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;
298 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
299 * which only requires access, and "set-[ac]time-to-X" which
300 * requires ownership.
301 * So if it looks like it might be "set both to the same time which
302 * is close to now", and if inode_change_ok fails, then we
303 * convert to "set to now" instead of "set to explicit time"
305 * We only call inode_change_ok as the last test as technically
306 * it is not an interface that we should be using. It is only
307 * valid if the filesystem does not define it's own i_op->setattr.
309 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
310 #define MAX_TOUCH_TIME_ERROR (30*60)
311 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
312 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
314 /* Looks probable. Now just make sure time is in the right ballpark.
315 * Solaris, at least, doesn't seem to care what the time request is.
316 * We require it be within 30 minutes of now.
318 time_t delta = iap->ia_atime.tv_sec - get_seconds();
319 if (delta<0) delta = -delta;
320 if (delta < MAX_TOUCH_TIME_ERROR &&
321 inode_change_ok(inode, iap) != 0) {
322 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
323 * this will cause notify_change to set these times to "now"
325 iap->ia_valid &= ~BOTH_TIME_SET;
329 /* The size case is special. It changes the file as well as the attributes. */
330 if (iap->ia_valid & ATTR_SIZE) {
331 if (iap->ia_size < inode->i_size) {
332 err = nfsd_permission(rqstp, fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
338 * If we are changing the size of the file, then
339 * we need to break all leases.
341 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
342 if (host_err == -EWOULDBLOCK)
343 host_err = -ETIMEDOUT;
344 if (host_err) /* ENOMEM or EWOULDBLOCK */
347 host_err = get_write_access(inode);
352 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
354 put_write_access(inode);
360 imode = inode->i_mode;
361 if (iap->ia_valid & ATTR_MODE) {
362 iap->ia_mode &= S_IALLUGO;
363 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
366 /* Revoke setuid/setgid bit on chown/chgrp */
367 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
368 iap->ia_valid |= ATTR_KILL_SUID;
369 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
370 iap->ia_valid |= ATTR_KILL_SGID;
372 /* Change the attributes. */
374 iap->ia_valid |= ATTR_CTIME;
376 err = nfserr_notsync;
377 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
379 host_err = notify_change(dentry, iap);
380 err = nfserrno(host_err);
384 put_write_access(inode);
386 if (EX_ISSYNC(fhp->fh_export))
387 write_inode_now(inode, 1);
392 err = nfserrno(host_err);
396 #if defined(CONFIG_NFSD_V2_ACL) || \
397 defined(CONFIG_NFSD_V3_ACL) || \
398 defined(CONFIG_NFSD_V4)
399 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
403 buflen = vfs_getxattr(dentry, key, NULL, 0);
407 *buf = kmalloc(buflen, GFP_KERNEL);
411 return vfs_getxattr(dentry, key, *buf, buflen);
415 #if defined(CONFIG_NFSD_V4)
417 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
424 buflen = posix_acl_xattr_size(pacl->a_count);
425 buf = kmalloc(buflen, GFP_KERNEL);
430 len = posix_acl_to_xattr(pacl, buf, buflen);
436 error = vfs_setxattr(dentry, key, buf, len, 0);
443 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
444 struct nfs4_acl *acl)
448 struct dentry *dentry;
450 struct posix_acl *pacl = NULL, *dpacl = NULL;
451 unsigned int flags = 0;
454 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
458 dentry = fhp->fh_dentry;
459 inode = dentry->d_inode;
460 if (S_ISDIR(inode->i_mode))
461 flags = NFS4_ACL_DIR;
463 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
464 if (host_error == -EINVAL) {
465 return nfserr_attrnotsupp;
466 } else if (host_error < 0)
469 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
473 if (S_ISDIR(inode->i_mode))
474 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
477 posix_acl_release(pacl);
478 posix_acl_release(dpacl);
480 if (host_error == -EOPNOTSUPP)
481 return nfserr_attrnotsupp;
483 return nfserrno(host_error);
486 static struct posix_acl *
487 _get_posix_acl(struct dentry *dentry, char *key)
490 struct posix_acl *pacl = NULL;
493 buflen = nfsd_getxattr(dentry, key, &buf);
497 return ERR_PTR(buflen);
499 pacl = posix_acl_from_xattr(buf, buflen);
505 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
507 struct inode *inode = dentry->d_inode;
509 struct posix_acl *pacl = NULL, *dpacl = NULL;
510 unsigned int flags = 0;
512 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
513 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
514 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
516 error = PTR_ERR(pacl);
521 if (S_ISDIR(inode->i_mode)) {
522 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
523 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
525 else if (IS_ERR(dpacl)) {
526 error = PTR_ERR(dpacl);
530 flags = NFS4_ACL_DIR;
533 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
535 error = PTR_ERR(*acl);
539 posix_acl_release(pacl);
540 posix_acl_release(dpacl);
544 #endif /* defined(CONFIG_NFS_V4) */
546 #ifdef CONFIG_NFSD_V3
548 * Check server access rights to a file system object
554 static struct accessmap nfs3_regaccess[] = {
555 { NFS3_ACCESS_READ, MAY_READ },
556 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
557 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
558 { NFS3_ACCESS_EXTEND, MAY_WRITE },
563 static struct accessmap nfs3_diraccess[] = {
564 { NFS3_ACCESS_READ, MAY_READ },
565 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
566 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
567 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
568 { NFS3_ACCESS_DELETE, MAY_REMOVE },
573 static struct accessmap nfs3_anyaccess[] = {
574 /* Some clients - Solaris 2.6 at least, make an access call
575 * to the server to check for access for things like /dev/null
576 * (which really, the server doesn't care about). So
577 * We provide simple access checking for them, looking
578 * mainly at mode bits, and we make sure to ignore read-only
581 { NFS3_ACCESS_READ, MAY_READ },
582 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
583 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
584 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
590 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
592 struct accessmap *map;
593 struct svc_export *export;
594 struct dentry *dentry;
595 u32 query, result = 0, sresult = 0;
598 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
602 export = fhp->fh_export;
603 dentry = fhp->fh_dentry;
605 if (S_ISREG(dentry->d_inode->i_mode))
606 map = nfs3_regaccess;
607 else if (S_ISDIR(dentry->d_inode->i_mode))
608 map = nfs3_diraccess;
610 map = nfs3_anyaccess;
614 for (; map->access; map++) {
615 if (map->access & query) {
618 sresult |= map->access;
620 err2 = nfsd_permission(rqstp, export, dentry, map->how);
623 result |= map->access;
626 /* the following error codes just mean the access was not allowed,
627 * rather than an error occurred */
631 /* simply don't "or" in the access bit. */
641 *supported = sresult;
646 #endif /* CONFIG_NFSD_V3 */
651 * Open an existing file or directory.
652 * The access argument indicates the type of open (read/write/lock)
653 * N.B. After this call fhp needs an fh_put
656 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
657 int access, struct file **filp)
659 struct dentry *dentry;
661 int flags = O_RDONLY|O_LARGEFILE;
666 * If we get here, then the client has already done an "open",
667 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
668 * in case a chmod has now revoked permission.
670 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
674 dentry = fhp->fh_dentry;
675 inode = dentry->d_inode;
677 /* Disallow write access to files with the append-only bit set
678 * or any access when mandatory locking enabled
681 if (IS_APPEND(inode) && (access & MAY_WRITE))
683 if (IS_ISMNDLK(inode))
690 * Check to see if there are any leases on this file.
691 * This may block while leases are broken.
693 host_err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
694 if (host_err == -EWOULDBLOCK)
695 host_err = -ETIMEDOUT;
696 if (host_err) /* NOMEM or WOULDBLOCK */
699 if (access & MAY_WRITE) {
700 if (access & MAY_READ)
701 flags = O_RDWR|O_LARGEFILE;
703 flags = O_WRONLY|O_LARGEFILE;
707 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
709 host_err = PTR_ERR(*filp);
711 err = nfserrno(host_err);
720 nfsd_close(struct file *filp)
727 * As this calls fsync (not fdatasync) there is no need for a write_inode
730 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
731 const struct file_operations *fop)
733 struct inode *inode = dp->d_inode;
734 int (*fsync) (struct file *, struct dentry *, int);
737 err = filemap_fdatawrite(inode->i_mapping);
738 if (err == 0 && fop && (fsync = fop->fsync))
739 err = fsync(filp, dp, 0);
741 err = filemap_fdatawait(inode->i_mapping);
748 nfsd_sync(struct file *filp)
751 struct inode *inode = filp->f_path.dentry->d_inode;
752 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
753 mutex_lock(&inode->i_mutex);
754 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
755 mutex_unlock(&inode->i_mutex);
761 nfsd_sync_dir(struct dentry *dp)
763 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
767 * Obtain the readahead parameters for the file
768 * specified by (dev, ino).
771 static inline struct raparms *
772 nfsd_get_raparms(dev_t dev, ino_t ino)
774 struct raparms *ra, **rap, **frap = NULL;
777 struct raparm_hbucket *rab;
779 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
780 rab = &raparm_hash[hash];
782 spin_lock(&rab->pb_lock);
783 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
784 if (ra->p_ino == ino && ra->p_dev == dev)
787 if (ra->p_count == 0)
790 depth = nfsdstats.ra_size*11/10;
792 spin_unlock(&rab->pb_lock);
802 if (rap != &rab->pb_head) {
804 ra->p_next = rab->pb_head;
808 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
809 spin_unlock(&rab->pb_lock);
814 * Grab and keep cached pages associated with a file in the svc_rqst
815 * so that they can be passed to the network sendmsg/sendpage routines
816 * directly. They will be released after the sending has completed.
819 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
820 struct splice_desc *sd)
822 struct svc_rqst *rqstp = sd->u.data;
823 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
824 struct page *page = buf->page;
828 ret = buf->ops->confirm(pipe, buf);
834 if (rqstp->rq_res.page_len == 0) {
839 rqstp->rq_res.page_base = buf->offset;
840 rqstp->rq_res.page_len = size;
841 } else if (page != pp[-1]) {
847 rqstp->rq_res.page_len += size;
849 rqstp->rq_res.page_len += size;
854 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
855 struct splice_desc *sd)
857 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
861 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
862 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
871 inode = file->f_path.dentry->d_inode;
873 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
874 (!lock_may_read(inode, offset, *count)))
878 /* Get readahead parameters */
879 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
882 file->f_ra = ra->p_ra;
884 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
885 struct splice_desc sd = {
892 rqstp->rq_resused = 1;
893 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
897 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
901 /* Write back readahead params */
903 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
904 spin_lock(&rab->pb_lock);
905 ra->p_ra = file->f_ra;
908 spin_unlock(&rab->pb_lock);
912 nfsdstats.io_read += host_err;
915 fsnotify_access(file->f_path.dentry);
917 err = nfserrno(host_err);
922 static void kill_suid(struct dentry *dentry)
925 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
927 mutex_lock(&dentry->d_inode->i_mutex);
928 notify_change(dentry, &ia);
929 mutex_unlock(&dentry->d_inode->i_mutex);
933 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
934 loff_t offset, struct kvec *vec, int vlen,
935 unsigned long cnt, int *stablep)
937 struct svc_export *exp;
938 struct dentry *dentry;
943 int stable = *stablep;
948 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
949 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
953 dentry = file->f_path.dentry;
954 inode = dentry->d_inode;
955 exp = fhp->fh_export;
958 * Request sync writes if
959 * - the sync export option has been set, or
960 * - the client requested O_SYNC behavior (NFSv3 feature).
961 * - The file system doesn't support fsync().
962 * When gathered writes have been configured for this volume,
963 * flushing the data to disk is handled separately below.
966 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
968 *stablep = 2; /* FILE_SYNC */
973 if (stable && !EX_WGATHER(exp))
974 file->f_flags |= O_SYNC;
976 /* Write the data. */
977 oldfs = get_fs(); set_fs(KERNEL_DS);
978 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
981 nfsdstats.io_write += cnt;
982 fsnotify_modify(file->f_path.dentry);
985 /* clear setuid/setgid flag after write */
986 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
989 if (host_err >= 0 && stable) {
990 static ino_t last_ino;
991 static dev_t last_dev;
994 * Gathered writes: If another process is currently
995 * writing to the file, there's a high chance
996 * this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the
998 * file with each write request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates
1001 * C. Juszak's idea of gathered writes, but it's a
1002 * nice and simple solution (IMHO), and it seems to
1005 if (EX_WGATHER(exp)) {
1006 if (atomic_read(&inode->i_writecount) > 1
1007 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1008 dprintk("nfsd: write defer %d\n", current->pid);
1010 dprintk("nfsd: write resume %d\n", current->pid);
1013 if (inode->i_state & I_DIRTY) {
1014 dprintk("nfsd: write sync %d\n", current->pid);
1015 host_err=nfsd_sync(file);
1018 wake_up(&inode->i_wait);
1021 last_ino = inode->i_ino;
1022 last_dev = inode->i_sb->s_dev;
1025 dprintk("nfsd: write complete host_err=%d\n", host_err);
1029 err = nfserrno(host_err);
1035 * Read data from a file. count must contain the requested read count
1036 * on entry. On return, *count contains the number of bytes actually read.
1037 * N.B. After this call fhp needs an fh_put
1040 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1041 loff_t offset, struct kvec *vec, int vlen,
1042 unsigned long *count)
1047 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1048 MAY_READ|MAY_OWNER_OVERRIDE);
1051 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1053 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1056 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1064 * Write data to a file.
1065 * The stable flag requests synchronous writes.
1066 * N.B. After this call fhp needs an fh_put
1069 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1070 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1076 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1077 MAY_WRITE|MAY_OWNER_OVERRIDE);
1080 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1083 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1088 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1096 #ifdef CONFIG_NFSD_V3
1098 * Commit all pending writes to stable storage.
1099 * Strictly speaking, we could sync just the indicated file region here,
1100 * but there's currently no way we can ask the VFS to do so.
1102 * Unfortunately we cannot lock the file to make sure we return full WCC
1103 * data to the client, as locking happens lower down in the filesystem.
1106 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1107 loff_t offset, unsigned long count)
1112 if ((u64)count > ~(u64)offset)
1113 return nfserr_inval;
1115 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1117 if (EX_ISSYNC(fhp->fh_export)) {
1118 if (file->f_op && file->f_op->fsync) {
1119 err = nfserrno(nfsd_sync(file));
1121 err = nfserr_notsupp;
1128 #endif /* CONFIG_NFSD_V3 */
1131 * Create a file (regular, directory, device, fifo); UNIX sockets
1132 * not yet implemented.
1133 * If the response fh has been verified, the parent directory should
1134 * already be locked. Note that the parent directory is left locked.
1136 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1139 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1140 char *fname, int flen, struct iattr *iap,
1141 int type, dev_t rdev, struct svc_fh *resfhp)
1143 struct dentry *dentry, *dchild = NULL;
1152 if (isdotent(fname, flen))
1155 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1159 dentry = fhp->fh_dentry;
1160 dirp = dentry->d_inode;
1162 err = nfserr_notdir;
1163 if(!dirp->i_op || !dirp->i_op->lookup)
1166 * Check whether the response file handle has been verified yet.
1167 * If it has, the parent directory should already be locked.
1169 if (!resfhp->fh_dentry) {
1170 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1171 fh_lock_nested(fhp, I_MUTEX_PARENT);
1172 dchild = lookup_one_len(fname, dentry, flen);
1173 host_err = PTR_ERR(dchild);
1176 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1180 /* called from nfsd_proc_create */
1181 dchild = dget(resfhp->fh_dentry);
1182 if (!fhp->fh_locked) {
1183 /* not actually possible */
1185 "nfsd_create: parent %s/%s not locked!\n",
1186 dentry->d_parent->d_name.name,
1187 dentry->d_name.name);
1193 * Make sure the child dentry is still negative ...
1196 if (dchild->d_inode) {
1197 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1198 dentry->d_name.name, dchild->d_name.name);
1202 if (!(iap->ia_valid & ATTR_MODE))
1204 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1207 * Get the dir op function pointer.
1212 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1215 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1221 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1224 printk("nfsd: bad file type %o in nfsd_create\n", type);
1230 if (EX_ISSYNC(fhp->fh_export)) {
1231 err = nfserrno(nfsd_sync_dir(dentry));
1232 write_inode_now(dchild->d_inode, 1);
1236 /* Set file attributes. Mode has already been set and
1237 * setting uid/gid works only for root. Irix appears to
1238 * send along the gid when it tries to implement setgid
1239 * directories via NFS.
1241 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1242 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1247 * Update the file handle to get the new inode info.
1250 err = fh_update(resfhp);
1252 if (dchild && !IS_ERR(dchild))
1257 err = nfserrno(host_err);
1261 #ifdef CONFIG_NFSD_V3
1263 * NFSv3 version of nfsd_create
1266 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1267 char *fname, int flen, struct iattr *iap,
1268 struct svc_fh *resfhp, int createmode, u32 *verifier,
1269 int *truncp, int *created)
1271 struct dentry *dentry, *dchild = NULL;
1275 __u32 v_mtime=0, v_atime=0;
1281 if (isdotent(fname, flen))
1283 if (!(iap->ia_valid & ATTR_MODE))
1285 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1289 dentry = fhp->fh_dentry;
1290 dirp = dentry->d_inode;
1292 /* Get all the sanity checks out of the way before
1293 * we lock the parent. */
1294 err = nfserr_notdir;
1295 if(!dirp->i_op || !dirp->i_op->lookup)
1297 fh_lock_nested(fhp, I_MUTEX_PARENT);
1300 * Compose the response file handle.
1302 dchild = lookup_one_len(fname, dentry, flen);
1303 host_err = PTR_ERR(dchild);
1307 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1311 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1312 /* solaris7 gets confused (bugid 4218508) if these have
1313 * the high bit set, so just clear the high bits. If this is
1314 * ever changed to use different attrs for storing the
1315 * verifier, then do_open_lookup() will also need to be fixed
1318 v_mtime = verifier[0]&0x7fffffff;
1319 v_atime = verifier[1]&0x7fffffff;
1322 if (dchild->d_inode) {
1325 switch (createmode) {
1326 case NFS3_CREATE_UNCHECKED:
1327 if (! S_ISREG(dchild->d_inode->i_mode))
1330 /* in nfsv4, we need to treat this case a little
1331 * differently. we don't want to truncate the
1332 * file now; this would be wrong if the OPEN
1333 * fails for some other reason. furthermore,
1334 * if the size is nonzero, we should ignore it
1335 * according to spec!
1337 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1340 iap->ia_valid &= ATTR_SIZE;
1344 case NFS3_CREATE_EXCLUSIVE:
1345 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1346 && dchild->d_inode->i_atime.tv_sec == v_atime
1347 && dchild->d_inode->i_size == 0 )
1350 case NFS3_CREATE_GUARDED:
1356 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1362 if (EX_ISSYNC(fhp->fh_export)) {
1363 err = nfserrno(nfsd_sync_dir(dentry));
1364 /* setattr will sync the child (or not) */
1367 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1368 /* Cram the verifier into atime/mtime */
1369 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1370 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1371 /* XXX someone who knows this better please fix it for nsec */
1372 iap->ia_mtime.tv_sec = v_mtime;
1373 iap->ia_atime.tv_sec = v_atime;
1374 iap->ia_mtime.tv_nsec = 0;
1375 iap->ia_atime.tv_nsec = 0;
1378 /* Set file attributes.
1379 * Irix appears to send along the gid when it tries to
1380 * implement setgid directories via NFS. Clear out all that cruft.
1383 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1384 __be32 err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1390 * Update the filehandle to get the new inode info.
1393 err = fh_update(resfhp);
1397 if (dchild && !IS_ERR(dchild))
1402 err = nfserrno(host_err);
1405 #endif /* CONFIG_NFSD_V3 */
1408 * Read a symlink. On entry, *lenp must contain the maximum path length that
1409 * fits into the buffer. On return, it contains the true length.
1410 * N.B. After this call fhp needs an fh_put
1413 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1415 struct dentry *dentry;
1416 struct inode *inode;
1421 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1425 dentry = fhp->fh_dentry;
1426 inode = dentry->d_inode;
1429 if (!inode->i_op || !inode->i_op->readlink)
1432 touch_atime(fhp->fh_export->ex_mnt, dentry);
1433 /* N.B. Why does this call need a get_fs()??
1434 * Remove the set_fs and watch the fireworks:-) --okir
1437 oldfs = get_fs(); set_fs(KERNEL_DS);
1438 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1449 err = nfserrno(host_err);
1454 * Create a symlink and look up its inode
1455 * N.B. After this call _both_ fhp and resfhp need an fh_put
1458 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1459 char *fname, int flen,
1460 char *path, int plen,
1461 struct svc_fh *resfhp,
1464 struct dentry *dentry, *dnew;
1473 if (isdotent(fname, flen))
1476 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1480 dentry = fhp->fh_dentry;
1481 dnew = lookup_one_len(fname, dentry, flen);
1482 host_err = PTR_ERR(dnew);
1487 /* Only the MODE ATTRibute is even vaguely meaningful */
1488 if (iap && (iap->ia_valid & ATTR_MODE))
1489 mode = iap->ia_mode & S_IALLUGO;
1491 if (unlikely(path[plen] != 0)) {
1492 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1493 if (path_alloced == NULL)
1496 strncpy(path_alloced, path, plen);
1497 path_alloced[plen] = 0;
1498 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1499 kfree(path_alloced);
1502 host_err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1505 if (EX_ISSYNC(fhp->fh_export))
1506 host_err = nfsd_sync_dir(dentry);
1508 err = nfserrno(host_err);
1511 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1513 if (err==0) err = cerr;
1518 err = nfserrno(host_err);
1524 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1527 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1528 char *name, int len, struct svc_fh *tfhp)
1530 struct dentry *ddir, *dnew, *dold;
1531 struct inode *dirp, *dest;
1535 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1538 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1546 if (isdotent(name, len))
1549 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1550 ddir = ffhp->fh_dentry;
1551 dirp = ddir->d_inode;
1553 dnew = lookup_one_len(name, ddir, len);
1554 host_err = PTR_ERR(dnew);
1558 dold = tfhp->fh_dentry;
1559 dest = dold->d_inode;
1561 host_err = vfs_link(dold, dirp, dnew);
1563 if (EX_ISSYNC(ffhp->fh_export)) {
1564 err = nfserrno(nfsd_sync_dir(ddir));
1565 write_inode_now(dest, 1);
1569 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1572 err = nfserrno(host_err);
1582 err = nfserrno(host_err);
1588 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1591 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1592 struct svc_fh *tfhp, char *tname, int tlen)
1594 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1595 struct inode *fdir, *tdir;
1599 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1602 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1606 fdentry = ffhp->fh_dentry;
1607 fdir = fdentry->d_inode;
1609 tdentry = tfhp->fh_dentry;
1610 tdir = tdentry->d_inode;
1612 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1613 if (ffhp->fh_export != tfhp->fh_export)
1617 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1620 /* cannot use fh_lock as we need deadlock protective ordering
1621 * so do it by hand */
1622 trap = lock_rename(tdentry, fdentry);
1623 ffhp->fh_locked = tfhp->fh_locked = 1;
1627 odentry = lookup_one_len(fname, fdentry, flen);
1628 host_err = PTR_ERR(odentry);
1629 if (IS_ERR(odentry))
1633 if (!odentry->d_inode)
1636 if (odentry == trap)
1639 ndentry = lookup_one_len(tname, tdentry, tlen);
1640 host_err = PTR_ERR(ndentry);
1641 if (IS_ERR(ndentry))
1643 host_err = -ENOTEMPTY;
1644 if (ndentry == trap)
1648 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1649 ((atomic_read(&odentry->d_count) > 1)
1650 || (atomic_read(&ndentry->d_count) > 1))) {
1654 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1655 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1656 host_err = nfsd_sync_dir(tdentry);
1658 host_err = nfsd_sync_dir(fdentry);
1666 err = nfserrno(host_err);
1668 /* we cannot reply on fh_unlock on the two filehandles,
1669 * as that would do the wrong thing if the two directories
1670 * were the same, so again we do it by hand
1672 fill_post_wcc(ffhp);
1673 fill_post_wcc(tfhp);
1674 unlock_rename(tdentry, fdentry);
1675 ffhp->fh_locked = tfhp->fh_locked = 0;
1682 * Unlink a file or directory
1683 * N.B. After this call fhp needs an fh_put
1686 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1687 char *fname, int flen)
1689 struct dentry *dentry, *rdentry;
1695 if (!flen || isdotent(fname, flen))
1697 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1701 fh_lock_nested(fhp, I_MUTEX_PARENT);
1702 dentry = fhp->fh_dentry;
1703 dirp = dentry->d_inode;
1705 rdentry = lookup_one_len(fname, dentry, flen);
1706 host_err = PTR_ERR(rdentry);
1707 if (IS_ERR(rdentry))
1710 if (!rdentry->d_inode) {
1717 type = rdentry->d_inode->i_mode & S_IFMT;
1719 if (type != S_IFDIR) { /* It's UNLINK */
1721 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1722 (atomic_read(&rdentry->d_count) > 1)) {
1726 host_err = vfs_unlink(dirp, rdentry);
1727 } else { /* It's RMDIR */
1728 host_err = vfs_rmdir(dirp, rdentry);
1735 if (EX_ISSYNC(fhp->fh_export))
1736 host_err = nfsd_sync_dir(dentry);
1739 err = nfserrno(host_err);
1745 * Read entries from a directory.
1746 * The NFSv3/4 verifier we ignore for now.
1749 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1750 struct readdir_cd *cdp, filldir_t func)
1755 loff_t offset = *offsetp;
1757 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1761 offset = vfs_llseek(file, offset, 0);
1763 err = nfserrno((int)offset);
1768 * Read the directory entries. This silly loop is necessary because
1769 * readdir() is not guaranteed to fill up the entire buffer, but
1770 * may choose to do less.
1774 cdp->err = nfserr_eof; /* will be cleared on successful read */
1775 host_err = vfs_readdir(file, func, cdp);
1776 } while (host_err >=0 && cdp->err == nfs_ok);
1778 err = nfserrno(host_err);
1781 *offsetp = vfs_llseek(file, 0, 1);
1783 if (err == nfserr_eof || err == nfserr_toosmall)
1784 err = nfs_ok; /* can still be found in ->err */
1792 * Get file system stats
1793 * N.B. After this call fhp needs an fh_put
1796 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1798 __be32 err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1799 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1804 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1806 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1810 * Check for a user's access permissions to this inode.
1813 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1814 struct dentry *dentry, int acc)
1816 struct inode *inode = dentry->d_inode;
1822 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1824 (acc & MAY_READ)? " read" : "",
1825 (acc & MAY_WRITE)? " write" : "",
1826 (acc & MAY_EXEC)? " exec" : "",
1827 (acc & MAY_SATTR)? " sattr" : "",
1828 (acc & MAY_TRUNC)? " trunc" : "",
1829 (acc & MAY_LOCK)? " lock" : "",
1830 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1832 IS_IMMUTABLE(inode)? " immut" : "",
1833 IS_APPEND(inode)? " append" : "",
1834 IS_RDONLY(inode)? " ro" : "");
1835 dprintk(" owner %d/%d user %d/%d\n",
1836 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1839 /* Normally we reject any write/sattr etc access on a read-only file
1840 * system. But if it is IRIX doing check on write-access for a
1841 * device special file, we ignore rofs.
1843 if (!(acc & MAY_LOCAL_ACCESS))
1844 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1845 if (exp_rdonly(rqstp, exp) || IS_RDONLY(inode))
1847 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1850 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1853 if (acc & MAY_LOCK) {
1854 /* If we cannot rely on authentication in NLM requests,
1855 * just allow locks, otherwise require read permission, or
1858 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1861 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1864 * The file owner always gets access permission for accesses that
1865 * would normally be checked at open time. This is to make
1866 * file access work even when the client has done a fchmod(fd, 0).
1868 * However, `cp foo bar' should fail nevertheless when bar is
1869 * readonly. A sensible way to do this might be to reject all
1870 * attempts to truncate a read-only file, because a creat() call
1871 * always implies file truncation.
1872 * ... but this isn't really fair. A process may reasonably call
1873 * ftruncate on an open file descriptor on a file with perm 000.
1874 * We must trust the client to do permission checking - using "ACCESS"
1877 if ((acc & MAY_OWNER_OVERRIDE) &&
1878 inode->i_uid == current->fsuid)
1881 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1883 /* Allow read access to binaries even when mode 111 */
1884 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1885 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1886 err = permission(inode, MAY_EXEC, NULL);
1888 return err? nfserrno(err) : 0;
1892 nfsd_racache_shutdown(void)
1896 dprintk("nfsd: freeing readahead buffers.\n");
1901 * Initialize readahead param cache
1904 nfsd_racache_init(int cache_size)
1913 if (cache_size < 2*RAPARM_HASH_SIZE)
1914 cache_size = 2*RAPARM_HASH_SIZE;
1915 raparml = kcalloc(cache_size, sizeof(struct raparms), GFP_KERNEL);
1919 "nfsd: Could not allocate memory read-ahead cache.\n");
1923 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
1924 for (i = 0 ; i < RAPARM_HASH_SIZE ; i++) {
1925 raparm_hash[i].pb_head = NULL;
1926 spin_lock_init(&raparm_hash[i].pb_lock);
1928 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
1929 for (i = 0; i < cache_size - 1; i++) {
1930 if (i % nperbucket == 0)
1931 raparm_hash[j++].pb_head = raparml + i;
1932 if (i % nperbucket < nperbucket-1)
1933 raparml[i].p_next = raparml + i + 1;
1936 nfsdstats.ra_size = cache_size;
1940 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1942 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1944 struct inode *inode = fhp->fh_dentry->d_inode;
1948 struct posix_acl *acl;
1950 if (!IS_POSIXACL(inode))
1951 return ERR_PTR(-EOPNOTSUPP);
1954 case ACL_TYPE_ACCESS:
1955 name = POSIX_ACL_XATTR_ACCESS;
1957 case ACL_TYPE_DEFAULT:
1958 name = POSIX_ACL_XATTR_DEFAULT;
1961 return ERR_PTR(-EOPNOTSUPP);
1964 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1966 return ERR_PTR(size);
1968 acl = posix_acl_from_xattr(value, size);
1974 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1976 struct inode *inode = fhp->fh_dentry->d_inode;
1982 if (!IS_POSIXACL(inode) || !inode->i_op ||
1983 !inode->i_op->setxattr || !inode->i_op->removexattr)
1986 case ACL_TYPE_ACCESS:
1987 name = POSIX_ACL_XATTR_ACCESS;
1989 case ACL_TYPE_DEFAULT:
1990 name = POSIX_ACL_XATTR_DEFAULT;
1996 if (acl && acl->a_count) {
1997 size = posix_acl_xattr_size(acl->a_count);
1998 value = kmalloc(size, GFP_KERNEL);
2001 error = posix_acl_to_xattr(acl, value, size);
2009 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2011 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2014 error = vfs_removexattr(fhp->fh_dentry, name);
2015 if (error == -ENODATA)
2024 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */