access for a
1850 * device special file, we ignore rofs.
1852 if (!(acc & MAY_LOCAL_ACCESS))
1853 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1854 if (exp_rdonly(rqstp, exp) || IS_RDONLY(inode))
1856 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1859 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1862 if (acc & MAY_LOCK) {
1863 /* If we cannot rely on authentication in NLM requests,
1864 * just allow locks, otherwise require read permission, or
1867 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1870 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1873 * The file owner always gets access permission for accesses that
1874 * would normally be checked at open time. This is to make
1875 * file access work even when the client has done a fchmod(fd, 0).
1877 * However, `cp foo bar' should fail nevertheless when bar is
1878 * readonly. A sensible way to do this might be to reject all
1879 * attempts to truncate a read-only file, because a creat() call
1880 * always implies file truncation.
1881 * ... but this isn't really fair. A process may reasonably call
1882 * ftruncate on an open file descriptor on a file with perm 000.
1883 * We must trust the client to do permission checking - using "ACCESS"
1886 if ((acc & MAY_OWNER_OVERRIDE) &&
1887 inode->i_uid == current->fsuid)
1890 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1892 /* Allow read access to binaries even when mode 111 */
1893 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1894 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1895 err = permission(inode, MAY_EXEC, NULL);
1897 return err? nfserrno(err) : 0;
1901 nfsd_racache_shutdown(void)
1905 dprintk("nfsd: freeing readahead buffers.\n");
1910 * Initialize readahead param cache
1913 nfsd_racache_init(int cache_size)
1922 if (cache_size < 2*RAPARM_HASH_SIZE)
1923 cache_size = 2*RAPARM_HASH_SIZE;
1924 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1928 "nfsd: Could not allocate memory read-ahead cache.\n");
1932 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1933 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1934 raparm_hash[i].pb_head = NULL;
1935 spin_lock_init(&raparm_hash[i].pb_lock);
1937 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
1938 for (i = 0; i < cache_size - 1; i++) {
1939 if (i % nperbucket == 0)
1940 raparm_hash[j++].pb_head = raparml + i;
1941 if (i % nperbucket < nperbucket-1)
1942 raparml[i].p_next = raparml + i + 1;
1945 nfsdstats.ra_size = cache_size;
1949 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1951 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1953 struct inode *inode = fhp->fh_dentry->d_inode;
1957 struct posix_acl *acl;
1959 if (!IS_POSIXACL(inode))
1960 return ERR_PTR(-EOPNOTSUPP);
1963 case ACL_TYPE_ACCESS:
1964 name = POSIX_ACL_XATTR_ACCESS;
1966 case ACL_TYPE_DEFAULT:
1967 name = POSIX_ACL_XATTR_DEFAULT;
1970 return ERR_PTR(-EOPNOTSUPP);
1973 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1975 return ERR_PTR(size);
1977 acl = posix_acl_from_xattr(value, size);
1983 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1985 struct inode *inode = fhp->fh_dentry->d_inode;
1991 if (!IS_POSIXACL(inode) || !inode->i_op ||
1992 !inode->i_op->setxattr || !inode->i_op->removexattr)
1995 case ACL_TYPE_ACCESS:
1996 name = POSIX_ACL_XATTR_ACCESS;
1998 case ACL_TYPE_DEFAULT:
1999 name = POSIX_ACL_XATTR_DEFAULT;
2005 if (acl && acl->a_count) {
2006 size = posix_acl_xattr_size(acl->a_count);
2007 value = kmalloc(size, GFP_KERNEL);
2010 error = posix_acl_to_xattr(acl, value, size);
2018 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2020 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2023 error = vfs_removexattr(fhp->fh_dentry, name);
2024 if (error == -ENODATA)
2033 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */