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 (!S_ISDIR(inode->i_mode) &&
370 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
371 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
372 iap->ia_valid |= ATTR_KILL_PRIV;
373 if (iap->ia_valid & ATTR_MODE) {
374 /* we're setting mode too, just clear the s*id bits */
375 iap->ia_mode &= ~S_ISUID;
376 if (iap->ia_mode & S_IXGRP)
377 iap->ia_mode &= ~S_ISGID;
379 /* set ATTR_KILL_* bits and let VFS handle it */
380 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
384 /* Change the attributes. */
386 iap->ia_valid |= ATTR_CTIME;
388 err = nfserr_notsync;
389 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
391 host_err = notify_change(dentry, iap);
392 err = nfserrno(host_err);
396 put_write_access(inode);
398 if (EX_ISSYNC(fhp->fh_export))
399 write_inode_now(inode, 1);
404 err = nfserrno(host_err);
408 #if defined(CONFIG_NFSD_V2_ACL) || \
409 defined(CONFIG_NFSD_V3_ACL) || \
410 defined(CONFIG_NFSD_V4)
411 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
416 buflen = vfs_getxattr(dentry, key, NULL, 0);
420 *buf = kmalloc(buflen, GFP_KERNEL);
424 ret = vfs_getxattr(dentry, key, *buf, buflen);
431 #if defined(CONFIG_NFSD_V4)
433 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
440 buflen = posix_acl_xattr_size(pacl->a_count);
441 buf = kmalloc(buflen, GFP_KERNEL);
446 len = posix_acl_to_xattr(pacl, buf, buflen);
452 error = vfs_setxattr(dentry, key, buf, len, 0);
459 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
460 struct nfs4_acl *acl)
464 struct dentry *dentry;
466 struct posix_acl *pacl = NULL, *dpacl = NULL;
467 unsigned int flags = 0;
470 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
474 dentry = fhp->fh_dentry;
475 inode = dentry->d_inode;
476 if (S_ISDIR(inode->i_mode))
477 flags = NFS4_ACL_DIR;
479 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
480 if (host_error == -EINVAL) {
481 return nfserr_attrnotsupp;
482 } else if (host_error < 0)
485 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
489 if (S_ISDIR(inode->i_mode))
490 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
493 posix_acl_release(pacl);
494 posix_acl_release(dpacl);
496 if (host_error == -EOPNOTSUPP)
497 return nfserr_attrnotsupp;
499 return nfserrno(host_error);
502 static struct posix_acl *
503 _get_posix_acl(struct dentry *dentry, char *key)
506 struct posix_acl *pacl = NULL;
509 buflen = nfsd_getxattr(dentry, key, &buf);
513 return ERR_PTR(buflen);
515 pacl = posix_acl_from_xattr(buf, buflen);
521 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
523 struct inode *inode = dentry->d_inode;
525 struct posix_acl *pacl = NULL, *dpacl = NULL;
526 unsigned int flags = 0;
528 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
529 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
530 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
532 error = PTR_ERR(pacl);
537 if (S_ISDIR(inode->i_mode)) {
538 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
539 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
541 else if (IS_ERR(dpacl)) {
542 error = PTR_ERR(dpacl);
546 flags = NFS4_ACL_DIR;
549 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
551 error = PTR_ERR(*acl);
555 posix_acl_release(pacl);
556 posix_acl_release(dpacl);
560 #endif /* defined(CONFIG_NFS_V4) */
562 #ifdef CONFIG_NFSD_V3
564 * Check server access rights to a file system object
570 static struct accessmap nfs3_regaccess[] = {
571 { NFS3_ACCESS_READ, NFSD_MAY_READ },
572 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
573 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
574 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
579 static struct accessmap nfs3_diraccess[] = {
580 { NFS3_ACCESS_READ, NFSD_MAY_READ },
581 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
582 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
583 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
584 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
589 static struct accessmap nfs3_anyaccess[] = {
590 /* Some clients - Solaris 2.6 at least, make an access call
591 * to the server to check for access for things like /dev/null
592 * (which really, the server doesn't care about). So
593 * We provide simple access checking for them, looking
594 * mainly at mode bits, and we make sure to ignore read-only
597 { NFS3_ACCESS_READ, NFSD_MAY_READ },
598 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
599 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
600 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
606 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
608 struct accessmap *map;
609 struct svc_export *export;
610 struct dentry *dentry;
611 u32 query, result = 0, sresult = 0;
614 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
618 export = fhp->fh_export;
619 dentry = fhp->fh_dentry;
621 if (S_ISREG(dentry->d_inode->i_mode))
622 map = nfs3_regaccess;
623 else if (S_ISDIR(dentry->d_inode->i_mode))
624 map = nfs3_diraccess;
626 map = nfs3_anyaccess;
630 for (; map->access; map++) {
631 if (map->access & query) {
634 sresult |= map->access;
636 err2 = nfsd_permission(rqstp, export, dentry, map->how);
639 result |= map->access;
642 /* the following error codes just mean the access was not allowed,
643 * rather than an error occurred */
647 /* simply don't "or" in the access bit. */
657 *supported = sresult;
662 #endif /* CONFIG_NFSD_V3 */
667 * Open an existing file or directory.
668 * The access argument indicates the type of open (read/write/lock)
669 * N.B. After this call fhp needs an fh_put
672 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
673 int access, struct file **filp)
675 const struct cred *cred = current_cred();
676 struct dentry *dentry;
678 int flags = O_RDONLY|O_LARGEFILE;
683 * If we get here, then the client has already done an "open",
684 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
685 * in case a chmod has now revoked permission.
687 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
691 dentry = fhp->fh_dentry;
692 inode = dentry->d_inode;
694 /* Disallow write access to files with the append-only bit set
695 * or any access when mandatory locking enabled
698 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
701 * We must ignore files (but only files) which might have mandatory
702 * locks on them because there is no way to know if the accesser has
705 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
712 * Check to see if there are any leases on this file.
713 * This may block while leases are broken.
715 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
716 if (host_err == -EWOULDBLOCK)
717 host_err = -ETIMEDOUT;
718 if (host_err) /* NOMEM or WOULDBLOCK */
721 if (access & NFSD_MAY_WRITE) {
722 if (access & NFSD_MAY_READ)
723 flags = O_RDWR|O_LARGEFILE;
725 flags = O_WRONLY|O_LARGEFILE;
729 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
732 host_err = PTR_ERR(*filp);
734 err = nfserrno(host_err);
743 nfsd_close(struct file *filp)
750 * As this calls fsync (not fdatasync) there is no need for a write_inode
753 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
754 const struct file_operations *fop)
756 struct inode *inode = dp->d_inode;
757 int (*fsync) (struct file *, struct dentry *, int);
760 err = filemap_fdatawrite(inode->i_mapping);
761 if (err == 0 && fop && (fsync = fop->fsync))
762 err = fsync(filp, dp, 0);
764 err = filemap_fdatawait(inode->i_mapping);
770 nfsd_sync(struct file *filp)
773 struct inode *inode = filp->f_path.dentry->d_inode;
774 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
775 mutex_lock(&inode->i_mutex);
776 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
777 mutex_unlock(&inode->i_mutex);
783 nfsd_sync_dir(struct dentry *dp)
785 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
789 * Obtain the readahead parameters for the file
790 * specified by (dev, ino).
793 static inline struct raparms *
794 nfsd_get_raparms(dev_t dev, ino_t ino)
796 struct raparms *ra, **rap, **frap = NULL;
799 struct raparm_hbucket *rab;
801 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
802 rab = &raparm_hash[hash];
804 spin_lock(&rab->pb_lock);
805 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
806 if (ra->p_ino == ino && ra->p_dev == dev)
809 if (ra->p_count == 0)
812 depth = nfsdstats.ra_size*11/10;
814 spin_unlock(&rab->pb_lock);
824 if (rap != &rab->pb_head) {
826 ra->p_next = rab->pb_head;
830 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
831 spin_unlock(&rab->pb_lock);
836 * Grab and keep cached pages associated with a file in the svc_rqst
837 * so that they can be passed to the network sendmsg/sendpage routines
838 * directly. They will be released after the sending has completed.
841 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
842 struct splice_desc *sd)
844 struct svc_rqst *rqstp = sd->u.data;
845 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
846 struct page *page = buf->page;
850 ret = buf->ops->confirm(pipe, buf);
856 if (rqstp->rq_res.page_len == 0) {
861 rqstp->rq_res.page_base = buf->offset;
862 rqstp->rq_res.page_len = size;
863 } else if (page != pp[-1]) {
869 rqstp->rq_res.page_len += size;
871 rqstp->rq_res.page_len += size;
876 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
877 struct splice_desc *sd)
879 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
882 static inline int svc_msnfs(struct svc_fh *ffhp)
885 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
892 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
893 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
902 inode = file->f_path.dentry->d_inode;
904 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
907 /* Get readahead parameters */
908 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
911 file->f_ra = ra->p_ra;
913 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
914 struct splice_desc sd = {
921 rqstp->rq_resused = 1;
922 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
926 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
930 /* Write back readahead params */
932 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
933 spin_lock(&rab->pb_lock);
934 ra->p_ra = file->f_ra;
937 spin_unlock(&rab->pb_lock);
941 nfsdstats.io_read += host_err;
944 fsnotify_access(file->f_path.dentry);
946 err = nfserrno(host_err);
951 static void kill_suid(struct dentry *dentry)
954 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
956 mutex_lock(&dentry->d_inode->i_mutex);
957 notify_change(dentry, &ia);
958 mutex_unlock(&dentry->d_inode->i_mutex);
962 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
963 loff_t offset, struct kvec *vec, int vlen,
964 unsigned long *cnt, int *stablep)
966 struct svc_export *exp;
967 struct dentry *dentry;
972 int stable = *stablep;
977 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
978 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
982 dentry = file->f_path.dentry;
983 inode = dentry->d_inode;
984 exp = fhp->fh_export;
987 * Request sync writes if
988 * - the sync export option has been set, or
989 * - the client requested O_SYNC behavior (NFSv3 feature).
990 * - The file system doesn't support fsync().
991 * When gathered writes have been configured for this volume,
992 * flushing the data to disk is handled separately below.
995 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
997 *stablep = 2; /* FILE_SYNC */
1000 if (!EX_ISSYNC(exp))
1002 if (stable && !EX_WGATHER(exp))
1003 file->f_flags |= O_SYNC;
1005 /* Write the data. */
1006 oldfs = get_fs(); set_fs(KERNEL_DS);
1007 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1009 if (host_err >= 0) {
1010 nfsdstats.io_write += host_err;
1011 fsnotify_modify(file->f_path.dentry);
1014 /* clear setuid/setgid flag after write */
1015 if (host_err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
1018 if (host_err >= 0 && stable) {
1019 static ino_t last_ino;
1020 static dev_t last_dev;
1023 * Gathered writes: If another process is currently
1024 * writing to the file, there's a high chance
1025 * this is another nfsd (triggered by a bulk write
1026 * from a client's biod). Rather than syncing the
1027 * file with each write request, we sleep for 10 msec.
1029 * I don't know if this roughly approximates
1030 * C. Juszak's idea of gathered writes, but it's a
1031 * nice and simple solution (IMHO), and it seems to
1034 if (EX_WGATHER(exp)) {
1035 if (atomic_read(&inode->i_writecount) > 1
1036 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1037 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1039 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1042 if (inode->i_state & I_DIRTY) {
1043 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1044 host_err=nfsd_sync(file);
1047 wake_up(&inode->i_wait);
1050 last_ino = inode->i_ino;
1051 last_dev = inode->i_sb->s_dev;
1054 dprintk("nfsd: write complete host_err=%d\n", host_err);
1055 if (host_err >= 0) {
1059 err = nfserrno(host_err);
1065 * Read data from a file. count must contain the requested read count
1066 * on entry. On return, *count contains the number of bytes actually read.
1067 * N.B. After this call fhp needs an fh_put
1070 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1071 loff_t offset, struct kvec *vec, int vlen,
1072 unsigned long *count)
1077 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1078 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1081 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1083 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1086 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1094 * Write data to a file.
1095 * The stable flag requests synchronous writes.
1096 * N.B. After this call fhp needs an fh_put
1099 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1100 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1106 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1107 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1110 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1113 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1118 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1126 #ifdef CONFIG_NFSD_V3
1128 * Commit all pending writes to stable storage.
1129 * Strictly speaking, we could sync just the indicated file region here,
1130 * but there's currently no way we can ask the VFS to do so.
1132 * Unfortunately we cannot lock the file to make sure we return full WCC
1133 * data to the client, as locking happens lower down in the filesystem.
1136 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1137 loff_t offset, unsigned long count)
1142 if ((u64)count > ~(u64)offset)
1143 return nfserr_inval;
1145 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1148 if (EX_ISSYNC(fhp->fh_export)) {
1149 if (file->f_op && file->f_op->fsync) {
1150 err = nfserrno(nfsd_sync(file));
1152 err = nfserr_notsupp;
1159 #endif /* CONFIG_NFSD_V3 */
1162 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1166 * Mode has already been set earlier in create:
1168 iap->ia_valid &= ~ATTR_MODE;
1170 * Setting uid/gid works only for root. Irix appears to
1171 * send along the gid on create when it tries to implement
1172 * setgid directories via NFS:
1174 if (current_fsuid() != 0)
1175 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1177 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1181 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1182 * setting size to 0 may fail for some specific file systems by the permission
1183 * checking which requires WRITE permission but the mode is 000.
1184 * we ignore the resizing(to 0) on the just new created file, since the size is
1185 * 0 after file created.
1187 * call this only after vfs_create() is called.
1190 nfsd_check_ignore_resizing(struct iattr *iap)
1192 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1193 iap->ia_valid &= ~ATTR_SIZE;
1197 * Create a file (regular, directory, device, fifo); UNIX sockets
1198 * not yet implemented.
1199 * If the response fh has been verified, the parent directory should
1200 * already be locked. Note that the parent directory is left locked.
1202 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1205 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1206 char *fname, int flen, struct iattr *iap,
1207 int type, dev_t rdev, struct svc_fh *resfhp)
1209 struct dentry *dentry, *dchild = NULL;
1219 if (isdotent(fname, flen))
1222 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1226 dentry = fhp->fh_dentry;
1227 dirp = dentry->d_inode;
1229 err = nfserr_notdir;
1230 if (!dirp->i_op->lookup)
1233 * Check whether the response file handle has been verified yet.
1234 * If it has, the parent directory should already be locked.
1236 if (!resfhp->fh_dentry) {
1237 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1238 fh_lock_nested(fhp, I_MUTEX_PARENT);
1239 dchild = lookup_one_len(fname, dentry, flen);
1240 host_err = PTR_ERR(dchild);
1243 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1247 /* called from nfsd_proc_create */
1248 dchild = dget(resfhp->fh_dentry);
1249 if (!fhp->fh_locked) {
1250 /* not actually possible */
1252 "nfsd_create: parent %s/%s not locked!\n",
1253 dentry->d_parent->d_name.name,
1254 dentry->d_name.name);
1260 * Make sure the child dentry is still negative ...
1263 if (dchild->d_inode) {
1264 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1265 dentry->d_name.name, dchild->d_name.name);
1269 if (!(iap->ia_valid & ATTR_MODE))
1271 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1274 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1275 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1280 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1285 * Get the dir op function pointer.
1290 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1292 nfsd_check_ignore_resizing(iap);
1295 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1301 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1305 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1309 if (EX_ISSYNC(fhp->fh_export)) {
1310 err = nfserrno(nfsd_sync_dir(dentry));
1311 write_inode_now(dchild->d_inode, 1);
1314 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1317 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1319 * Update the file handle to get the new inode info.
1322 err = fh_update(resfhp);
1324 if (dchild && !IS_ERR(dchild))
1329 err = nfserrno(host_err);
1333 #ifdef CONFIG_NFSD_V3
1335 * NFSv3 version of nfsd_create
1338 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1339 char *fname, int flen, struct iattr *iap,
1340 struct svc_fh *resfhp, int createmode, u32 *verifier,
1341 int *truncp, int *created)
1343 struct dentry *dentry, *dchild = NULL;
1348 __u32 v_mtime=0, v_atime=0;
1354 if (isdotent(fname, flen))
1356 if (!(iap->ia_valid & ATTR_MODE))
1358 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1362 dentry = fhp->fh_dentry;
1363 dirp = dentry->d_inode;
1365 /* Get all the sanity checks out of the way before
1366 * we lock the parent. */
1367 err = nfserr_notdir;
1368 if (!dirp->i_op->lookup)
1370 fh_lock_nested(fhp, I_MUTEX_PARENT);
1373 * Compose the response file handle.
1375 dchild = lookup_one_len(fname, dentry, flen);
1376 host_err = PTR_ERR(dchild);
1380 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1384 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1385 /* solaris7 gets confused (bugid 4218508) if these have
1386 * the high bit set, so just clear the high bits. If this is
1387 * ever changed to use different attrs for storing the
1388 * verifier, then do_open_lookup() will also need to be fixed
1391 v_mtime = verifier[0]&0x7fffffff;
1392 v_atime = verifier[1]&0x7fffffff;
1395 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1398 if (dchild->d_inode) {
1401 switch (createmode) {
1402 case NFS3_CREATE_UNCHECKED:
1403 if (! S_ISREG(dchild->d_inode->i_mode))
1406 /* in nfsv4, we need to treat this case a little
1407 * differently. we don't want to truncate the
1408 * file now; this would be wrong if the OPEN
1409 * fails for some other reason. furthermore,
1410 * if the size is nonzero, we should ignore it
1411 * according to spec!
1413 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1416 iap->ia_valid &= ATTR_SIZE;
1420 case NFS3_CREATE_EXCLUSIVE:
1421 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1422 && dchild->d_inode->i_atime.tv_sec == v_atime
1423 && dchild->d_inode->i_size == 0 )
1426 case NFS3_CREATE_GUARDED:
1429 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1433 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1435 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1441 if (EX_ISSYNC(fhp->fh_export)) {
1442 err = nfserrno(nfsd_sync_dir(dentry));
1443 /* setattr will sync the child (or not) */
1446 nfsd_check_ignore_resizing(iap);
1448 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1449 /* Cram the verifier into atime/mtime */
1450 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1451 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1452 /* XXX someone who knows this better please fix it for nsec */
1453 iap->ia_mtime.tv_sec = v_mtime;
1454 iap->ia_atime.tv_sec = v_atime;
1455 iap->ia_mtime.tv_nsec = 0;
1456 iap->ia_atime.tv_nsec = 0;
1460 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1464 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1466 * Update the filehandle to get the new inode info.
1469 err = fh_update(resfhp);
1473 if (dchild && !IS_ERR(dchild))
1478 err = nfserrno(host_err);
1481 #endif /* CONFIG_NFSD_V3 */
1484 * Read a symlink. On entry, *lenp must contain the maximum path length that
1485 * fits into the buffer. On return, it contains the true length.
1486 * N.B. After this call fhp needs an fh_put
1489 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1491 struct dentry *dentry;
1492 struct inode *inode;
1497 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1501 dentry = fhp->fh_dentry;
1502 inode = dentry->d_inode;
1505 if (!inode->i_op->readlink)
1508 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1509 /* N.B. Why does this call need a get_fs()??
1510 * Remove the set_fs and watch the fireworks:-) --okir
1513 oldfs = get_fs(); set_fs(KERNEL_DS);
1514 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1525 err = nfserrno(host_err);
1530 * Create a symlink and look up its inode
1531 * N.B. After this call _both_ fhp and resfhp need an fh_put
1534 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1535 char *fname, int flen,
1536 char *path, int plen,
1537 struct svc_fh *resfhp,
1540 struct dentry *dentry, *dnew;
1548 if (isdotent(fname, flen))
1551 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1555 dentry = fhp->fh_dentry;
1556 dnew = lookup_one_len(fname, dentry, flen);
1557 host_err = PTR_ERR(dnew);
1561 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1565 if (unlikely(path[plen] != 0)) {
1566 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1567 if (path_alloced == NULL)
1570 strncpy(path_alloced, path, plen);
1571 path_alloced[plen] = 0;
1572 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1573 kfree(path_alloced);
1576 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1579 if (EX_ISSYNC(fhp->fh_export))
1580 host_err = nfsd_sync_dir(dentry);
1582 err = nfserrno(host_err);
1585 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1587 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1589 if (err==0) err = cerr;
1594 err = nfserrno(host_err);
1600 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1603 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1604 char *name, int len, struct svc_fh *tfhp)
1606 struct dentry *ddir, *dnew, *dold;
1607 struct inode *dirp, *dest;
1611 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1614 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1622 if (isdotent(name, len))
1625 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1626 ddir = ffhp->fh_dentry;
1627 dirp = ddir->d_inode;
1629 dnew = lookup_one_len(name, ddir, len);
1630 host_err = PTR_ERR(dnew);
1634 dold = tfhp->fh_dentry;
1635 dest = dold->d_inode;
1637 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1639 err = nfserrno(host_err);
1642 host_err = vfs_link(dold, dirp, dnew);
1644 if (EX_ISSYNC(ffhp->fh_export)) {
1645 err = nfserrno(nfsd_sync_dir(ddir));
1646 write_inode_now(dest, 1);
1650 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1653 err = nfserrno(host_err);
1655 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1664 err = nfserrno(host_err);
1670 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1673 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1674 struct svc_fh *tfhp, char *tname, int tlen)
1676 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1677 struct inode *fdir, *tdir;
1681 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1684 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1688 fdentry = ffhp->fh_dentry;
1689 fdir = fdentry->d_inode;
1691 tdentry = tfhp->fh_dentry;
1692 tdir = tdentry->d_inode;
1694 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1695 if (ffhp->fh_export != tfhp->fh_export)
1699 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1702 /* cannot use fh_lock as we need deadlock protective ordering
1703 * so do it by hand */
1704 trap = lock_rename(tdentry, fdentry);
1705 ffhp->fh_locked = tfhp->fh_locked = 1;
1709 odentry = lookup_one_len(fname, fdentry, flen);
1710 host_err = PTR_ERR(odentry);
1711 if (IS_ERR(odentry))
1715 if (!odentry->d_inode)
1718 if (odentry == trap)
1721 ndentry = lookup_one_len(tname, tdentry, tlen);
1722 host_err = PTR_ERR(ndentry);
1723 if (IS_ERR(ndentry))
1725 host_err = -ENOTEMPTY;
1726 if (ndentry == trap)
1729 if (svc_msnfs(ffhp) &&
1730 ((atomic_read(&odentry->d_count) > 1)
1731 || (atomic_read(&ndentry->d_count) > 1))) {
1737 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1739 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1743 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1744 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1745 host_err = nfsd_sync_dir(tdentry);
1747 host_err = nfsd_sync_dir(fdentry);
1750 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1757 err = nfserrno(host_err);
1759 /* we cannot reply on fh_unlock on the two filehandles,
1760 * as that would do the wrong thing if the two directories
1761 * were the same, so again we do it by hand
1763 fill_post_wcc(ffhp);
1764 fill_post_wcc(tfhp);
1765 unlock_rename(tdentry, fdentry);
1766 ffhp->fh_locked = tfhp->fh_locked = 0;
1773 * Unlink a file or directory
1774 * N.B. After this call fhp needs an fh_put
1777 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1778 char *fname, int flen)
1780 struct dentry *dentry, *rdentry;
1786 if (!flen || isdotent(fname, flen))
1788 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1792 fh_lock_nested(fhp, I_MUTEX_PARENT);
1793 dentry = fhp->fh_dentry;
1794 dirp = dentry->d_inode;
1796 rdentry = lookup_one_len(fname, dentry, flen);
1797 host_err = PTR_ERR(rdentry);
1798 if (IS_ERR(rdentry))
1801 if (!rdentry->d_inode) {
1808 type = rdentry->d_inode->i_mode & S_IFMT;
1810 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1814 if (type != S_IFDIR) { /* It's UNLINK */
1816 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1817 (atomic_read(&rdentry->d_count) > 1)) {
1821 host_err = vfs_unlink(dirp, rdentry);
1822 } else { /* It's RMDIR */
1823 host_err = vfs_rmdir(dirp, rdentry);
1830 if (EX_ISSYNC(fhp->fh_export))
1831 host_err = nfsd_sync_dir(dentry);
1834 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1836 err = nfserrno(host_err);
1842 * We do this buffering because we must not call back into the file
1843 * system's ->lookup() method from the filldir callback. That may well
1844 * deadlock a number of file systems.
1846 * This is based heavily on the implementation of same in XFS.
1848 struct buffered_dirent {
1852 unsigned int d_type;
1856 struct readdir_data {
1862 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1863 loff_t offset, u64 ino, unsigned int d_type)
1865 struct readdir_data *buf = __buf;
1866 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1867 unsigned int reclen;
1869 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1870 if (buf->used + reclen > PAGE_SIZE) {
1875 de->namlen = namlen;
1876 de->offset = offset;
1878 de->d_type = d_type;
1879 memcpy(de->name, name, namlen);
1880 buf->used += reclen;
1885 static int nfsd_buffered_readdir(struct file *file, filldir_t func,
1886 struct readdir_cd *cdp, loff_t *offsetp)
1888 struct readdir_data buf;
1889 struct buffered_dirent *de;
1894 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1901 unsigned int reclen;
1903 cdp->err = nfserr_eof; /* will be cleared on successful read */
1907 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1919 de = (struct buffered_dirent *)buf.dirent;
1921 offset = de->offset;
1923 if (func(cdp, de->name, de->namlen, de->offset,
1924 de->ino, de->d_type))
1927 if (cdp->err != nfs_ok)
1930 reclen = ALIGN(sizeof(*de) + de->namlen,
1933 de = (struct buffered_dirent *)((char *)de + reclen);
1935 offset = vfs_llseek(file, 0, SEEK_CUR);
1939 free_page((unsigned long)(buf.dirent));
1942 return nfserrno(host_err);
1949 * Read entries from a directory.
1950 * The NFSv3/4 verifier we ignore for now.
1953 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1954 struct readdir_cd *cdp, filldir_t func)
1958 loff_t offset = *offsetp;
1960 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
1964 offset = vfs_llseek(file, offset, 0);
1966 err = nfserrno((int)offset);
1970 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1972 if (err == nfserr_eof || err == nfserr_toosmall)
1973 err = nfs_ok; /* can still be found in ->err */
1981 * Get file system stats
1982 * N.B. After this call fhp needs an fh_put
1985 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1987 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1988 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1993 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1995 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1999 * Check for a user's access permissions to this inode.
2002 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2003 struct dentry *dentry, int acc)
2005 struct inode *inode = dentry->d_inode;
2008 if (acc == NFSD_MAY_NOP)
2011 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2013 (acc & NFSD_MAY_READ)? " read" : "",
2014 (acc & NFSD_MAY_WRITE)? " write" : "",
2015 (acc & NFSD_MAY_EXEC)? " exec" : "",
2016 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2017 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2018 (acc & NFSD_MAY_LOCK)? " lock" : "",
2019 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2021 IS_IMMUTABLE(inode)? " immut" : "",
2022 IS_APPEND(inode)? " append" : "",
2023 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2024 dprintk(" owner %d/%d user %d/%d\n",
2025 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2028 /* Normally we reject any write/sattr etc access on a read-only file
2029 * system. But if it is IRIX doing check on write-access for a
2030 * device special file, we ignore rofs.
2032 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2033 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2034 if (exp_rdonly(rqstp, exp) ||
2035 __mnt_is_readonly(exp->ex_path.mnt))
2037 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2040 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2043 if (acc & NFSD_MAY_LOCK) {
2044 /* If we cannot rely on authentication in NLM requests,
2045 * just allow locks, otherwise require read permission, or
2048 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2051 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2054 * The file owner always gets access permission for accesses that
2055 * would normally be checked at open time. This is to make
2056 * file access work even when the client has done a fchmod(fd, 0).
2058 * However, `cp foo bar' should fail nevertheless when bar is
2059 * readonly. A sensible way to do this might be to reject all
2060 * attempts to truncate a read-only file, because a creat() call
2061 * always implies file truncation.
2062 * ... but this isn't really fair. A process may reasonably call
2063 * ftruncate on an open file descriptor on a file with perm 000.
2064 * We must trust the client to do permission checking - using "ACCESS"
2067 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2068 inode->i_uid == current_fsuid())
2071 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2072 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2074 /* Allow read access to binaries even when mode 111 */
2075 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2076 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2077 err = inode_permission(inode, MAY_EXEC);
2079 return err? nfserrno(err) : 0;
2083 nfsd_racache_shutdown(void)
2085 struct raparms *raparm, *last_raparm;
2088 dprintk("nfsd: freeing readahead buffers.\n");
2090 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2091 raparm = raparm_hash[i].pb_head;
2093 last_raparm = raparm;
2094 raparm = raparm->p_next;
2097 raparm_hash[i].pb_head = NULL;
2101 * Initialize readahead param cache
2104 nfsd_racache_init(int cache_size)
2109 struct raparms **raparm = NULL;
2112 if (raparm_hash[0].pb_head)
2114 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2117 cache_size = nperbucket * RAPARM_HASH_SIZE;
2119 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2121 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2122 spin_lock_init(&raparm_hash[i].pb_lock);
2124 raparm = &raparm_hash[i].pb_head;
2125 for (j = 0; j < nperbucket; j++) {
2126 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2129 raparm = &(*raparm)->p_next;
2134 nfsdstats.ra_size = cache_size;
2138 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2139 nfsd_racache_shutdown();
2143 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2145 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2147 struct inode *inode = fhp->fh_dentry->d_inode;
2151 struct posix_acl *acl;
2153 if (!IS_POSIXACL(inode))
2154 return ERR_PTR(-EOPNOTSUPP);
2157 case ACL_TYPE_ACCESS:
2158 name = POSIX_ACL_XATTR_ACCESS;
2160 case ACL_TYPE_DEFAULT:
2161 name = POSIX_ACL_XATTR_DEFAULT;
2164 return ERR_PTR(-EOPNOTSUPP);
2167 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2169 return ERR_PTR(size);
2171 acl = posix_acl_from_xattr(value, size);
2177 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2179 struct inode *inode = fhp->fh_dentry->d_inode;
2185 if (!IS_POSIXACL(inode) ||
2186 !inode->i_op->setxattr || !inode->i_op->removexattr)
2189 case ACL_TYPE_ACCESS:
2190 name = POSIX_ACL_XATTR_ACCESS;
2192 case ACL_TYPE_DEFAULT:
2193 name = POSIX_ACL_XATTR_DEFAULT;
2199 if (acl && acl->a_count) {
2200 size = posix_acl_xattr_size(acl->a_count);
2201 value = kmalloc(size, GFP_KERNEL);
2204 error = posix_acl_to_xattr(acl, value, size);
2211 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2215 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2217 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2220 error = vfs_removexattr(fhp->fh_dentry, name);
2221 if (error == -ENODATA)
2225 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2231 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */