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>
58 #include <linux/ima.h>
61 #include <asm/uaccess.h>
63 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
67 * This is a cache of readahead params that help us choose the proper
68 * readahead strategy. Initially, we set all readahead parameters to 0
69 * and let the VFS handle things.
70 * If you increase the number of cached files very much, you'll need to
71 * add a hash table here.
74 struct raparms *p_next;
79 struct file_ra_state p_ra;
80 unsigned int p_hindex;
83 struct raparm_hbucket {
84 struct raparms *pb_head;
86 } ____cacheline_aligned_in_smp;
88 #define RAPARM_HASH_BITS 4
89 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
90 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
91 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
94 nfsd_v4client(struct svc_rqst *rq)
96 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4;
100 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
102 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
103 * or nfs_ok having possibly changed *dpp and *expp
106 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
107 struct svc_export **expp)
109 struct svc_export *exp = *expp, *exp2 = NULL;
110 struct dentry *dentry = *dpp;
111 struct path path = {.mnt = mntget(exp->ex_path.mnt),
112 .dentry = dget(dentry)};
115 while (d_mountpoint(path.dentry) && follow_down(&path))
118 exp2 = rqst_exp_get_by_name(rqstp, &path);
120 if (PTR_ERR(exp2) != -ENOENT)
125 if (nfsd_v4client(rqstp) ||
126 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
127 /* successfully crossed mount point */
129 * This is subtle: path.dentry is *not* on path.mnt
130 * at this point. The only reason we are safe is that
131 * original mnt is pinned down by exp, so we should
132 * put path *before* putting exp
135 path.dentry = dentry;
145 static void follow_to_parent(struct path *path)
149 while (path->dentry == path->mnt->mnt_root && follow_up(path))
151 dp = dget_parent(path->dentry);
156 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
158 struct svc_export *exp2;
159 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
160 .dentry = dget(dparent)};
162 follow_to_parent(&path);
164 exp2 = rqst_exp_parent(rqstp, &path);
165 if (PTR_ERR(exp2) == -ENOENT) {
166 *dentryp = dget(dparent);
167 } else if (IS_ERR(exp2)) {
169 return PTR_ERR(exp2);
171 *dentryp = dget(path.dentry);
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
190 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
192 /* Obtain dentry and export. */
193 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
197 dparent = fhp->fh_dentry;
198 exp = fhp->fh_export;
201 /* Lookup the name, but don't follow links */
202 if (isdotent(name, len)) {
204 dentry = dget(dparent);
205 else if (dparent != exp->ex_path.dentry)
206 dentry = dget_parent(dparent);
207 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
208 dentry = dget(dparent); /* .. == . just like at / */
210 /* checking mountpoint crossing is very different when stepping up */
211 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
217 dentry = lookup_one_len(name, dparent, len);
218 host_err = PTR_ERR(dentry);
222 * check if we have crossed a mount point ...
224 if (d_mountpoint(dentry)) {
225 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
231 *dentry_ret = dentry;
237 return nfserrno(host_err);
241 * Look up one component of a pathname.
242 * N.B. After this call _both_ fhp and resfh need an fh_put
244 * If the lookup would cross a mountpoint, and the mounted filesystem
245 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
246 * accepted as it stands and the mounted directory is
247 * returned. Otherwise the covered directory is returned.
248 * NOTE: this mountpoint crossing is not supported properly by all
249 * clients and is explicitly disallowed for NFSv3
250 * NeilBrown <neilb@cse.unsw.edu.au>
253 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
254 unsigned int len, struct svc_fh *resfh)
256 struct svc_export *exp;
257 struct dentry *dentry;
260 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
263 err = check_nfsd_access(exp, rqstp);
267 * Note: we compose the file handle now, but as the
268 * dentry may be negative, it may need to be updated.
270 err = fh_compose(resfh, exp, dentry, fhp);
271 if (!err && !dentry->d_inode)
281 * Set various file attributes.
282 * N.B. After this call fhp needs an fh_put
285 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
286 int check_guard, time_t guardtime)
288 struct dentry *dentry;
290 int accmode = NFSD_MAY_SATTR;
296 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
297 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
298 if (iap->ia_valid & ATTR_SIZE)
302 err = fh_verify(rqstp, fhp, ftype, accmode);
306 dentry = fhp->fh_dentry;
307 inode = dentry->d_inode;
309 /* Ignore any mode updates on symlinks */
310 if (S_ISLNK(inode->i_mode))
311 iap->ia_valid &= ~ATTR_MODE;
317 * NFSv2 does not differentiate between "set-[ac]time-to-now"
318 * which only requires access, and "set-[ac]time-to-X" which
319 * requires ownership.
320 * So if it looks like it might be "set both to the same time which
321 * is close to now", and if inode_change_ok fails, then we
322 * convert to "set to now" instead of "set to explicit time"
324 * We only call inode_change_ok as the last test as technically
325 * it is not an interface that we should be using. It is only
326 * valid if the filesystem does not define it's own i_op->setattr.
328 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
329 #define MAX_TOUCH_TIME_ERROR (30*60)
330 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
331 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
335 * Now just make sure time is in the right ballpark.
336 * Solaris, at least, doesn't seem to care what the time
337 * request is. We require it be within 30 minutes of now.
339 time_t delta = iap->ia_atime.tv_sec - get_seconds();
342 if (delta < MAX_TOUCH_TIME_ERROR &&
343 inode_change_ok(inode, iap) != 0) {
345 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
346 * This will cause notify_change to set these times
349 iap->ia_valid &= ~BOTH_TIME_SET;
354 * The size case is special.
355 * It changes the file as well as the attributes.
357 if (iap->ia_valid & ATTR_SIZE) {
358 if (iap->ia_size < inode->i_size) {
359 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
360 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
366 * If we are changing the size of the file, then
367 * we need to break all leases.
369 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
370 if (host_err == -EWOULDBLOCK)
371 host_err = -ETIMEDOUT;
372 if (host_err) /* ENOMEM or EWOULDBLOCK */
375 host_err = get_write_access(inode);
380 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
382 put_write_access(inode);
388 /* sanitize the mode change */
389 if (iap->ia_valid & ATTR_MODE) {
390 iap->ia_mode &= S_IALLUGO;
391 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
394 /* Revoke setuid/setgid on chown */
395 if (!S_ISDIR(inode->i_mode) &&
396 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
397 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
398 iap->ia_valid |= ATTR_KILL_PRIV;
399 if (iap->ia_valid & ATTR_MODE) {
400 /* we're setting mode too, just clear the s*id bits */
401 iap->ia_mode &= ~S_ISUID;
402 if (iap->ia_mode & S_IXGRP)
403 iap->ia_mode &= ~S_ISGID;
405 /* set ATTR_KILL_* bits and let VFS handle it */
406 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
410 /* Change the attributes. */
412 iap->ia_valid |= ATTR_CTIME;
414 err = nfserr_notsync;
415 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
417 host_err = notify_change(dentry, iap);
418 err = nfserrno(host_err);
422 put_write_access(inode);
424 if (EX_ISSYNC(fhp->fh_export))
425 write_inode_now(inode, 1);
430 err = nfserrno(host_err);
434 #if defined(CONFIG_NFSD_V2_ACL) || \
435 defined(CONFIG_NFSD_V3_ACL) || \
436 defined(CONFIG_NFSD_V4)
437 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
442 buflen = vfs_getxattr(dentry, key, NULL, 0);
446 *buf = kmalloc(buflen, GFP_KERNEL);
450 ret = vfs_getxattr(dentry, key, *buf, buflen);
457 #if defined(CONFIG_NFSD_V4)
459 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
466 buflen = posix_acl_xattr_size(pacl->a_count);
467 buf = kmalloc(buflen, GFP_KERNEL);
472 len = posix_acl_to_xattr(pacl, buf, buflen);
478 error = vfs_setxattr(dentry, key, buf, len, 0);
485 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
486 struct nfs4_acl *acl)
490 struct dentry *dentry;
492 struct posix_acl *pacl = NULL, *dpacl = NULL;
493 unsigned int flags = 0;
496 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
500 dentry = fhp->fh_dentry;
501 inode = dentry->d_inode;
502 if (S_ISDIR(inode->i_mode))
503 flags = NFS4_ACL_DIR;
505 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
506 if (host_error == -EINVAL) {
507 return nfserr_attrnotsupp;
508 } else if (host_error < 0)
511 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
515 if (S_ISDIR(inode->i_mode))
516 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
519 posix_acl_release(pacl);
520 posix_acl_release(dpacl);
522 if (host_error == -EOPNOTSUPP)
523 return nfserr_attrnotsupp;
525 return nfserrno(host_error);
528 static struct posix_acl *
529 _get_posix_acl(struct dentry *dentry, char *key)
532 struct posix_acl *pacl = NULL;
535 buflen = nfsd_getxattr(dentry, key, &buf);
539 return ERR_PTR(buflen);
541 pacl = posix_acl_from_xattr(buf, buflen);
547 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
549 struct inode *inode = dentry->d_inode;
551 struct posix_acl *pacl = NULL, *dpacl = NULL;
552 unsigned int flags = 0;
554 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
555 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
556 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
558 error = PTR_ERR(pacl);
563 if (S_ISDIR(inode->i_mode)) {
564 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
565 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
567 else if (IS_ERR(dpacl)) {
568 error = PTR_ERR(dpacl);
572 flags = NFS4_ACL_DIR;
575 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
577 error = PTR_ERR(*acl);
581 posix_acl_release(pacl);
582 posix_acl_release(dpacl);
586 #endif /* defined(CONFIG_NFS_V4) */
588 #ifdef CONFIG_NFSD_V3
590 * Check server access rights to a file system object
596 static struct accessmap nfs3_regaccess[] = {
597 { NFS3_ACCESS_READ, NFSD_MAY_READ },
598 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
599 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
600 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
605 static struct accessmap nfs3_diraccess[] = {
606 { NFS3_ACCESS_READ, NFSD_MAY_READ },
607 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
608 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
609 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
610 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
615 static struct accessmap nfs3_anyaccess[] = {
616 /* Some clients - Solaris 2.6 at least, make an access call
617 * to the server to check for access for things like /dev/null
618 * (which really, the server doesn't care about). So
619 * We provide simple access checking for them, looking
620 * mainly at mode bits, and we make sure to ignore read-only
623 { NFS3_ACCESS_READ, NFSD_MAY_READ },
624 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
625 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
626 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
632 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
634 struct accessmap *map;
635 struct svc_export *export;
636 struct dentry *dentry;
637 u32 query, result = 0, sresult = 0;
640 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
644 export = fhp->fh_export;
645 dentry = fhp->fh_dentry;
647 if (S_ISREG(dentry->d_inode->i_mode))
648 map = nfs3_regaccess;
649 else if (S_ISDIR(dentry->d_inode->i_mode))
650 map = nfs3_diraccess;
652 map = nfs3_anyaccess;
656 for (; map->access; map++) {
657 if (map->access & query) {
660 sresult |= map->access;
662 err2 = nfsd_permission(rqstp, export, dentry, map->how);
665 result |= map->access;
668 /* the following error codes just mean the access was not allowed,
669 * rather than an error occurred */
673 /* simply don't "or" in the access bit. */
683 *supported = sresult;
688 #endif /* CONFIG_NFSD_V3 */
693 * Open an existing file or directory.
694 * The access argument indicates the type of open (read/write/lock)
695 * N.B. After this call fhp needs an fh_put
698 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
699 int access, struct file **filp)
701 struct dentry *dentry;
703 int flags = O_RDONLY|O_LARGEFILE;
707 validate_process_creds();
710 * If we get here, then the client has already done an "open",
711 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
712 * in case a chmod has now revoked permission.
714 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
718 dentry = fhp->fh_dentry;
719 inode = dentry->d_inode;
721 /* Disallow write access to files with the append-only bit set
722 * or any access when mandatory locking enabled
725 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
728 * We must ignore files (but only files) which might have mandatory
729 * locks on them because there is no way to know if the accesser has
732 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
739 * Check to see if there are any leases on this file.
740 * This may block while leases are broken.
742 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
743 if (host_err == -EWOULDBLOCK)
744 host_err = -ETIMEDOUT;
745 if (host_err) /* NOMEM or WOULDBLOCK */
748 if (access & NFSD_MAY_WRITE) {
749 if (access & NFSD_MAY_READ)
750 flags = O_RDWR|O_LARGEFILE;
752 flags = O_WRONLY|O_LARGEFILE;
756 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
757 flags, current_cred());
759 host_err = PTR_ERR(*filp);
761 ima_counts_get(*filp);
763 err = nfserrno(host_err);
765 validate_process_creds();
773 nfsd_close(struct file *filp)
780 * As this calls fsync (not fdatasync) there is no need for a write_inode
783 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
784 const struct file_operations *fop)
786 struct inode *inode = dp->d_inode;
787 int (*fsync) (struct file *, struct dentry *, int);
790 err = filemap_fdatawrite(inode->i_mapping);
791 if (err == 0 && fop && (fsync = fop->fsync))
792 err = fsync(filp, dp, 0);
794 err = filemap_fdatawait(inode->i_mapping);
800 nfsd_sync(struct file *filp)
803 struct inode *inode = filp->f_path.dentry->d_inode;
804 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
805 mutex_lock(&inode->i_mutex);
806 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
807 mutex_unlock(&inode->i_mutex);
813 nfsd_sync_dir(struct dentry *dp)
815 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
819 * Obtain the readahead parameters for the file
820 * specified by (dev, ino).
823 static inline struct raparms *
824 nfsd_get_raparms(dev_t dev, ino_t ino)
826 struct raparms *ra, **rap, **frap = NULL;
829 struct raparm_hbucket *rab;
831 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
832 rab = &raparm_hash[hash];
834 spin_lock(&rab->pb_lock);
835 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
836 if (ra->p_ino == ino && ra->p_dev == dev)
839 if (ra->p_count == 0)
842 depth = nfsdstats.ra_size*11/10;
844 spin_unlock(&rab->pb_lock);
854 if (rap != &rab->pb_head) {
856 ra->p_next = rab->pb_head;
860 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
861 spin_unlock(&rab->pb_lock);
866 * Grab and keep cached pages associated with a file in the svc_rqst
867 * so that they can be passed to the network sendmsg/sendpage routines
868 * directly. They will be released after the sending has completed.
871 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
872 struct splice_desc *sd)
874 struct svc_rqst *rqstp = sd->u.data;
875 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
876 struct page *page = buf->page;
880 ret = buf->ops->confirm(pipe, buf);
886 if (rqstp->rq_res.page_len == 0) {
891 rqstp->rq_res.page_base = buf->offset;
892 rqstp->rq_res.page_len = size;
893 } else if (page != pp[-1]) {
899 rqstp->rq_res.page_len += size;
901 rqstp->rq_res.page_len += size;
906 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
907 struct splice_desc *sd)
909 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
912 static inline int svc_msnfs(struct svc_fh *ffhp)
915 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
922 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
923 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
932 inode = file->f_path.dentry->d_inode;
934 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
937 /* Get readahead parameters */
938 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
941 file->f_ra = ra->p_ra;
943 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
944 struct splice_desc sd = {
951 rqstp->rq_resused = 1;
952 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
956 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
960 /* Write back readahead params */
962 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
963 spin_lock(&rab->pb_lock);
964 ra->p_ra = file->f_ra;
967 spin_unlock(&rab->pb_lock);
971 nfsdstats.io_read += host_err;
974 fsnotify_access(file->f_path.dentry);
976 err = nfserrno(host_err);
981 static void kill_suid(struct dentry *dentry)
984 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
986 mutex_lock(&dentry->d_inode->i_mutex);
987 notify_change(dentry, &ia);
988 mutex_unlock(&dentry->d_inode->i_mutex);
992 * Gathered writes: If another process is currently writing to the file,
993 * there's a high chance this is another nfsd (triggered by a bulk write
994 * from a client's biod). Rather than syncing the file with each write
995 * request, we sleep for 10 msec.
997 * I don't know if this roughly approximates C. Juszak's idea of
998 * gathered writes, but it's a nice and simple solution (IMHO), and it
1001 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1002 * better tool (separate unstable writes and commits) for solving this
1005 static int wait_for_concurrent_writes(struct file *file)
1007 struct inode *inode = file->f_path.dentry->d_inode;
1008 static ino_t last_ino;
1009 static dev_t last_dev;
1012 if (atomic_read(&inode->i_writecount) > 1
1013 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1014 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1016 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1019 if (inode->i_state & I_DIRTY) {
1020 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1021 err = nfsd_sync(file);
1023 last_ino = inode->i_ino;
1024 last_dev = inode->i_sb->s_dev;
1029 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1030 loff_t offset, struct kvec *vec, int vlen,
1031 unsigned long *cnt, int *stablep)
1033 struct svc_export *exp;
1034 struct dentry *dentry;
1035 struct inode *inode;
1039 int stable = *stablep;
1045 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1046 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1050 dentry = file->f_path.dentry;
1051 inode = dentry->d_inode;
1052 exp = fhp->fh_export;
1055 * Request sync writes if
1056 * - the sync export option has been set, or
1057 * - the client requested O_SYNC behavior (NFSv3 feature).
1058 * - The file system doesn't support fsync().
1059 * When NFSv2 gathered writes have been configured for this volume,
1060 * flushing the data to disk is handled separately below.
1062 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1064 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1066 *stablep = 2; /* FILE_SYNC */
1069 if (!EX_ISSYNC(exp))
1071 if (stable && !use_wgather) {
1072 spin_lock(&file->f_lock);
1073 file->f_flags |= O_SYNC;
1074 spin_unlock(&file->f_lock);
1077 /* Write the data. */
1078 oldfs = get_fs(); set_fs(KERNEL_DS);
1079 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1084 nfsdstats.io_write += host_err;
1085 fsnotify_modify(file->f_path.dentry);
1087 /* clear setuid/setgid flag after write */
1088 if (inode->i_mode & (S_ISUID | S_ISGID))
1091 if (stable && use_wgather)
1092 host_err = wait_for_concurrent_writes(file);
1095 dprintk("nfsd: write complete host_err=%d\n", host_err);
1099 err = nfserrno(host_err);
1105 * Read data from a file. count must contain the requested read count
1106 * on entry. On return, *count contains the number of bytes actually read.
1107 * N.B. After this call fhp needs an fh_put
1110 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1111 loff_t offset, struct kvec *vec, int vlen,
1112 unsigned long *count)
1117 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1118 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1121 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1123 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1126 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1134 * Write data to a file.
1135 * The stable flag requests synchronous writes.
1136 * N.B. After this call fhp needs an fh_put
1139 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1140 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1146 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1147 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1150 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1153 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1158 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1166 #ifdef CONFIG_NFSD_V3
1168 * Commit all pending writes to stable storage.
1169 * Strictly speaking, we could sync just the indicated file region here,
1170 * but there's currently no way we can ask the VFS to do so.
1172 * Unfortunately we cannot lock the file to make sure we return full WCC
1173 * data to the client, as locking happens lower down in the filesystem.
1176 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1177 loff_t offset, unsigned long count)
1182 if ((u64)count > ~(u64)offset)
1183 return nfserr_inval;
1185 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1188 if (EX_ISSYNC(fhp->fh_export)) {
1189 if (file->f_op && file->f_op->fsync) {
1190 err = nfserrno(nfsd_sync(file));
1192 err = nfserr_notsupp;
1199 #endif /* CONFIG_NFSD_V3 */
1202 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1206 * Mode has already been set earlier in create:
1208 iap->ia_valid &= ~ATTR_MODE;
1210 * Setting uid/gid works only for root. Irix appears to
1211 * send along the gid on create when it tries to implement
1212 * setgid directories via NFS:
1214 if (current_fsuid() != 0)
1215 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1217 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1221 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1222 * setting size to 0 may fail for some specific file systems by the permission
1223 * checking which requires WRITE permission but the mode is 000.
1224 * we ignore the resizing(to 0) on the just new created file, since the size is
1225 * 0 after file created.
1227 * call this only after vfs_create() is called.
1230 nfsd_check_ignore_resizing(struct iattr *iap)
1232 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1233 iap->ia_valid &= ~ATTR_SIZE;
1237 * Create a file (regular, directory, device, fifo); UNIX sockets
1238 * not yet implemented.
1239 * If the response fh has been verified, the parent directory should
1240 * already be locked. Note that the parent directory is left locked.
1242 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1245 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1246 char *fname, int flen, struct iattr *iap,
1247 int type, dev_t rdev, struct svc_fh *resfhp)
1249 struct dentry *dentry, *dchild = NULL;
1259 if (isdotent(fname, flen))
1262 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1266 dentry = fhp->fh_dentry;
1267 dirp = dentry->d_inode;
1269 err = nfserr_notdir;
1270 if (!dirp->i_op->lookup)
1273 * Check whether the response file handle has been verified yet.
1274 * If it has, the parent directory should already be locked.
1276 if (!resfhp->fh_dentry) {
1277 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1278 fh_lock_nested(fhp, I_MUTEX_PARENT);
1279 dchild = lookup_one_len(fname, dentry, flen);
1280 host_err = PTR_ERR(dchild);
1283 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1287 /* called from nfsd_proc_create */
1288 dchild = dget(resfhp->fh_dentry);
1289 if (!fhp->fh_locked) {
1290 /* not actually possible */
1292 "nfsd_create: parent %s/%s not locked!\n",
1293 dentry->d_parent->d_name.name,
1294 dentry->d_name.name);
1300 * Make sure the child dentry is still negative ...
1303 if (dchild->d_inode) {
1304 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1305 dentry->d_name.name, dchild->d_name.name);
1309 if (!(iap->ia_valid & ATTR_MODE))
1311 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1314 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1315 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1320 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1325 * Get the dir op function pointer.
1330 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1332 nfsd_check_ignore_resizing(iap);
1335 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1341 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1345 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1349 if (EX_ISSYNC(fhp->fh_export)) {
1350 err = nfserrno(nfsd_sync_dir(dentry));
1351 write_inode_now(dchild->d_inode, 1);
1354 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1357 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1359 * Update the file handle to get the new inode info.
1362 err = fh_update(resfhp);
1364 if (dchild && !IS_ERR(dchild))
1369 err = nfserrno(host_err);
1373 #ifdef CONFIG_NFSD_V3
1375 * NFSv3 version of nfsd_create
1378 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1379 char *fname, int flen, struct iattr *iap,
1380 struct svc_fh *resfhp, int createmode, u32 *verifier,
1381 int *truncp, int *created)
1383 struct dentry *dentry, *dchild = NULL;
1388 __u32 v_mtime=0, v_atime=0;
1394 if (isdotent(fname, flen))
1396 if (!(iap->ia_valid & ATTR_MODE))
1398 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1402 dentry = fhp->fh_dentry;
1403 dirp = dentry->d_inode;
1405 /* Get all the sanity checks out of the way before
1406 * we lock the parent. */
1407 err = nfserr_notdir;
1408 if (!dirp->i_op->lookup)
1410 fh_lock_nested(fhp, I_MUTEX_PARENT);
1413 * Compose the response file handle.
1415 dchild = lookup_one_len(fname, dentry, flen);
1416 host_err = PTR_ERR(dchild);
1420 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1424 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1425 /* solaris7 gets confused (bugid 4218508) if these have
1426 * the high bit set, so just clear the high bits. If this is
1427 * ever changed to use different attrs for storing the
1428 * verifier, then do_open_lookup() will also need to be fixed
1431 v_mtime = verifier[0]&0x7fffffff;
1432 v_atime = verifier[1]&0x7fffffff;
1435 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1438 if (dchild->d_inode) {
1441 switch (createmode) {
1442 case NFS3_CREATE_UNCHECKED:
1443 if (! S_ISREG(dchild->d_inode->i_mode))
1446 /* in nfsv4, we need to treat this case a little
1447 * differently. we don't want to truncate the
1448 * file now; this would be wrong if the OPEN
1449 * fails for some other reason. furthermore,
1450 * if the size is nonzero, we should ignore it
1451 * according to spec!
1453 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1456 iap->ia_valid &= ATTR_SIZE;
1460 case NFS3_CREATE_EXCLUSIVE:
1461 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1462 && dchild->d_inode->i_atime.tv_sec == v_atime
1463 && dchild->d_inode->i_size == 0 )
1466 case NFS3_CREATE_GUARDED:
1469 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1473 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1475 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1481 if (EX_ISSYNC(fhp->fh_export)) {
1482 err = nfserrno(nfsd_sync_dir(dentry));
1483 /* setattr will sync the child (or not) */
1486 nfsd_check_ignore_resizing(iap);
1488 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1489 /* Cram the verifier into atime/mtime */
1490 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1491 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1492 /* XXX someone who knows this better please fix it for nsec */
1493 iap->ia_mtime.tv_sec = v_mtime;
1494 iap->ia_atime.tv_sec = v_atime;
1495 iap->ia_mtime.tv_nsec = 0;
1496 iap->ia_atime.tv_nsec = 0;
1500 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1504 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1506 * Update the filehandle to get the new inode info.
1509 err = fh_update(resfhp);
1513 if (dchild && !IS_ERR(dchild))
1518 err = nfserrno(host_err);
1521 #endif /* CONFIG_NFSD_V3 */
1524 * Read a symlink. On entry, *lenp must contain the maximum path length that
1525 * fits into the buffer. On return, it contains the true length.
1526 * N.B. After this call fhp needs an fh_put
1529 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1531 struct dentry *dentry;
1532 struct inode *inode;
1537 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1541 dentry = fhp->fh_dentry;
1542 inode = dentry->d_inode;
1545 if (!inode->i_op->readlink)
1548 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1549 /* N.B. Why does this call need a get_fs()??
1550 * Remove the set_fs and watch the fireworks:-) --okir
1553 oldfs = get_fs(); set_fs(KERNEL_DS);
1554 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1565 err = nfserrno(host_err);
1570 * Create a symlink and look up its inode
1571 * N.B. After this call _both_ fhp and resfhp need an fh_put
1574 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1575 char *fname, int flen,
1576 char *path, int plen,
1577 struct svc_fh *resfhp,
1580 struct dentry *dentry, *dnew;
1588 if (isdotent(fname, flen))
1591 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1595 dentry = fhp->fh_dentry;
1596 dnew = lookup_one_len(fname, dentry, flen);
1597 host_err = PTR_ERR(dnew);
1601 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1605 if (unlikely(path[plen] != 0)) {
1606 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1607 if (path_alloced == NULL)
1610 strncpy(path_alloced, path, plen);
1611 path_alloced[plen] = 0;
1612 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1613 kfree(path_alloced);
1616 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1619 if (EX_ISSYNC(fhp->fh_export))
1620 host_err = nfsd_sync_dir(dentry);
1622 err = nfserrno(host_err);
1625 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1627 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1629 if (err==0) err = cerr;
1634 err = nfserrno(host_err);
1640 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1643 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1644 char *name, int len, struct svc_fh *tfhp)
1646 struct dentry *ddir, *dnew, *dold;
1647 struct inode *dirp, *dest;
1651 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1654 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1662 if (isdotent(name, len))
1665 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1666 ddir = ffhp->fh_dentry;
1667 dirp = ddir->d_inode;
1669 dnew = lookup_one_len(name, ddir, len);
1670 host_err = PTR_ERR(dnew);
1674 dold = tfhp->fh_dentry;
1675 dest = dold->d_inode;
1677 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1679 err = nfserrno(host_err);
1682 host_err = vfs_link(dold, dirp, dnew);
1684 if (EX_ISSYNC(ffhp->fh_export)) {
1685 err = nfserrno(nfsd_sync_dir(ddir));
1686 write_inode_now(dest, 1);
1690 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1693 err = nfserrno(host_err);
1695 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1704 err = nfserrno(host_err);
1710 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1713 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1714 struct svc_fh *tfhp, char *tname, int tlen)
1716 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1717 struct inode *fdir, *tdir;
1721 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1724 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1728 fdentry = ffhp->fh_dentry;
1729 fdir = fdentry->d_inode;
1731 tdentry = tfhp->fh_dentry;
1732 tdir = tdentry->d_inode;
1734 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1735 if (ffhp->fh_export != tfhp->fh_export)
1739 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1742 /* cannot use fh_lock as we need deadlock protective ordering
1743 * so do it by hand */
1744 trap = lock_rename(tdentry, fdentry);
1745 ffhp->fh_locked = tfhp->fh_locked = 1;
1749 odentry = lookup_one_len(fname, fdentry, flen);
1750 host_err = PTR_ERR(odentry);
1751 if (IS_ERR(odentry))
1755 if (!odentry->d_inode)
1758 if (odentry == trap)
1761 ndentry = lookup_one_len(tname, tdentry, tlen);
1762 host_err = PTR_ERR(ndentry);
1763 if (IS_ERR(ndentry))
1765 host_err = -ENOTEMPTY;
1766 if (ndentry == trap)
1769 if (svc_msnfs(ffhp) &&
1770 ((atomic_read(&odentry->d_count) > 1)
1771 || (atomic_read(&ndentry->d_count) > 1))) {
1777 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1779 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1783 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1784 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1785 host_err = nfsd_sync_dir(tdentry);
1787 host_err = nfsd_sync_dir(fdentry);
1790 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1797 err = nfserrno(host_err);
1799 /* we cannot reply on fh_unlock on the two filehandles,
1800 * as that would do the wrong thing if the two directories
1801 * were the same, so again we do it by hand
1803 fill_post_wcc(ffhp);
1804 fill_post_wcc(tfhp);
1805 unlock_rename(tdentry, fdentry);
1806 ffhp->fh_locked = tfhp->fh_locked = 0;
1813 * Unlink a file or directory
1814 * N.B. After this call fhp needs an fh_put
1817 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1818 char *fname, int flen)
1820 struct dentry *dentry, *rdentry;
1826 if (!flen || isdotent(fname, flen))
1828 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1832 fh_lock_nested(fhp, I_MUTEX_PARENT);
1833 dentry = fhp->fh_dentry;
1834 dirp = dentry->d_inode;
1836 rdentry = lookup_one_len(fname, dentry, flen);
1837 host_err = PTR_ERR(rdentry);
1838 if (IS_ERR(rdentry))
1841 if (!rdentry->d_inode) {
1848 type = rdentry->d_inode->i_mode & S_IFMT;
1850 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1854 if (type != S_IFDIR) { /* It's UNLINK */
1856 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1857 (atomic_read(&rdentry->d_count) > 1)) {
1861 host_err = vfs_unlink(dirp, rdentry);
1862 } else { /* It's RMDIR */
1863 host_err = vfs_rmdir(dirp, rdentry);
1870 if (EX_ISSYNC(fhp->fh_export))
1871 host_err = nfsd_sync_dir(dentry);
1874 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1876 err = nfserrno(host_err);
1882 * We do this buffering because we must not call back into the file
1883 * system's ->lookup() method from the filldir callback. That may well
1884 * deadlock a number of file systems.
1886 * This is based heavily on the implementation of same in XFS.
1888 struct buffered_dirent {
1892 unsigned int d_type;
1896 struct readdir_data {
1902 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1903 loff_t offset, u64 ino, unsigned int d_type)
1905 struct readdir_data *buf = __buf;
1906 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1907 unsigned int reclen;
1909 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1910 if (buf->used + reclen > PAGE_SIZE) {
1915 de->namlen = namlen;
1916 de->offset = offset;
1918 de->d_type = d_type;
1919 memcpy(de->name, name, namlen);
1920 buf->used += reclen;
1925 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1926 struct readdir_cd *cdp, loff_t *offsetp)
1928 struct readdir_data buf;
1929 struct buffered_dirent *de;
1934 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1936 return nfserrno(-ENOMEM);
1941 struct inode *dir_inode = file->f_path.dentry->d_inode;
1942 unsigned int reclen;
1944 cdp->err = nfserr_eof; /* will be cleared on successful read */
1948 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1961 * Various filldir functions may end up calling back into
1962 * lookup_one_len() and the file system's ->lookup() method.
1963 * These expect i_mutex to be held, as it would within readdir.
1965 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1969 de = (struct buffered_dirent *)buf.dirent;
1971 offset = de->offset;
1973 if (func(cdp, de->name, de->namlen, de->offset,
1974 de->ino, de->d_type))
1977 if (cdp->err != nfs_ok)
1980 reclen = ALIGN(sizeof(*de) + de->namlen,
1983 de = (struct buffered_dirent *)((char *)de + reclen);
1985 mutex_unlock(&dir_inode->i_mutex);
1986 if (size > 0) /* We bailed out early */
1989 offset = vfs_llseek(file, 0, SEEK_CUR);
1992 free_page((unsigned long)(buf.dirent));
1995 return nfserrno(host_err);
2002 * Read entries from a directory.
2003 * The NFSv3/4 verifier we ignore for now.
2006 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2007 struct readdir_cd *cdp, filldir_t func)
2011 loff_t offset = *offsetp;
2013 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2017 offset = vfs_llseek(file, offset, 0);
2019 err = nfserrno((int)offset);
2023 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2025 if (err == nfserr_eof || err == nfserr_toosmall)
2026 err = nfs_ok; /* can still be found in ->err */
2034 * Get file system stats
2035 * N.B. After this call fhp needs an fh_put
2038 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2040 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2041 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2046 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2048 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2052 * Check for a user's access permissions to this inode.
2055 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2056 struct dentry *dentry, int acc)
2058 struct inode *inode = dentry->d_inode;
2062 if (acc == NFSD_MAY_NOP)
2065 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2067 (acc & NFSD_MAY_READ)? " read" : "",
2068 (acc & NFSD_MAY_WRITE)? " write" : "",
2069 (acc & NFSD_MAY_EXEC)? " exec" : "",
2070 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2071 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2072 (acc & NFSD_MAY_LOCK)? " lock" : "",
2073 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2075 IS_IMMUTABLE(inode)? " immut" : "",
2076 IS_APPEND(inode)? " append" : "",
2077 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2078 dprintk(" owner %d/%d user %d/%d\n",
2079 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2082 /* Normally we reject any write/sattr etc access on a read-only file
2083 * system. But if it is IRIX doing check on write-access for a
2084 * device special file, we ignore rofs.
2086 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2087 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2088 if (exp_rdonly(rqstp, exp) ||
2089 __mnt_is_readonly(exp->ex_path.mnt))
2091 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2094 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2097 if (acc & NFSD_MAY_LOCK) {
2098 /* If we cannot rely on authentication in NLM requests,
2099 * just allow locks, otherwise require read permission, or
2102 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2105 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2108 * The file owner always gets access permission for accesses that
2109 * would normally be checked at open time. This is to make
2110 * file access work even when the client has done a fchmod(fd, 0).
2112 * However, `cp foo bar' should fail nevertheless when bar is
2113 * readonly. A sensible way to do this might be to reject all
2114 * attempts to truncate a read-only file, because a creat() call
2115 * always implies file truncation.
2116 * ... but this isn't really fair. A process may reasonably call
2117 * ftruncate on an open file descriptor on a file with perm 000.
2118 * We must trust the client to do permission checking - using "ACCESS"
2121 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2122 inode->i_uid == current_fsuid())
2125 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2126 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2128 /* Allow read access to binaries even when mode 111 */
2129 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2130 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2131 err = inode_permission(inode, MAY_EXEC);
2135 /* Do integrity (permission) checking now, but defer incrementing
2136 * IMA counts to the actual file open.
2138 path.mnt = exp->ex_path.mnt;
2139 path.dentry = dentry;
2140 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2143 return err? nfserrno(err) : 0;
2147 nfsd_racache_shutdown(void)
2149 struct raparms *raparm, *last_raparm;
2152 dprintk("nfsd: freeing readahead buffers.\n");
2154 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2155 raparm = raparm_hash[i].pb_head;
2157 last_raparm = raparm;
2158 raparm = raparm->p_next;
2161 raparm_hash[i].pb_head = NULL;
2165 * Initialize readahead param cache
2168 nfsd_racache_init(int cache_size)
2173 struct raparms **raparm = NULL;
2176 if (raparm_hash[0].pb_head)
2178 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2181 cache_size = nperbucket * RAPARM_HASH_SIZE;
2183 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2185 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2186 spin_lock_init(&raparm_hash[i].pb_lock);
2188 raparm = &raparm_hash[i].pb_head;
2189 for (j = 0; j < nperbucket; j++) {
2190 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2193 raparm = &(*raparm)->p_next;
2198 nfsdstats.ra_size = cache_size;
2202 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2203 nfsd_racache_shutdown();
2207 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2209 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2211 struct inode *inode = fhp->fh_dentry->d_inode;
2215 struct posix_acl *acl;
2217 if (!IS_POSIXACL(inode))
2218 return ERR_PTR(-EOPNOTSUPP);
2221 case ACL_TYPE_ACCESS:
2222 name = POSIX_ACL_XATTR_ACCESS;
2224 case ACL_TYPE_DEFAULT:
2225 name = POSIX_ACL_XATTR_DEFAULT;
2228 return ERR_PTR(-EOPNOTSUPP);
2231 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2233 return ERR_PTR(size);
2235 acl = posix_acl_from_xattr(value, size);
2241 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2243 struct inode *inode = fhp->fh_dentry->d_inode;
2249 if (!IS_POSIXACL(inode) ||
2250 !inode->i_op->setxattr || !inode->i_op->removexattr)
2253 case ACL_TYPE_ACCESS:
2254 name = POSIX_ACL_XATTR_ACCESS;
2256 case ACL_TYPE_DEFAULT:
2257 name = POSIX_ACL_XATTR_DEFAULT;
2263 if (acl && acl->a_count) {
2264 size = posix_acl_xattr_size(acl->a_count);
2265 value = kmalloc(size, GFP_KERNEL);
2268 error = posix_acl_to_xattr(acl, value, size);
2275 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2279 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2281 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2284 error = vfs_removexattr(fhp->fh_dentry, name);
2285 if (error == -ENODATA)
2289 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2295 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */