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
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
72 struct raparms *p_next;
77 struct file_ra_state p_ra;
78 unsigned int p_hindex;
81 struct raparm_hbucket {
82 struct raparms *pb_head;
84 } ____cacheline_aligned_in_smp;
86 #define RAPARM_HASH_BITS 4
87 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
88 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
89 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
95 * or nfs_ok having possibly changed *dpp and *expp
98 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
99 struct svc_export **expp)
101 struct svc_export *exp = *expp, *exp2 = NULL;
102 struct dentry *dentry = *dpp;
103 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
104 struct dentry *mounts = dget(dentry);
107 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
109 exp2 = rqst_exp_get_by_name(rqstp, mnt, mounts);
111 if (PTR_ERR(exp2) != -ENOENT)
117 if ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
118 /* successfully crossed mount point */
133 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
134 const char *name, unsigned int len,
135 struct svc_export **exp_ret, struct dentry **dentry_ret)
137 struct svc_export *exp;
138 struct dentry *dparent;
139 struct dentry *dentry;
143 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
145 /* Obtain dentry and export. */
146 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
150 dparent = fhp->fh_dentry;
151 exp = fhp->fh_export;
154 /* Lookup the name, but don't follow links */
155 if (isdotent(name, len)) {
157 dentry = dget(dparent);
158 else if (dparent != exp->ex_path.dentry)
159 dentry = dget_parent(dparent);
160 else if (!EX_NOHIDE(exp))
161 dentry = dget(dparent); /* .. == . just like at / */
163 /* checking mountpoint crossing is very different when stepping up */
164 struct svc_export *exp2 = NULL;
166 struct vfsmount *mnt = mntget(exp->ex_path.mnt);
167 dentry = dget(dparent);
168 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
170 dp = dget_parent(dentry);
174 exp2 = rqst_exp_parent(rqstp, mnt, dentry);
175 if (PTR_ERR(exp2) == -ENOENT) {
177 dentry = dget(dparent);
178 } else if (IS_ERR(exp2)) {
179 host_err = PTR_ERR(exp2);
191 dentry = lookup_one_len(name, dparent, len);
192 host_err = PTR_ERR(dentry);
196 * check if we have crossed a mount point ...
198 if (d_mountpoint(dentry)) {
199 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
205 *dentry_ret = dentry;
211 return nfserrno(host_err);
215 * Look up one component of a pathname.
216 * N.B. After this call _both_ fhp and resfh need an fh_put
218 * If the lookup would cross a mountpoint, and the mounted filesystem
219 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
220 * accepted as it stands and the mounted directory is
221 * returned. Otherwise the covered directory is returned.
222 * NOTE: this mountpoint crossing is not supported properly by all
223 * clients and is explicitly disallowed for NFSv3
224 * NeilBrown <neilb@cse.unsw.edu.au>
227 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
228 unsigned int len, struct svc_fh *resfh)
230 struct svc_export *exp;
231 struct dentry *dentry;
234 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
237 err = check_nfsd_access(exp, rqstp);
241 * Note: we compose the file handle now, but as the
242 * dentry may be negative, it may need to be updated.
244 err = fh_compose(resfh, exp, dentry, fhp);
245 if (!err && !dentry->d_inode)
255 * Set various file attributes.
256 * N.B. After this call fhp needs an fh_put
259 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
260 int check_guard, time_t guardtime)
262 struct dentry *dentry;
264 int accmode = NFSD_MAY_SATTR;
270 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
271 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
272 if (iap->ia_valid & ATTR_SIZE)
276 err = fh_verify(rqstp, fhp, ftype, accmode);
280 dentry = fhp->fh_dentry;
281 inode = dentry->d_inode;
283 /* Ignore any mode updates on symlinks */
284 if (S_ISLNK(inode->i_mode))
285 iap->ia_valid &= ~ATTR_MODE;
291 * NFSv2 does not differentiate between "set-[ac]time-to-now"
292 * which only requires access, and "set-[ac]time-to-X" which
293 * requires ownership.
294 * So if it looks like it might be "set both to the same time which
295 * is close to now", and if inode_change_ok fails, then we
296 * convert to "set to now" instead of "set to explicit time"
298 * We only call inode_change_ok as the last test as technically
299 * it is not an interface that we should be using. It is only
300 * valid if the filesystem does not define it's own i_op->setattr.
302 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
303 #define MAX_TOUCH_TIME_ERROR (30*60)
304 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
305 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
309 * Now just make sure time is in the right ballpark.
310 * Solaris, at least, doesn't seem to care what the time
311 * request is. We require it be within 30 minutes of now.
313 time_t delta = iap->ia_atime.tv_sec - get_seconds();
316 if (delta < MAX_TOUCH_TIME_ERROR &&
317 inode_change_ok(inode, iap) != 0) {
319 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
320 * This will cause notify_change to set these times
323 iap->ia_valid &= ~BOTH_TIME_SET;
328 * The size case is special.
329 * It changes the file as well as the attributes.
331 if (iap->ia_valid & ATTR_SIZE) {
332 if (iap->ia_size < inode->i_size) {
333 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
334 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
340 * If we are changing the size of the file, then
341 * we need to break all leases.
343 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
344 if (host_err == -EWOULDBLOCK)
345 host_err = -ETIMEDOUT;
346 if (host_err) /* ENOMEM or EWOULDBLOCK */
349 host_err = get_write_access(inode);
354 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
356 put_write_access(inode);
362 /* sanitize the mode change */
363 if (iap->ia_valid & ATTR_MODE) {
364 iap->ia_mode &= S_IALLUGO;
365 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
368 /* Revoke setuid/setgid on chown */
369 if (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
370 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)) {
371 iap->ia_valid |= ATTR_KILL_PRIV;
372 if (iap->ia_valid & ATTR_MODE) {
373 /* we're setting mode too, just clear the s*id bits */
374 iap->ia_mode &= ~S_ISUID;
375 if (iap->ia_mode & S_IXGRP)
376 iap->ia_mode &= ~S_ISGID;
378 /* set ATTR_KILL_* bits and let VFS handle it */
379 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
383 /* Change the attributes. */
385 iap->ia_valid |= ATTR_CTIME;
387 err = nfserr_notsync;
388 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
390 host_err = notify_change(dentry, iap);
391 err = nfserrno(host_err);
395 put_write_access(inode);
397 if (EX_ISSYNC(fhp->fh_export))
398 write_inode_now(inode, 1);
403 err = nfserrno(host_err);
407 #if defined(CONFIG_NFSD_V2_ACL) || \
408 defined(CONFIG_NFSD_V3_ACL) || \
409 defined(CONFIG_NFSD_V4)
410 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
415 buflen = vfs_getxattr(dentry, key, NULL, 0);
419 *buf = kmalloc(buflen, GFP_KERNEL);
423 ret = vfs_getxattr(dentry, key, *buf, buflen);
430 #if defined(CONFIG_NFSD_V4)
432 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
439 buflen = posix_acl_xattr_size(pacl->a_count);
440 buf = kmalloc(buflen, GFP_KERNEL);
445 len = posix_acl_to_xattr(pacl, buf, buflen);
451 error = vfs_setxattr(dentry, key, buf, len, 0);
458 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
459 struct nfs4_acl *acl)
463 struct dentry *dentry;
465 struct posix_acl *pacl = NULL, *dpacl = NULL;
466 unsigned int flags = 0;
469 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
473 dentry = fhp->fh_dentry;
474 inode = dentry->d_inode;
475 if (S_ISDIR(inode->i_mode))
476 flags = NFS4_ACL_DIR;
478 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
479 if (host_error == -EINVAL) {
480 return nfserr_attrnotsupp;
481 } else if (host_error < 0)
484 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
488 if (S_ISDIR(inode->i_mode))
489 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
492 posix_acl_release(pacl);
493 posix_acl_release(dpacl);
495 if (host_error == -EOPNOTSUPP)
496 return nfserr_attrnotsupp;
498 return nfserrno(host_error);
501 static struct posix_acl *
502 _get_posix_acl(struct dentry *dentry, char *key)
505 struct posix_acl *pacl = NULL;
508 buflen = nfsd_getxattr(dentry, key, &buf);
512 return ERR_PTR(buflen);
514 pacl = posix_acl_from_xattr(buf, buflen);
520 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
522 struct inode *inode = dentry->d_inode;
524 struct posix_acl *pacl = NULL, *dpacl = NULL;
525 unsigned int flags = 0;
527 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
528 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
529 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
531 error = PTR_ERR(pacl);
536 if (S_ISDIR(inode->i_mode)) {
537 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
538 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
540 else if (IS_ERR(dpacl)) {
541 error = PTR_ERR(dpacl);
545 flags = NFS4_ACL_DIR;
548 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
550 error = PTR_ERR(*acl);
554 posix_acl_release(pacl);
555 posix_acl_release(dpacl);
559 #endif /* defined(CONFIG_NFS_V4) */
561 #ifdef CONFIG_NFSD_V3
563 * Check server access rights to a file system object
569 static struct accessmap nfs3_regaccess[] = {
570 { NFS3_ACCESS_READ, NFSD_MAY_READ },
571 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
572 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
573 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
578 static struct accessmap nfs3_diraccess[] = {
579 { NFS3_ACCESS_READ, NFSD_MAY_READ },
580 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
581 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
582 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
583 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
588 static struct accessmap nfs3_anyaccess[] = {
589 /* Some clients - Solaris 2.6 at least, make an access call
590 * to the server to check for access for things like /dev/null
591 * (which really, the server doesn't care about). So
592 * We provide simple access checking for them, looking
593 * mainly at mode bits, and we make sure to ignore read-only
596 { NFS3_ACCESS_READ, NFSD_MAY_READ },
597 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
598 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
599 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
605 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
607 struct accessmap *map;
608 struct svc_export *export;
609 struct dentry *dentry;
610 u32 query, result = 0, sresult = 0;
613 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
617 export = fhp->fh_export;
618 dentry = fhp->fh_dentry;
620 if (S_ISREG(dentry->d_inode->i_mode))
621 map = nfs3_regaccess;
622 else if (S_ISDIR(dentry->d_inode->i_mode))
623 map = nfs3_diraccess;
625 map = nfs3_anyaccess;
629 for (; map->access; map++) {
630 if (map->access & query) {
633 sresult |= map->access;
635 err2 = nfsd_permission(rqstp, export, dentry, map->how);
638 result |= map->access;
641 /* the following error codes just mean the access was not allowed,
642 * rather than an error occurred */
646 /* simply don't "or" in the access bit. */
656 *supported = sresult;
661 #endif /* CONFIG_NFSD_V3 */
666 * Open an existing file or directory.
667 * The access argument indicates the type of open (read/write/lock)
668 * N.B. After this call fhp needs an fh_put
671 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
672 int access, struct file **filp)
674 const struct cred *cred = current_cred();
675 struct dentry *dentry;
677 int flags = O_RDONLY|O_LARGEFILE;
682 * If we get here, then the client has already done an "open",
683 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
684 * in case a chmod has now revoked permission.
686 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
690 dentry = fhp->fh_dentry;
691 inode = dentry->d_inode;
693 /* Disallow write access to files with the append-only bit set
694 * or any access when mandatory locking enabled
697 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
700 * We must ignore files (but only files) which might have mandatory
701 * locks on them because there is no way to know if the accesser has
704 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
711 * Check to see if there are any leases on this file.
712 * This may block while leases are broken.
714 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
715 if (host_err == -EWOULDBLOCK)
716 host_err = -ETIMEDOUT;
717 if (host_err) /* NOMEM or WOULDBLOCK */
720 if (access & NFSD_MAY_WRITE) {
721 if (access & NFSD_MAY_READ)
722 flags = O_RDWR|O_LARGEFILE;
724 flags = O_WRONLY|O_LARGEFILE;
728 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
731 host_err = PTR_ERR(*filp);
733 err = nfserrno(host_err);
742 nfsd_close(struct file *filp)
749 * As this calls fsync (not fdatasync) there is no need for a write_inode
752 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
753 const struct file_operations *fop)
755 struct inode *inode = dp->d_inode;
756 int (*fsync) (struct file *, struct dentry *, int);
759 err = filemap_fdatawrite(inode->i_mapping);
760 if (err == 0 && fop && (fsync = fop->fsync))
761 err = fsync(filp, dp, 0);
763 err = filemap_fdatawait(inode->i_mapping);
769 nfsd_sync(struct file *filp)
772 struct inode *inode = filp->f_path.dentry->d_inode;
773 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
774 mutex_lock(&inode->i_mutex);
775 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
776 mutex_unlock(&inode->i_mutex);
782 nfsd_sync_dir(struct dentry *dp)
784 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
788 * Obtain the readahead parameters for the file
789 * specified by (dev, ino).
792 static inline struct raparms *
793 nfsd_get_raparms(dev_t dev, ino_t ino)
795 struct raparms *ra, **rap, **frap = NULL;
798 struct raparm_hbucket *rab;
800 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
801 rab = &raparm_hash[hash];
803 spin_lock(&rab->pb_lock);
804 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
805 if (ra->p_ino == ino && ra->p_dev == dev)
808 if (ra->p_count == 0)
811 depth = nfsdstats.ra_size*11/10;
813 spin_unlock(&rab->pb_lock);
823 if (rap != &rab->pb_head) {
825 ra->p_next = rab->pb_head;
829 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
830 spin_unlock(&rab->pb_lock);
835 * Grab and keep cached pages associated with a file in the svc_rqst
836 * so that they can be passed to the network sendmsg/sendpage routines
837 * directly. They will be released after the sending has completed.
840 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
841 struct splice_desc *sd)
843 struct svc_rqst *rqstp = sd->u.data;
844 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
845 struct page *page = buf->page;
849 ret = buf->ops->confirm(pipe, buf);
855 if (rqstp->rq_res.page_len == 0) {
860 rqstp->rq_res.page_base = buf->offset;
861 rqstp->rq_res.page_len = size;
862 } else if (page != pp[-1]) {
868 rqstp->rq_res.page_len += size;
870 rqstp->rq_res.page_len += size;
875 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
876 struct splice_desc *sd)
878 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
881 static inline int svc_msnfs(struct svc_fh *ffhp)
884 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
891 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
892 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
901 inode = file->f_path.dentry->d_inode;
903 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
906 /* Get readahead parameters */
907 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
910 file->f_ra = ra->p_ra;
912 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
913 struct splice_desc sd = {
920 rqstp->rq_resused = 1;
921 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
925 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
929 /* Write back readahead params */
931 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
932 spin_lock(&rab->pb_lock);
933 ra->p_ra = file->f_ra;
936 spin_unlock(&rab->pb_lock);
940 nfsdstats.io_read += host_err;
943 fsnotify_access(file->f_path.dentry);
945 err = nfserrno(host_err);
950 static void kill_suid(struct dentry *dentry)
953 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
955 mutex_lock(&dentry->d_inode->i_mutex);
956 notify_change(dentry, &ia);
957 mutex_unlock(&dentry->d_inode->i_mutex);
961 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
962 loff_t offset, struct kvec *vec, int vlen,
963 unsigned long cnt, int *stablep)
965 struct svc_export *exp;
966 struct dentry *dentry;
971 int stable = *stablep;
976 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
977 (!lock_may_write(file->f_path.dentry->d_inode, offset, cnt)))
981 dentry = file->f_path.dentry;
982 inode = dentry->d_inode;
983 exp = fhp->fh_export;
986 * Request sync writes if
987 * - the sync export option has been set, or
988 * - the client requested O_SYNC behavior (NFSv3 feature).
989 * - The file system doesn't support fsync().
990 * When gathered writes have been configured for this volume,
991 * flushing the data to disk is handled separately below.
994 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
996 *stablep = 2; /* FILE_SYNC */
1001 if (stable && !EX_WGATHER(exp))
1002 file->f_flags |= O_SYNC;
1004 /* Write the data. */
1005 oldfs = get_fs(); set_fs(KERNEL_DS);
1006 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1008 if (host_err >= 0) {
1009 nfsdstats.io_write += cnt;
1010 fsnotify_modify(file->f_path.dentry);
1013 /* clear setuid/setgid flag after write */
1014 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1017 if (host_err >= 0 && stable) {
1018 static ino_t last_ino;
1019 static dev_t last_dev;
1022 * Gathered writes: If another process is currently
1023 * writing to the file, there's a high chance
1024 * this is another nfsd (triggered by a bulk write
1025 * from a client's biod). Rather than syncing the
1026 * file with each write request, we sleep for 10 msec.
1028 * I don't know if this roughly approximates
1029 * C. Juszak's idea of gathered writes, but it's a
1030 * nice and simple solution (IMHO), and it seems to
1033 if (EX_WGATHER(exp)) {
1034 if (atomic_read(&inode->i_writecount) > 1
1035 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1036 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1038 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1041 if (inode->i_state & I_DIRTY) {
1042 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1043 host_err=nfsd_sync(file);
1046 wake_up(&inode->i_wait);
1049 last_ino = inode->i_ino;
1050 last_dev = inode->i_sb->s_dev;
1053 dprintk("nfsd: write complete host_err=%d\n", host_err);
1057 err = nfserrno(host_err);
1063 * Read data from a file. count must contain the requested read count
1064 * on entry. On return, *count contains the number of bytes actually read.
1065 * N.B. After this call fhp needs an fh_put
1068 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1069 loff_t offset, struct kvec *vec, int vlen,
1070 unsigned long *count)
1075 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1076 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1079 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1081 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1084 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1092 * Write data to a file.
1093 * The stable flag requests synchronous writes.
1094 * N.B. After this call fhp needs an fh_put
1097 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1098 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1104 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1105 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1108 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1111 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1116 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1124 #ifdef CONFIG_NFSD_V3
1126 * Commit all pending writes to stable storage.
1127 * Strictly speaking, we could sync just the indicated file region here,
1128 * but there's currently no way we can ask the VFS to do so.
1130 * Unfortunately we cannot lock the file to make sure we return full WCC
1131 * data to the client, as locking happens lower down in the filesystem.
1134 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1135 loff_t offset, unsigned long count)
1140 if ((u64)count > ~(u64)offset)
1141 return nfserr_inval;
1143 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1146 if (EX_ISSYNC(fhp->fh_export)) {
1147 if (file->f_op && file->f_op->fsync) {
1148 err = nfserrno(nfsd_sync(file));
1150 err = nfserr_notsupp;
1157 #endif /* CONFIG_NFSD_V3 */
1160 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1164 * Mode has already been set earlier in create:
1166 iap->ia_valid &= ~ATTR_MODE;
1168 * Setting uid/gid works only for root. Irix appears to
1169 * send along the gid on create when it tries to implement
1170 * setgid directories via NFS:
1172 if (current_fsuid() != 0)
1173 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1175 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1180 * Create a file (regular, directory, device, fifo); UNIX sockets
1181 * not yet implemented.
1182 * If the response fh has been verified, the parent directory should
1183 * already be locked. Note that the parent directory is left locked.
1185 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1188 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1189 char *fname, int flen, struct iattr *iap,
1190 int type, dev_t rdev, struct svc_fh *resfhp)
1192 struct dentry *dentry, *dchild = NULL;
1202 if (isdotent(fname, flen))
1205 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1209 dentry = fhp->fh_dentry;
1210 dirp = dentry->d_inode;
1212 err = nfserr_notdir;
1213 if (!dirp->i_op->lookup)
1216 * Check whether the response file handle has been verified yet.
1217 * If it has, the parent directory should already be locked.
1219 if (!resfhp->fh_dentry) {
1220 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1221 fh_lock_nested(fhp, I_MUTEX_PARENT);
1222 dchild = lookup_one_len(fname, dentry, flen);
1223 host_err = PTR_ERR(dchild);
1226 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1230 /* called from nfsd_proc_create */
1231 dchild = dget(resfhp->fh_dentry);
1232 if (!fhp->fh_locked) {
1233 /* not actually possible */
1235 "nfsd_create: parent %s/%s not locked!\n",
1236 dentry->d_parent->d_name.name,
1237 dentry->d_name.name);
1243 * Make sure the child dentry is still negative ...
1246 if (dchild->d_inode) {
1247 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1248 dentry->d_name.name, dchild->d_name.name);
1252 if (!(iap->ia_valid & ATTR_MODE))
1254 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1257 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1258 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1263 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1268 * Get the dir op function pointer.
1273 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1276 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1282 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1286 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1290 if (EX_ISSYNC(fhp->fh_export)) {
1291 err = nfserrno(nfsd_sync_dir(dentry));
1292 write_inode_now(dchild->d_inode, 1);
1295 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1298 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1300 * Update the file handle to get the new inode info.
1303 err = fh_update(resfhp);
1305 if (dchild && !IS_ERR(dchild))
1310 err = nfserrno(host_err);
1314 #ifdef CONFIG_NFSD_V3
1316 * NFSv3 version of nfsd_create
1319 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1320 char *fname, int flen, struct iattr *iap,
1321 struct svc_fh *resfhp, int createmode, u32 *verifier,
1322 int *truncp, int *created)
1324 struct dentry *dentry, *dchild = NULL;
1329 __u32 v_mtime=0, v_atime=0;
1335 if (isdotent(fname, flen))
1337 if (!(iap->ia_valid & ATTR_MODE))
1339 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1343 dentry = fhp->fh_dentry;
1344 dirp = dentry->d_inode;
1346 /* Get all the sanity checks out of the way before
1347 * we lock the parent. */
1348 err = nfserr_notdir;
1349 if (!dirp->i_op->lookup)
1351 fh_lock_nested(fhp, I_MUTEX_PARENT);
1354 * Compose the response file handle.
1356 dchild = lookup_one_len(fname, dentry, flen);
1357 host_err = PTR_ERR(dchild);
1361 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1365 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1366 /* solaris7 gets confused (bugid 4218508) if these have
1367 * the high bit set, so just clear the high bits. If this is
1368 * ever changed to use different attrs for storing the
1369 * verifier, then do_open_lookup() will also need to be fixed
1372 v_mtime = verifier[0]&0x7fffffff;
1373 v_atime = verifier[1]&0x7fffffff;
1376 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1379 if (dchild->d_inode) {
1382 switch (createmode) {
1383 case NFS3_CREATE_UNCHECKED:
1384 if (! S_ISREG(dchild->d_inode->i_mode))
1387 /* in nfsv4, we need to treat this case a little
1388 * differently. we don't want to truncate the
1389 * file now; this would be wrong if the OPEN
1390 * fails for some other reason. furthermore,
1391 * if the size is nonzero, we should ignore it
1392 * according to spec!
1394 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1397 iap->ia_valid &= ATTR_SIZE;
1401 case NFS3_CREATE_EXCLUSIVE:
1402 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1403 && dchild->d_inode->i_atime.tv_sec == v_atime
1404 && dchild->d_inode->i_size == 0 )
1407 case NFS3_CREATE_GUARDED:
1410 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1414 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1416 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1422 if (EX_ISSYNC(fhp->fh_export)) {
1423 err = nfserrno(nfsd_sync_dir(dentry));
1424 /* setattr will sync the child (or not) */
1427 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1428 /* Cram the verifier into atime/mtime */
1429 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1430 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1431 /* XXX someone who knows this better please fix it for nsec */
1432 iap->ia_mtime.tv_sec = v_mtime;
1433 iap->ia_atime.tv_sec = v_atime;
1434 iap->ia_mtime.tv_nsec = 0;
1435 iap->ia_atime.tv_nsec = 0;
1439 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1443 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1445 * Update the filehandle to get the new inode info.
1448 err = fh_update(resfhp);
1452 if (dchild && !IS_ERR(dchild))
1457 err = nfserrno(host_err);
1460 #endif /* CONFIG_NFSD_V3 */
1463 * Read a symlink. On entry, *lenp must contain the maximum path length that
1464 * fits into the buffer. On return, it contains the true length.
1465 * N.B. After this call fhp needs an fh_put
1468 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1470 struct dentry *dentry;
1471 struct inode *inode;
1476 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1480 dentry = fhp->fh_dentry;
1481 inode = dentry->d_inode;
1484 if (!inode->i_op->readlink)
1487 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1488 /* N.B. Why does this call need a get_fs()??
1489 * Remove the set_fs and watch the fireworks:-) --okir
1492 oldfs = get_fs(); set_fs(KERNEL_DS);
1493 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1504 err = nfserrno(host_err);
1509 * Create a symlink and look up its inode
1510 * N.B. After this call _both_ fhp and resfhp need an fh_put
1513 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1514 char *fname, int flen,
1515 char *path, int plen,
1516 struct svc_fh *resfhp,
1519 struct dentry *dentry, *dnew;
1527 if (isdotent(fname, flen))
1530 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1534 dentry = fhp->fh_dentry;
1535 dnew = lookup_one_len(fname, dentry, flen);
1536 host_err = PTR_ERR(dnew);
1540 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1544 if (unlikely(path[plen] != 0)) {
1545 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1546 if (path_alloced == NULL)
1549 strncpy(path_alloced, path, plen);
1550 path_alloced[plen] = 0;
1551 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1552 kfree(path_alloced);
1555 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1558 if (EX_ISSYNC(fhp->fh_export))
1559 host_err = nfsd_sync_dir(dentry);
1561 err = nfserrno(host_err);
1564 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1566 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1568 if (err==0) err = cerr;
1573 err = nfserrno(host_err);
1579 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1582 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1583 char *name, int len, struct svc_fh *tfhp)
1585 struct dentry *ddir, *dnew, *dold;
1586 struct inode *dirp, *dest;
1590 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1593 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1601 if (isdotent(name, len))
1604 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1605 ddir = ffhp->fh_dentry;
1606 dirp = ddir->d_inode;
1608 dnew = lookup_one_len(name, ddir, len);
1609 host_err = PTR_ERR(dnew);
1613 dold = tfhp->fh_dentry;
1614 dest = dold->d_inode;
1616 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1618 err = nfserrno(host_err);
1621 host_err = vfs_link(dold, dirp, dnew);
1623 if (EX_ISSYNC(ffhp->fh_export)) {
1624 err = nfserrno(nfsd_sync_dir(ddir));
1625 write_inode_now(dest, 1);
1629 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1632 err = nfserrno(host_err);
1634 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1643 err = nfserrno(host_err);
1649 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1652 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1653 struct svc_fh *tfhp, char *tname, int tlen)
1655 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1656 struct inode *fdir, *tdir;
1660 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1663 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1667 fdentry = ffhp->fh_dentry;
1668 fdir = fdentry->d_inode;
1670 tdentry = tfhp->fh_dentry;
1671 tdir = tdentry->d_inode;
1673 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1674 if (ffhp->fh_export != tfhp->fh_export)
1678 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1681 /* cannot use fh_lock as we need deadlock protective ordering
1682 * so do it by hand */
1683 trap = lock_rename(tdentry, fdentry);
1684 ffhp->fh_locked = tfhp->fh_locked = 1;
1688 odentry = lookup_one_len(fname, fdentry, flen);
1689 host_err = PTR_ERR(odentry);
1690 if (IS_ERR(odentry))
1694 if (!odentry->d_inode)
1697 if (odentry == trap)
1700 ndentry = lookup_one_len(tname, tdentry, tlen);
1701 host_err = PTR_ERR(ndentry);
1702 if (IS_ERR(ndentry))
1704 host_err = -ENOTEMPTY;
1705 if (ndentry == trap)
1708 if (svc_msnfs(ffhp) &&
1709 ((atomic_read(&odentry->d_count) > 1)
1710 || (atomic_read(&ndentry->d_count) > 1))) {
1716 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1718 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1722 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1723 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1724 host_err = nfsd_sync_dir(tdentry);
1726 host_err = nfsd_sync_dir(fdentry);
1729 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1736 err = nfserrno(host_err);
1738 /* we cannot reply on fh_unlock on the two filehandles,
1739 * as that would do the wrong thing if the two directories
1740 * were the same, so again we do it by hand
1742 fill_post_wcc(ffhp);
1743 fill_post_wcc(tfhp);
1744 unlock_rename(tdentry, fdentry);
1745 ffhp->fh_locked = tfhp->fh_locked = 0;
1752 * Unlink a file or directory
1753 * N.B. After this call fhp needs an fh_put
1756 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1757 char *fname, int flen)
1759 struct dentry *dentry, *rdentry;
1765 if (!flen || isdotent(fname, flen))
1767 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1771 fh_lock_nested(fhp, I_MUTEX_PARENT);
1772 dentry = fhp->fh_dentry;
1773 dirp = dentry->d_inode;
1775 rdentry = lookup_one_len(fname, dentry, flen);
1776 host_err = PTR_ERR(rdentry);
1777 if (IS_ERR(rdentry))
1780 if (!rdentry->d_inode) {
1787 type = rdentry->d_inode->i_mode & S_IFMT;
1789 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1793 if (type != S_IFDIR) { /* It's UNLINK */
1795 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1796 (atomic_read(&rdentry->d_count) > 1)) {
1800 host_err = vfs_unlink(dirp, rdentry);
1801 } else { /* It's RMDIR */
1802 host_err = vfs_rmdir(dirp, rdentry);
1809 if (EX_ISSYNC(fhp->fh_export))
1810 host_err = nfsd_sync_dir(dentry);
1813 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1815 err = nfserrno(host_err);
1821 * We do this buffering because we must not call back into the file
1822 * system's ->lookup() method from the filldir callback. That may well
1823 * deadlock a number of file systems.
1825 * This is based heavily on the implementation of same in XFS.
1827 struct buffered_dirent {
1831 unsigned int d_type;
1835 struct readdir_data {
1841 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1842 loff_t offset, u64 ino, unsigned int d_type)
1844 struct readdir_data *buf = __buf;
1845 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1846 unsigned int reclen;
1848 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1849 if (buf->used + reclen > PAGE_SIZE) {
1854 de->namlen = namlen;
1855 de->offset = offset;
1857 de->d_type = d_type;
1858 memcpy(de->name, name, namlen);
1859 buf->used += reclen;
1864 static int nfsd_buffered_readdir(struct file *file, filldir_t func,
1865 struct readdir_cd *cdp, loff_t *offsetp)
1867 struct readdir_data buf;
1868 struct buffered_dirent *de;
1873 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1880 unsigned int reclen;
1882 cdp->err = nfserr_eof; /* will be cleared on successful read */
1886 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1898 de = (struct buffered_dirent *)buf.dirent;
1900 offset = de->offset;
1902 if (func(cdp, de->name, de->namlen, de->offset,
1903 de->ino, de->d_type))
1906 if (cdp->err != nfs_ok)
1909 reclen = ALIGN(sizeof(*de) + de->namlen,
1912 de = (struct buffered_dirent *)((char *)de + reclen);
1914 offset = vfs_llseek(file, 0, SEEK_CUR);
1918 free_page((unsigned long)(buf.dirent));
1921 return nfserrno(host_err);
1928 * Read entries from a directory.
1929 * The NFSv3/4 verifier we ignore for now.
1932 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1933 struct readdir_cd *cdp, filldir_t func)
1937 loff_t offset = *offsetp;
1939 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1943 offset = vfs_llseek(file, offset, 0);
1945 err = nfserrno((int)offset);
1949 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1951 if (err == nfserr_eof || err == nfserr_toosmall)
1952 err = nfs_ok; /* can still be found in ->err */
1960 * Get file system stats
1961 * N.B. After this call fhp needs an fh_put
1964 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1966 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1967 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1972 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1974 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1978 * Check for a user's access permissions to this inode.
1981 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1982 struct dentry *dentry, int acc)
1984 struct inode *inode = dentry->d_inode;
1987 if (acc == NFSD_MAY_NOP)
1990 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1992 (acc & NFSD_MAY_READ)? " read" : "",
1993 (acc & NFSD_MAY_WRITE)? " write" : "",
1994 (acc & NFSD_MAY_EXEC)? " exec" : "",
1995 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1996 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1997 (acc & NFSD_MAY_LOCK)? " lock" : "",
1998 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2000 IS_IMMUTABLE(inode)? " immut" : "",
2001 IS_APPEND(inode)? " append" : "",
2002 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2003 dprintk(" owner %d/%d user %d/%d\n",
2004 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2007 /* Normally we reject any write/sattr etc access on a read-only file
2008 * system. But if it is IRIX doing check on write-access for a
2009 * device special file, we ignore rofs.
2011 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2012 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2013 if (exp_rdonly(rqstp, exp) ||
2014 __mnt_is_readonly(exp->ex_path.mnt))
2016 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2019 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2022 if (acc & NFSD_MAY_LOCK) {
2023 /* If we cannot rely on authentication in NLM requests,
2024 * just allow locks, otherwise require read permission, or
2027 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2030 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2033 * The file owner always gets access permission for accesses that
2034 * would normally be checked at open time. This is to make
2035 * file access work even when the client has done a fchmod(fd, 0).
2037 * However, `cp foo bar' should fail nevertheless when bar is
2038 * readonly. A sensible way to do this might be to reject all
2039 * attempts to truncate a read-only file, because a creat() call
2040 * always implies file truncation.
2041 * ... but this isn't really fair. A process may reasonably call
2042 * ftruncate on an open file descriptor on a file with perm 000.
2043 * We must trust the client to do permission checking - using "ACCESS"
2046 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2047 inode->i_uid == current_fsuid())
2050 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2051 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2053 /* Allow read access to binaries even when mode 111 */
2054 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2055 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2056 err = inode_permission(inode, MAY_EXEC);
2058 return err? nfserrno(err) : 0;
2062 nfsd_racache_shutdown(void)
2064 struct raparms *raparm, *last_raparm;
2067 dprintk("nfsd: freeing readahead buffers.\n");
2069 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2070 raparm = raparm_hash[i].pb_head;
2072 last_raparm = raparm;
2073 raparm = raparm->p_next;
2076 raparm_hash[i].pb_head = NULL;
2080 * Initialize readahead param cache
2083 nfsd_racache_init(int cache_size)
2088 struct raparms **raparm = NULL;
2091 if (raparm_hash[0].pb_head)
2093 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2096 cache_size = nperbucket * RAPARM_HASH_SIZE;
2098 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2100 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2101 spin_lock_init(&raparm_hash[i].pb_lock);
2103 raparm = &raparm_hash[i].pb_head;
2104 for (j = 0; j < nperbucket; j++) {
2105 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2108 raparm = &(*raparm)->p_next;
2113 nfsdstats.ra_size = cache_size;
2117 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2118 nfsd_racache_shutdown();
2122 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2124 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2126 struct inode *inode = fhp->fh_dentry->d_inode;
2130 struct posix_acl *acl;
2132 if (!IS_POSIXACL(inode))
2133 return ERR_PTR(-EOPNOTSUPP);
2136 case ACL_TYPE_ACCESS:
2137 name = POSIX_ACL_XATTR_ACCESS;
2139 case ACL_TYPE_DEFAULT:
2140 name = POSIX_ACL_XATTR_DEFAULT;
2143 return ERR_PTR(-EOPNOTSUPP);
2146 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2148 return ERR_PTR(size);
2150 acl = posix_acl_from_xattr(value, size);
2156 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2158 struct inode *inode = fhp->fh_dentry->d_inode;
2164 if (!IS_POSIXACL(inode) ||
2165 !inode->i_op->setxattr || !inode->i_op->removexattr)
2168 case ACL_TYPE_ACCESS:
2169 name = POSIX_ACL_XATTR_ACCESS;
2171 case ACL_TYPE_DEFAULT:
2172 name = POSIX_ACL_XATTR_DEFAULT;
2178 if (acl && acl->a_count) {
2179 size = posix_acl_xattr_size(acl->a_count);
2180 value = kmalloc(size, GFP_KERNEL);
2183 error = posix_acl_to_xattr(acl, value, size);
2190 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2194 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2196 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2199 error = vfs_removexattr(fhp->fh_dentry, name);
2200 if (error == -ENODATA)
2204 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2210 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */