1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/quotaops.h>
24 #include <linux/fsnotify.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/xattr.h>
27 #include <linux/jhash.h>
28 #include <linux/ima.h>
29 #include <asm/uaccess.h>
33 #endif /* CONFIG_NFSD_V3 */
36 #include <linux/nfs4_acl.h>
37 #include <linux/nfsd_idmap.h>
38 #endif /* CONFIG_NFSD_V4 */
43 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
47 * This is a cache of readahead params that help us choose the proper
48 * readahead strategy. Initially, we set all readahead parameters to 0
49 * and let the VFS handle things.
50 * If you increase the number of cached files very much, you'll need to
51 * add a hash table here.
54 struct raparms *p_next;
59 struct file_ra_state p_ra;
60 unsigned int p_hindex;
63 struct raparm_hbucket {
64 struct raparms *pb_head;
66 } ____cacheline_aligned_in_smp;
68 #define RAPARM_HASH_BITS 4
69 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
70 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
71 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
74 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
76 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
77 * or nfs_ok having possibly changed *dpp and *expp
80 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
81 struct svc_export **expp)
83 struct svc_export *exp = *expp, *exp2 = NULL;
84 struct dentry *dentry = *dpp;
85 struct path path = {.mnt = mntget(exp->ex_path.mnt),
86 .dentry = dget(dentry)};
89 while (d_mountpoint(path.dentry) && follow_down(&path))
92 exp2 = rqst_exp_get_by_name(rqstp, &path);
96 * We normally allow NFS clients to continue
97 * "underneath" a mountpoint that is not exported.
98 * The exception is V4ROOT, where no traversal is ever
99 * allowed without an explicit export of the new
102 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
107 if (nfsd_v4client(rqstp) ||
108 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
109 /* successfully crossed mount point */
111 * This is subtle: path.dentry is *not* on path.mnt
112 * at this point. The only reason we are safe is that
113 * original mnt is pinned down by exp, so we should
114 * put path *before* putting exp
117 path.dentry = dentry;
127 static void follow_to_parent(struct path *path)
131 while (path->dentry == path->mnt->mnt_root && follow_up(path))
133 dp = dget_parent(path->dentry);
138 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
140 struct svc_export *exp2;
141 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
142 .dentry = dget(dparent)};
144 follow_to_parent(&path);
146 exp2 = rqst_exp_parent(rqstp, &path);
147 if (PTR_ERR(exp2) == -ENOENT) {
148 *dentryp = dget(dparent);
149 } else if (IS_ERR(exp2)) {
151 return PTR_ERR(exp2);
153 *dentryp = dget(path.dentry);
162 * For nfsd purposes, we treat V4ROOT exports as though there was an
163 * export at *every* directory.
165 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
167 if (d_mountpoint(dentry))
169 if (!(exp->ex_flags & NFSEXP_V4ROOT))
171 return dentry->d_inode != NULL;
175 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
176 const char *name, unsigned int len,
177 struct svc_export **exp_ret, struct dentry **dentry_ret)
179 struct svc_export *exp;
180 struct dentry *dparent;
181 struct dentry *dentry;
185 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
187 /* Obtain dentry and export. */
188 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
192 dparent = fhp->fh_dentry;
193 exp = fhp->fh_export;
196 /* Lookup the name, but don't follow links */
197 if (isdotent(name, len)) {
199 dentry = dget(dparent);
200 else if (dparent != exp->ex_path.dentry)
201 dentry = dget_parent(dparent);
202 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
203 dentry = dget(dparent); /* .. == . just like at / */
205 /* checking mountpoint crossing is very different when stepping up */
206 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
212 dentry = lookup_one_len(name, dparent, len);
213 host_err = PTR_ERR(dentry);
217 * check if we have crossed a mount point ...
219 if (nfsd_mountpoint(dentry, exp)) {
220 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
226 *dentry_ret = dentry;
232 return nfserrno(host_err);
236 * Look up one component of a pathname.
237 * N.B. After this call _both_ fhp and resfh need an fh_put
239 * If the lookup would cross a mountpoint, and the mounted filesystem
240 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
241 * accepted as it stands and the mounted directory is
242 * returned. Otherwise the covered directory is returned.
243 * NOTE: this mountpoint crossing is not supported properly by all
244 * clients and is explicitly disallowed for NFSv3
245 * NeilBrown <neilb@cse.unsw.edu.au>
248 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
249 unsigned int len, struct svc_fh *resfh)
251 struct svc_export *exp;
252 struct dentry *dentry;
255 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
258 err = check_nfsd_access(exp, rqstp);
262 * Note: we compose the file handle now, but as the
263 * dentry may be negative, it may need to be updated.
265 err = fh_compose(resfh, exp, dentry, fhp);
266 if (!err && !dentry->d_inode)
276 * Set various file attributes.
277 * N.B. After this call fhp needs an fh_put
280 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
281 int check_guard, time_t guardtime)
283 struct dentry *dentry;
285 int accmode = NFSD_MAY_SATTR;
291 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
292 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
293 if (iap->ia_valid & ATTR_SIZE)
297 err = fh_verify(rqstp, fhp, ftype, accmode);
301 dentry = fhp->fh_dentry;
302 inode = dentry->d_inode;
304 /* Ignore any mode updates on symlinks */
305 if (S_ISLNK(inode->i_mode))
306 iap->ia_valid &= ~ATTR_MODE;
312 * NFSv2 does not differentiate between "set-[ac]time-to-now"
313 * which only requires access, and "set-[ac]time-to-X" which
314 * requires ownership.
315 * So if it looks like it might be "set both to the same time which
316 * is close to now", and if inode_change_ok fails, then we
317 * convert to "set to now" instead of "set to explicit time"
319 * We only call inode_change_ok as the last test as technically
320 * it is not an interface that we should be using. It is only
321 * valid if the filesystem does not define it's own i_op->setattr.
323 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
324 #define MAX_TOUCH_TIME_ERROR (30*60)
325 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
326 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
330 * Now just make sure time is in the right ballpark.
331 * Solaris, at least, doesn't seem to care what the time
332 * request is. We require it be within 30 minutes of now.
334 time_t delta = iap->ia_atime.tv_sec - get_seconds();
337 if (delta < MAX_TOUCH_TIME_ERROR &&
338 inode_change_ok(inode, iap) != 0) {
340 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
341 * This will cause notify_change to set these times
344 iap->ia_valid &= ~BOTH_TIME_SET;
349 * The size case is special.
350 * It changes the file as well as the attributes.
352 if (iap->ia_valid & ATTR_SIZE) {
353 if (iap->ia_size < inode->i_size) {
354 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
355 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
361 * If we are changing the size of the file, then
362 * we need to break all leases.
364 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
365 if (host_err == -EWOULDBLOCK)
366 host_err = -ETIMEDOUT;
367 if (host_err) /* ENOMEM or EWOULDBLOCK */
370 host_err = get_write_access(inode);
375 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
377 put_write_access(inode);
383 /* sanitize the mode change */
384 if (iap->ia_valid & ATTR_MODE) {
385 iap->ia_mode &= S_IALLUGO;
386 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
389 /* Revoke setuid/setgid on chown */
390 if (!S_ISDIR(inode->i_mode) &&
391 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
392 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
393 iap->ia_valid |= ATTR_KILL_PRIV;
394 if (iap->ia_valid & ATTR_MODE) {
395 /* we're setting mode too, just clear the s*id bits */
396 iap->ia_mode &= ~S_ISUID;
397 if (iap->ia_mode & S_IXGRP)
398 iap->ia_mode &= ~S_ISGID;
400 /* set ATTR_KILL_* bits and let VFS handle it */
401 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
405 /* Change the attributes. */
407 iap->ia_valid |= ATTR_CTIME;
409 err = nfserr_notsync;
410 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
412 host_err = notify_change(dentry, iap);
413 err = nfserrno(host_err);
417 put_write_access(inode);
419 if (EX_ISSYNC(fhp->fh_export))
420 write_inode_now(inode, 1);
425 err = nfserrno(host_err);
429 #if defined(CONFIG_NFSD_V2_ACL) || \
430 defined(CONFIG_NFSD_V3_ACL) || \
431 defined(CONFIG_NFSD_V4)
432 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
437 buflen = vfs_getxattr(dentry, key, NULL, 0);
441 *buf = kmalloc(buflen, GFP_KERNEL);
445 ret = vfs_getxattr(dentry, key, *buf, buflen);
452 #if defined(CONFIG_NFSD_V4)
454 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
461 buflen = posix_acl_xattr_size(pacl->a_count);
462 buf = kmalloc(buflen, GFP_KERNEL);
467 len = posix_acl_to_xattr(pacl, buf, buflen);
473 error = vfs_setxattr(dentry, key, buf, len, 0);
480 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
481 struct nfs4_acl *acl)
485 struct dentry *dentry;
487 struct posix_acl *pacl = NULL, *dpacl = NULL;
488 unsigned int flags = 0;
491 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
495 dentry = fhp->fh_dentry;
496 inode = dentry->d_inode;
497 if (S_ISDIR(inode->i_mode))
498 flags = NFS4_ACL_DIR;
500 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
501 if (host_error == -EINVAL) {
502 return nfserr_attrnotsupp;
503 } else if (host_error < 0)
506 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
510 if (S_ISDIR(inode->i_mode))
511 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
514 posix_acl_release(pacl);
515 posix_acl_release(dpacl);
517 if (host_error == -EOPNOTSUPP)
518 return nfserr_attrnotsupp;
520 return nfserrno(host_error);
523 static struct posix_acl *
524 _get_posix_acl(struct dentry *dentry, char *key)
527 struct posix_acl *pacl = NULL;
530 buflen = nfsd_getxattr(dentry, key, &buf);
534 return ERR_PTR(buflen);
536 pacl = posix_acl_from_xattr(buf, buflen);
542 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
544 struct inode *inode = dentry->d_inode;
546 struct posix_acl *pacl = NULL, *dpacl = NULL;
547 unsigned int flags = 0;
549 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
550 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
551 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
553 error = PTR_ERR(pacl);
558 if (S_ISDIR(inode->i_mode)) {
559 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
560 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
562 else if (IS_ERR(dpacl)) {
563 error = PTR_ERR(dpacl);
567 flags = NFS4_ACL_DIR;
570 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
572 error = PTR_ERR(*acl);
576 posix_acl_release(pacl);
577 posix_acl_release(dpacl);
581 #endif /* defined(CONFIG_NFS_V4) */
583 #ifdef CONFIG_NFSD_V3
585 * Check server access rights to a file system object
591 static struct accessmap nfs3_regaccess[] = {
592 { NFS3_ACCESS_READ, NFSD_MAY_READ },
593 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
594 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
595 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
600 static struct accessmap nfs3_diraccess[] = {
601 { NFS3_ACCESS_READ, NFSD_MAY_READ },
602 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
603 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
604 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
605 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
610 static struct accessmap nfs3_anyaccess[] = {
611 /* Some clients - Solaris 2.6 at least, make an access call
612 * to the server to check for access for things like /dev/null
613 * (which really, the server doesn't care about). So
614 * We provide simple access checking for them, looking
615 * mainly at mode bits, and we make sure to ignore read-only
618 { NFS3_ACCESS_READ, NFSD_MAY_READ },
619 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
620 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
621 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
627 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
629 struct accessmap *map;
630 struct svc_export *export;
631 struct dentry *dentry;
632 u32 query, result = 0, sresult = 0;
635 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
639 export = fhp->fh_export;
640 dentry = fhp->fh_dentry;
642 if (S_ISREG(dentry->d_inode->i_mode))
643 map = nfs3_regaccess;
644 else if (S_ISDIR(dentry->d_inode->i_mode))
645 map = nfs3_diraccess;
647 map = nfs3_anyaccess;
651 for (; map->access; map++) {
652 if (map->access & query) {
655 sresult |= map->access;
657 err2 = nfsd_permission(rqstp, export, dentry, map->how);
660 result |= map->access;
663 /* the following error codes just mean the access was not allowed,
664 * rather than an error occurred */
668 /* simply don't "or" in the access bit. */
678 *supported = sresult;
683 #endif /* CONFIG_NFSD_V3 */
688 * Open an existing file or directory.
689 * The access argument indicates the type of open (read/write/lock)
690 * N.B. After this call fhp needs an fh_put
693 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
694 int access, struct file **filp)
696 struct dentry *dentry;
698 int flags = O_RDONLY|O_LARGEFILE;
702 validate_process_creds();
705 * If we get here, then the client has already done an "open",
706 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
707 * in case a chmod has now revoked permission.
709 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
713 dentry = fhp->fh_dentry;
714 inode = dentry->d_inode;
716 /* Disallow write access to files with the append-only bit set
717 * or any access when mandatory locking enabled
720 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
723 * We must ignore files (but only files) which might have mandatory
724 * locks on them because there is no way to know if the accesser has
727 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
734 * Check to see if there are any leases on this file.
735 * This may block while leases are broken.
737 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
738 if (host_err == -EWOULDBLOCK)
739 host_err = -ETIMEDOUT;
740 if (host_err) /* NOMEM or WOULDBLOCK */
743 if (access & NFSD_MAY_WRITE) {
744 if (access & NFSD_MAY_READ)
745 flags = O_RDWR|O_LARGEFILE;
747 flags = O_WRONLY|O_LARGEFILE;
751 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
752 flags, current_cred());
754 host_err = PTR_ERR(*filp);
756 err = nfserrno(host_err);
758 validate_process_creds();
766 nfsd_close(struct file *filp)
773 * As this calls fsync (not fdatasync) there is no need for a write_inode
776 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
777 const struct file_operations *fop)
779 struct inode *inode = dp->d_inode;
780 int (*fsync) (struct file *, struct dentry *, int);
783 err = filemap_write_and_wait(inode->i_mapping);
784 if (err == 0 && fop && (fsync = fop->fsync))
785 err = fsync(filp, dp, 0);
790 nfsd_sync(struct file *filp)
793 struct inode *inode = filp->f_path.dentry->d_inode;
794 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
795 mutex_lock(&inode->i_mutex);
796 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
797 mutex_unlock(&inode->i_mutex);
803 nfsd_sync_dir(struct dentry *dp)
805 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
809 * Obtain the readahead parameters for the file
810 * specified by (dev, ino).
813 static inline struct raparms *
814 nfsd_get_raparms(dev_t dev, ino_t ino)
816 struct raparms *ra, **rap, **frap = NULL;
819 struct raparm_hbucket *rab;
821 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
822 rab = &raparm_hash[hash];
824 spin_lock(&rab->pb_lock);
825 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
826 if (ra->p_ino == ino && ra->p_dev == dev)
829 if (ra->p_count == 0)
832 depth = nfsdstats.ra_size*11/10;
834 spin_unlock(&rab->pb_lock);
844 if (rap != &rab->pb_head) {
846 ra->p_next = rab->pb_head;
850 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
851 spin_unlock(&rab->pb_lock);
856 * Grab and keep cached pages associated with a file in the svc_rqst
857 * so that they can be passed to the network sendmsg/sendpage routines
858 * directly. They will be released after the sending has completed.
861 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
862 struct splice_desc *sd)
864 struct svc_rqst *rqstp = sd->u.data;
865 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
866 struct page *page = buf->page;
870 ret = buf->ops->confirm(pipe, buf);
876 if (rqstp->rq_res.page_len == 0) {
881 rqstp->rq_res.page_base = buf->offset;
882 rqstp->rq_res.page_len = size;
883 } else if (page != pp[-1]) {
889 rqstp->rq_res.page_len += size;
891 rqstp->rq_res.page_len += size;
896 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
897 struct splice_desc *sd)
899 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
902 static inline int svc_msnfs(struct svc_fh *ffhp)
905 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
912 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
913 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
922 inode = file->f_path.dentry->d_inode;
924 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
927 /* Get readahead parameters */
928 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
931 file->f_ra = ra->p_ra;
933 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
934 struct splice_desc sd = {
941 rqstp->rq_resused = 1;
942 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
946 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
950 /* Write back readahead params */
952 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
953 spin_lock(&rab->pb_lock);
954 ra->p_ra = file->f_ra;
957 spin_unlock(&rab->pb_lock);
961 nfsdstats.io_read += host_err;
964 fsnotify_access(file->f_path.dentry);
966 err = nfserrno(host_err);
971 static void kill_suid(struct dentry *dentry)
974 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
976 mutex_lock(&dentry->d_inode->i_mutex);
977 notify_change(dentry, &ia);
978 mutex_unlock(&dentry->d_inode->i_mutex);
982 * Gathered writes: If another process is currently writing to the file,
983 * there's a high chance this is another nfsd (triggered by a bulk write
984 * from a client's biod). Rather than syncing the file with each write
985 * request, we sleep for 10 msec.
987 * I don't know if this roughly approximates C. Juszak's idea of
988 * gathered writes, but it's a nice and simple solution (IMHO), and it
991 * Note: we do this only in the NFSv2 case, since v3 and higher have a
992 * better tool (separate unstable writes and commits) for solving this
995 static int wait_for_concurrent_writes(struct file *file)
997 struct inode *inode = file->f_path.dentry->d_inode;
998 static ino_t last_ino;
999 static dev_t last_dev;
1002 if (atomic_read(&inode->i_writecount) > 1
1003 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1004 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1006 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1009 if (inode->i_state & I_DIRTY) {
1010 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1011 err = nfsd_sync(file);
1013 last_ino = inode->i_ino;
1014 last_dev = inode->i_sb->s_dev;
1019 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1020 loff_t offset, struct kvec *vec, int vlen,
1021 unsigned long *cnt, int *stablep)
1023 struct svc_export *exp;
1024 struct dentry *dentry;
1025 struct inode *inode;
1029 int stable = *stablep;
1035 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1036 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1040 dentry = file->f_path.dentry;
1041 inode = dentry->d_inode;
1042 exp = fhp->fh_export;
1045 * Request sync writes if
1046 * - the sync export option has been set, or
1047 * - the client requested O_SYNC behavior (NFSv3 feature).
1048 * - The file system doesn't support fsync().
1049 * When NFSv2 gathered writes have been configured for this volume,
1050 * flushing the data to disk is handled separately below.
1052 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1054 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1056 *stablep = 2; /* FILE_SYNC */
1059 if (!EX_ISSYNC(exp))
1061 if (stable && !use_wgather) {
1062 spin_lock(&file->f_lock);
1063 file->f_flags |= O_SYNC;
1064 spin_unlock(&file->f_lock);
1067 /* Write the data. */
1068 oldfs = get_fs(); set_fs(KERNEL_DS);
1069 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1074 nfsdstats.io_write += host_err;
1075 fsnotify_modify(file->f_path.dentry);
1077 /* clear setuid/setgid flag after write */
1078 if (inode->i_mode & (S_ISUID | S_ISGID))
1081 if (stable && use_wgather)
1082 host_err = wait_for_concurrent_writes(file);
1085 dprintk("nfsd: write complete host_err=%d\n", host_err);
1089 err = nfserrno(host_err);
1095 * Read data from a file. count must contain the requested read count
1096 * on entry. On return, *count contains the number of bytes actually read.
1097 * N.B. After this call fhp needs an fh_put
1100 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1101 loff_t offset, struct kvec *vec, int vlen,
1102 unsigned long *count)
1107 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1108 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1111 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1113 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1116 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1124 * Write data to a file.
1125 * The stable flag requests synchronous writes.
1126 * N.B. After this call fhp needs an fh_put
1129 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1130 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1136 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1137 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1140 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1143 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1148 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1156 #ifdef CONFIG_NFSD_V3
1158 * Commit all pending writes to stable storage.
1159 * Strictly speaking, we could sync just the indicated file region here,
1160 * but there's currently no way we can ask the VFS to do so.
1162 * Unfortunately we cannot lock the file to make sure we return full WCC
1163 * data to the client, as locking happens lower down in the filesystem.
1166 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1167 loff_t offset, unsigned long count)
1172 if ((u64)count > ~(u64)offset)
1173 return nfserr_inval;
1175 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1178 if (EX_ISSYNC(fhp->fh_export)) {
1179 if (file->f_op && file->f_op->fsync) {
1180 err = nfserrno(nfsd_sync(file));
1182 err = nfserr_notsupp;
1189 #endif /* CONFIG_NFSD_V3 */
1192 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1196 * Mode has already been set earlier in create:
1198 iap->ia_valid &= ~ATTR_MODE;
1200 * Setting uid/gid works only for root. Irix appears to
1201 * send along the gid on create when it tries to implement
1202 * setgid directories via NFS:
1204 if (current_fsuid() != 0)
1205 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1207 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1211 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1212 * setting size to 0 may fail for some specific file systems by the permission
1213 * checking which requires WRITE permission but the mode is 000.
1214 * we ignore the resizing(to 0) on the just new created file, since the size is
1215 * 0 after file created.
1217 * call this only after vfs_create() is called.
1220 nfsd_check_ignore_resizing(struct iattr *iap)
1222 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1223 iap->ia_valid &= ~ATTR_SIZE;
1227 * Create a file (regular, directory, device, fifo); UNIX sockets
1228 * not yet implemented.
1229 * If the response fh has been verified, the parent directory should
1230 * already be locked. Note that the parent directory is left locked.
1232 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1235 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1236 char *fname, int flen, struct iattr *iap,
1237 int type, dev_t rdev, struct svc_fh *resfhp)
1239 struct dentry *dentry, *dchild = NULL;
1249 if (isdotent(fname, flen))
1252 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1256 dentry = fhp->fh_dentry;
1257 dirp = dentry->d_inode;
1259 err = nfserr_notdir;
1260 if (!dirp->i_op->lookup)
1263 * Check whether the response file handle has been verified yet.
1264 * If it has, the parent directory should already be locked.
1266 if (!resfhp->fh_dentry) {
1267 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1268 fh_lock_nested(fhp, I_MUTEX_PARENT);
1269 dchild = lookup_one_len(fname, dentry, flen);
1270 host_err = PTR_ERR(dchild);
1273 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1277 /* called from nfsd_proc_create */
1278 dchild = dget(resfhp->fh_dentry);
1279 if (!fhp->fh_locked) {
1280 /* not actually possible */
1282 "nfsd_create: parent %s/%s not locked!\n",
1283 dentry->d_parent->d_name.name,
1284 dentry->d_name.name);
1290 * Make sure the child dentry is still negative ...
1293 if (dchild->d_inode) {
1294 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1295 dentry->d_name.name, dchild->d_name.name);
1299 if (!(iap->ia_valid & ATTR_MODE))
1301 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1304 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1305 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1310 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1315 * Get the dir op function pointer.
1320 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1322 nfsd_check_ignore_resizing(iap);
1325 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1331 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1335 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1339 if (EX_ISSYNC(fhp->fh_export)) {
1340 err = nfserrno(nfsd_sync_dir(dentry));
1341 write_inode_now(dchild->d_inode, 1);
1344 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1347 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1349 * Update the file handle to get the new inode info.
1352 err = fh_update(resfhp);
1354 if (dchild && !IS_ERR(dchild))
1359 err = nfserrno(host_err);
1363 #ifdef CONFIG_NFSD_V3
1365 * NFSv3 version of nfsd_create
1368 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1369 char *fname, int flen, struct iattr *iap,
1370 struct svc_fh *resfhp, int createmode, u32 *verifier,
1371 int *truncp, int *created)
1373 struct dentry *dentry, *dchild = NULL;
1378 __u32 v_mtime=0, v_atime=0;
1384 if (isdotent(fname, flen))
1386 if (!(iap->ia_valid & ATTR_MODE))
1388 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1392 dentry = fhp->fh_dentry;
1393 dirp = dentry->d_inode;
1395 /* Get all the sanity checks out of the way before
1396 * we lock the parent. */
1397 err = nfserr_notdir;
1398 if (!dirp->i_op->lookup)
1400 fh_lock_nested(fhp, I_MUTEX_PARENT);
1403 * Compose the response file handle.
1405 dchild = lookup_one_len(fname, dentry, flen);
1406 host_err = PTR_ERR(dchild);
1410 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1414 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1415 /* solaris7 gets confused (bugid 4218508) if these have
1416 * the high bit set, so just clear the high bits. If this is
1417 * ever changed to use different attrs for storing the
1418 * verifier, then do_open_lookup() will also need to be fixed
1421 v_mtime = verifier[0]&0x7fffffff;
1422 v_atime = verifier[1]&0x7fffffff;
1425 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1428 if (dchild->d_inode) {
1431 switch (createmode) {
1432 case NFS3_CREATE_UNCHECKED:
1433 if (! S_ISREG(dchild->d_inode->i_mode))
1436 /* in nfsv4, we need to treat this case a little
1437 * differently. we don't want to truncate the
1438 * file now; this would be wrong if the OPEN
1439 * fails for some other reason. furthermore,
1440 * if the size is nonzero, we should ignore it
1441 * according to spec!
1443 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1446 iap->ia_valid &= ATTR_SIZE;
1450 case NFS3_CREATE_EXCLUSIVE:
1451 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1452 && dchild->d_inode->i_atime.tv_sec == v_atime
1453 && dchild->d_inode->i_size == 0 )
1456 case NFS3_CREATE_GUARDED:
1459 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1463 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1465 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1471 if (EX_ISSYNC(fhp->fh_export)) {
1472 err = nfserrno(nfsd_sync_dir(dentry));
1473 /* setattr will sync the child (or not) */
1476 nfsd_check_ignore_resizing(iap);
1478 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1479 /* Cram the verifier into atime/mtime */
1480 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1481 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1482 /* XXX someone who knows this better please fix it for nsec */
1483 iap->ia_mtime.tv_sec = v_mtime;
1484 iap->ia_atime.tv_sec = v_atime;
1485 iap->ia_mtime.tv_nsec = 0;
1486 iap->ia_atime.tv_nsec = 0;
1490 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1494 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1496 * Update the filehandle to get the new inode info.
1499 err = fh_update(resfhp);
1503 if (dchild && !IS_ERR(dchild))
1508 err = nfserrno(host_err);
1511 #endif /* CONFIG_NFSD_V3 */
1514 * Read a symlink. On entry, *lenp must contain the maximum path length that
1515 * fits into the buffer. On return, it contains the true length.
1516 * N.B. After this call fhp needs an fh_put
1519 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1521 struct dentry *dentry;
1522 struct inode *inode;
1527 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1531 dentry = fhp->fh_dentry;
1532 inode = dentry->d_inode;
1535 if (!inode->i_op->readlink)
1538 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1539 /* N.B. Why does this call need a get_fs()??
1540 * Remove the set_fs and watch the fireworks:-) --okir
1543 oldfs = get_fs(); set_fs(KERNEL_DS);
1544 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1555 err = nfserrno(host_err);
1560 * Create a symlink and look up its inode
1561 * N.B. After this call _both_ fhp and resfhp need an fh_put
1564 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1565 char *fname, int flen,
1566 char *path, int plen,
1567 struct svc_fh *resfhp,
1570 struct dentry *dentry, *dnew;
1578 if (isdotent(fname, flen))
1581 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1585 dentry = fhp->fh_dentry;
1586 dnew = lookup_one_len(fname, dentry, flen);
1587 host_err = PTR_ERR(dnew);
1591 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1595 if (unlikely(path[plen] != 0)) {
1596 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1597 if (path_alloced == NULL)
1600 strncpy(path_alloced, path, plen);
1601 path_alloced[plen] = 0;
1602 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1603 kfree(path_alloced);
1606 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1609 if (EX_ISSYNC(fhp->fh_export))
1610 host_err = nfsd_sync_dir(dentry);
1612 err = nfserrno(host_err);
1615 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1617 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1619 if (err==0) err = cerr;
1624 err = nfserrno(host_err);
1630 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1633 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1634 char *name, int len, struct svc_fh *tfhp)
1636 struct dentry *ddir, *dnew, *dold;
1637 struct inode *dirp, *dest;
1641 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1644 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1652 if (isdotent(name, len))
1655 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1656 ddir = ffhp->fh_dentry;
1657 dirp = ddir->d_inode;
1659 dnew = lookup_one_len(name, ddir, len);
1660 host_err = PTR_ERR(dnew);
1664 dold = tfhp->fh_dentry;
1665 dest = dold->d_inode;
1667 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1669 err = nfserrno(host_err);
1672 host_err = vfs_link(dold, dirp, dnew);
1674 if (EX_ISSYNC(ffhp->fh_export)) {
1675 err = nfserrno(nfsd_sync_dir(ddir));
1676 write_inode_now(dest, 1);
1680 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1683 err = nfserrno(host_err);
1685 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1694 err = nfserrno(host_err);
1700 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1703 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1704 struct svc_fh *tfhp, char *tname, int tlen)
1706 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1707 struct inode *fdir, *tdir;
1711 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1714 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1718 fdentry = ffhp->fh_dentry;
1719 fdir = fdentry->d_inode;
1721 tdentry = tfhp->fh_dentry;
1722 tdir = tdentry->d_inode;
1724 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1725 if (ffhp->fh_export != tfhp->fh_export)
1729 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1732 /* cannot use fh_lock as we need deadlock protective ordering
1733 * so do it by hand */
1734 trap = lock_rename(tdentry, fdentry);
1735 ffhp->fh_locked = tfhp->fh_locked = 1;
1739 odentry = lookup_one_len(fname, fdentry, flen);
1740 host_err = PTR_ERR(odentry);
1741 if (IS_ERR(odentry))
1745 if (!odentry->d_inode)
1748 if (odentry == trap)
1751 ndentry = lookup_one_len(tname, tdentry, tlen);
1752 host_err = PTR_ERR(ndentry);
1753 if (IS_ERR(ndentry))
1755 host_err = -ENOTEMPTY;
1756 if (ndentry == trap)
1759 if (svc_msnfs(ffhp) &&
1760 ((atomic_read(&odentry->d_count) > 1)
1761 || (atomic_read(&ndentry->d_count) > 1))) {
1767 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1769 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1773 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1774 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1775 host_err = nfsd_sync_dir(tdentry);
1777 host_err = nfsd_sync_dir(fdentry);
1780 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1787 err = nfserrno(host_err);
1789 /* we cannot reply on fh_unlock on the two filehandles,
1790 * as that would do the wrong thing if the two directories
1791 * were the same, so again we do it by hand
1793 fill_post_wcc(ffhp);
1794 fill_post_wcc(tfhp);
1795 unlock_rename(tdentry, fdentry);
1796 ffhp->fh_locked = tfhp->fh_locked = 0;
1803 * Unlink a file or directory
1804 * N.B. After this call fhp needs an fh_put
1807 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1808 char *fname, int flen)
1810 struct dentry *dentry, *rdentry;
1816 if (!flen || isdotent(fname, flen))
1818 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1822 fh_lock_nested(fhp, I_MUTEX_PARENT);
1823 dentry = fhp->fh_dentry;
1824 dirp = dentry->d_inode;
1826 rdentry = lookup_one_len(fname, dentry, flen);
1827 host_err = PTR_ERR(rdentry);
1828 if (IS_ERR(rdentry))
1831 if (!rdentry->d_inode) {
1838 type = rdentry->d_inode->i_mode & S_IFMT;
1840 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1844 if (type != S_IFDIR) { /* It's UNLINK */
1846 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1847 (atomic_read(&rdentry->d_count) > 1)) {
1851 host_err = vfs_unlink(dirp, rdentry);
1852 } else { /* It's RMDIR */
1853 host_err = vfs_rmdir(dirp, rdentry);
1860 if (EX_ISSYNC(fhp->fh_export))
1861 host_err = nfsd_sync_dir(dentry);
1864 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1866 err = nfserrno(host_err);
1872 * We do this buffering because we must not call back into the file
1873 * system's ->lookup() method from the filldir callback. That may well
1874 * deadlock a number of file systems.
1876 * This is based heavily on the implementation of same in XFS.
1878 struct buffered_dirent {
1882 unsigned int d_type;
1886 struct readdir_data {
1892 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1893 loff_t offset, u64 ino, unsigned int d_type)
1895 struct readdir_data *buf = __buf;
1896 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1897 unsigned int reclen;
1899 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1900 if (buf->used + reclen > PAGE_SIZE) {
1905 de->namlen = namlen;
1906 de->offset = offset;
1908 de->d_type = d_type;
1909 memcpy(de->name, name, namlen);
1910 buf->used += reclen;
1915 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1916 struct readdir_cd *cdp, loff_t *offsetp)
1918 struct readdir_data buf;
1919 struct buffered_dirent *de;
1924 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1926 return nfserrno(-ENOMEM);
1931 struct inode *dir_inode = file->f_path.dentry->d_inode;
1932 unsigned int reclen;
1934 cdp->err = nfserr_eof; /* will be cleared on successful read */
1938 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1951 * Various filldir functions may end up calling back into
1952 * lookup_one_len() and the file system's ->lookup() method.
1953 * These expect i_mutex to be held, as it would within readdir.
1955 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1959 de = (struct buffered_dirent *)buf.dirent;
1961 offset = de->offset;
1963 if (func(cdp, de->name, de->namlen, de->offset,
1964 de->ino, de->d_type))
1967 if (cdp->err != nfs_ok)
1970 reclen = ALIGN(sizeof(*de) + de->namlen,
1973 de = (struct buffered_dirent *)((char *)de + reclen);
1975 mutex_unlock(&dir_inode->i_mutex);
1976 if (size > 0) /* We bailed out early */
1979 offset = vfs_llseek(file, 0, SEEK_CUR);
1982 free_page((unsigned long)(buf.dirent));
1985 return nfserrno(host_err);
1992 * Read entries from a directory.
1993 * The NFSv3/4 verifier we ignore for now.
1996 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1997 struct readdir_cd *cdp, filldir_t func)
2001 loff_t offset = *offsetp;
2003 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2007 offset = vfs_llseek(file, offset, 0);
2009 err = nfserrno((int)offset);
2013 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2015 if (err == nfserr_eof || err == nfserr_toosmall)
2016 err = nfs_ok; /* can still be found in ->err */
2024 * Get file system stats
2025 * N.B. After this call fhp needs an fh_put
2028 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2030 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2031 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2036 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2038 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2042 * Check for a user's access permissions to this inode.
2045 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2046 struct dentry *dentry, int acc)
2048 struct inode *inode = dentry->d_inode;
2052 if (acc == NFSD_MAY_NOP)
2055 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2057 (acc & NFSD_MAY_READ)? " read" : "",
2058 (acc & NFSD_MAY_WRITE)? " write" : "",
2059 (acc & NFSD_MAY_EXEC)? " exec" : "",
2060 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2061 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2062 (acc & NFSD_MAY_LOCK)? " lock" : "",
2063 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2065 IS_IMMUTABLE(inode)? " immut" : "",
2066 IS_APPEND(inode)? " append" : "",
2067 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2068 dprintk(" owner %d/%d user %d/%d\n",
2069 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2072 /* Normally we reject any write/sattr etc access on a read-only file
2073 * system. But if it is IRIX doing check on write-access for a
2074 * device special file, we ignore rofs.
2076 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2077 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2078 if (exp_rdonly(rqstp, exp) ||
2079 __mnt_is_readonly(exp->ex_path.mnt))
2081 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2084 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2087 if (acc & NFSD_MAY_LOCK) {
2088 /* If we cannot rely on authentication in NLM requests,
2089 * just allow locks, otherwise require read permission, or
2092 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2095 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2098 * The file owner always gets access permission for accesses that
2099 * would normally be checked at open time. This is to make
2100 * file access work even when the client has done a fchmod(fd, 0).
2102 * However, `cp foo bar' should fail nevertheless when bar is
2103 * readonly. A sensible way to do this might be to reject all
2104 * attempts to truncate a read-only file, because a creat() call
2105 * always implies file truncation.
2106 * ... but this isn't really fair. A process may reasonably call
2107 * ftruncate on an open file descriptor on a file with perm 000.
2108 * We must trust the client to do permission checking - using "ACCESS"
2111 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2112 inode->i_uid == current_fsuid())
2115 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2116 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2118 /* Allow read access to binaries even when mode 111 */
2119 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2120 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2121 err = inode_permission(inode, MAY_EXEC);
2125 /* Do integrity (permission) checking now, but defer incrementing
2126 * IMA counts to the actual file open.
2128 path.mnt = exp->ex_path.mnt;
2129 path.dentry = dentry;
2130 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC));
2132 return err? nfserrno(err) : 0;
2136 nfsd_racache_shutdown(void)
2138 struct raparms *raparm, *last_raparm;
2141 dprintk("nfsd: freeing readahead buffers.\n");
2143 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2144 raparm = raparm_hash[i].pb_head;
2146 last_raparm = raparm;
2147 raparm = raparm->p_next;
2150 raparm_hash[i].pb_head = NULL;
2154 * Initialize readahead param cache
2157 nfsd_racache_init(int cache_size)
2162 struct raparms **raparm = NULL;
2165 if (raparm_hash[0].pb_head)
2167 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2170 cache_size = nperbucket * RAPARM_HASH_SIZE;
2172 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2174 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2175 spin_lock_init(&raparm_hash[i].pb_lock);
2177 raparm = &raparm_hash[i].pb_head;
2178 for (j = 0; j < nperbucket; j++) {
2179 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2182 raparm = &(*raparm)->p_next;
2187 nfsdstats.ra_size = cache_size;
2191 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2192 nfsd_racache_shutdown();
2196 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2198 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2200 struct inode *inode = fhp->fh_dentry->d_inode;
2204 struct posix_acl *acl;
2206 if (!IS_POSIXACL(inode))
2207 return ERR_PTR(-EOPNOTSUPP);
2210 case ACL_TYPE_ACCESS:
2211 name = POSIX_ACL_XATTR_ACCESS;
2213 case ACL_TYPE_DEFAULT:
2214 name = POSIX_ACL_XATTR_DEFAULT;
2217 return ERR_PTR(-EOPNOTSUPP);
2220 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2222 return ERR_PTR(size);
2224 acl = posix_acl_from_xattr(value, size);
2230 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2232 struct inode *inode = fhp->fh_dentry->d_inode;
2238 if (!IS_POSIXACL(inode) ||
2239 !inode->i_op->setxattr || !inode->i_op->removexattr)
2242 case ACL_TYPE_ACCESS:
2243 name = POSIX_ACL_XATTR_ACCESS;
2245 case ACL_TYPE_DEFAULT:
2246 name = POSIX_ACL_XATTR_DEFAULT;
2252 if (acl && acl->a_count) {
2253 size = posix_acl_xattr_size(acl->a_count);
2254 value = kmalloc(size, GFP_KERNEL);
2257 error = posix_acl_to_xattr(acl, value, size);
2264 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2268 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2270 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2273 error = vfs_removexattr(fhp->fh_dentry, name);
2274 if (error == -ENODATA)
2278 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2284 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */