2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/falloc.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
32 #include <linux/security.h>
36 #endif /* CONFIG_NFSD_V3 */
41 #endif /* CONFIG_NFSD_V4 */
46 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
50 * This is a cache of readahead params that help us choose the proper
51 * readahead strategy. Initially, we set all readahead parameters to 0
52 * and let the VFS handle things.
53 * If you increase the number of cached files very much, you'll need to
54 * add a hash table here.
57 struct raparms *p_next;
62 struct file_ra_state p_ra;
63 unsigned int p_hindex;
66 struct raparm_hbucket {
67 struct raparms *pb_head;
69 } ____cacheline_aligned_in_smp;
71 #define RAPARM_HASH_BITS 4
72 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
73 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
74 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
77 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
79 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
80 * or nfs_ok having possibly changed *dpp and *expp
83 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
84 struct svc_export **expp)
86 struct svc_export *exp = *expp, *exp2 = NULL;
87 struct dentry *dentry = *dpp;
88 struct path path = {.mnt = mntget(exp->ex_path.mnt),
89 .dentry = dget(dentry)};
92 err = follow_down(&path);
95 if (path.mnt == exp->ex_path.mnt && path.dentry == dentry &&
96 nfsd_mountpoint(dentry, exp) == 2) {
97 /* This is only a mountpoint in some other namespace */
102 exp2 = rqst_exp_get_by_name(rqstp, &path);
106 * We normally allow NFS clients to continue
107 * "underneath" a mountpoint that is not exported.
108 * The exception is V4ROOT, where no traversal is ever
109 * allowed without an explicit export of the new
112 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
117 if (nfsd_v4client(rqstp) ||
118 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
119 /* successfully crossed mount point */
121 * This is subtle: path.dentry is *not* on path.mnt
122 * at this point. The only reason we are safe is that
123 * original mnt is pinned down by exp, so we should
124 * put path *before* putting exp
127 path.dentry = dentry;
137 static void follow_to_parent(struct path *path)
141 while (path->dentry == path->mnt->mnt_root && follow_up(path))
143 dp = dget_parent(path->dentry);
148 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
150 struct svc_export *exp2;
151 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
152 .dentry = dget(dparent)};
154 follow_to_parent(&path);
156 exp2 = rqst_exp_parent(rqstp, &path);
157 if (PTR_ERR(exp2) == -ENOENT) {
158 *dentryp = dget(dparent);
159 } else if (IS_ERR(exp2)) {
161 return PTR_ERR(exp2);
163 *dentryp = dget(path.dentry);
172 * For nfsd purposes, we treat V4ROOT exports as though there was an
173 * export at *every* directory.
175 * '1' if this dentry *must* be an export point,
176 * '2' if it might be, if there is really a mount here, and
177 * '0' if there is no chance of an export point here.
179 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
181 if (!d_inode(dentry))
183 if (exp->ex_flags & NFSEXP_V4ROOT)
185 if (nfsd4_is_junction(dentry))
187 if (d_mountpoint(dentry))
189 * Might only be a mountpoint in a different namespace,
190 * but we need to check.
197 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
198 const char *name, unsigned int len,
199 struct svc_export **exp_ret, struct dentry **dentry_ret)
201 struct svc_export *exp;
202 struct dentry *dparent;
203 struct dentry *dentry;
206 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
208 dparent = fhp->fh_dentry;
209 exp = exp_get(fhp->fh_export);
211 /* Lookup the name, but don't follow links */
212 if (isdotent(name, len)) {
214 dentry = dget(dparent);
215 else if (dparent != exp->ex_path.dentry)
216 dentry = dget_parent(dparent);
217 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
218 dentry = dget(dparent); /* .. == . just like at / */
220 /* checking mountpoint crossing is very different when stepping up */
221 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
227 * In the nfsd4_open() case, this may be held across
228 * subsequent open and delegation acquisition which may
229 * need to take the child's i_mutex:
231 fh_lock_nested(fhp, I_MUTEX_PARENT);
232 dentry = lookup_one_len(name, dparent, len);
233 host_err = PTR_ERR(dentry);
237 * check if we have crossed a mount point ...
239 if (nfsd_mountpoint(dentry, exp)) {
240 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
246 *dentry_ret = dentry;
252 return nfserrno(host_err);
256 * Look up one component of a pathname.
257 * N.B. After this call _both_ fhp and resfh need an fh_put
259 * If the lookup would cross a mountpoint, and the mounted filesystem
260 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
261 * accepted as it stands and the mounted directory is
262 * returned. Otherwise the covered directory is returned.
263 * NOTE: this mountpoint crossing is not supported properly by all
264 * clients and is explicitly disallowed for NFSv3
265 * NeilBrown <neilb@cse.unsw.edu.au>
268 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
269 unsigned int len, struct svc_fh *resfh)
271 struct svc_export *exp;
272 struct dentry *dentry;
275 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
278 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
281 err = check_nfsd_access(exp, rqstp);
285 * Note: we compose the file handle now, but as the
286 * dentry may be negative, it may need to be updated.
288 err = fh_compose(resfh, exp, dentry, fhp);
289 if (!err && d_really_is_negative(dentry))
298 * Commit metadata changes to stable storage.
301 commit_metadata(struct svc_fh *fhp)
303 struct inode *inode = d_inode(fhp->fh_dentry);
304 const struct export_operations *export_ops = inode->i_sb->s_export_op;
306 if (!EX_ISSYNC(fhp->fh_export))
309 if (export_ops->commit_metadata)
310 return export_ops->commit_metadata(inode);
311 return sync_inode_metadata(inode, 1);
315 * Go over the attributes and take care of the small differences between
316 * NFS semantics and what Linux expects.
319 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
321 /* sanitize the mode change */
322 if (iap->ia_valid & ATTR_MODE) {
323 iap->ia_mode &= S_IALLUGO;
324 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
327 /* Revoke setuid/setgid on chown */
328 if (!S_ISDIR(inode->i_mode) &&
329 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
330 iap->ia_valid |= ATTR_KILL_PRIV;
331 if (iap->ia_valid & ATTR_MODE) {
332 /* we're setting mode too, just clear the s*id bits */
333 iap->ia_mode &= ~S_ISUID;
334 if (iap->ia_mode & S_IXGRP)
335 iap->ia_mode &= ~S_ISGID;
337 /* set ATTR_KILL_* bits and let VFS handle it */
338 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
344 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
347 struct inode *inode = d_inode(fhp->fh_dentry);
350 if (iap->ia_size < inode->i_size) {
353 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
354 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
359 host_err = get_write_access(inode);
363 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
365 goto out_put_write_access;
368 out_put_write_access:
369 put_write_access(inode);
371 return nfserrno(host_err);
375 * Set various file attributes. After this call fhp needs an fh_put.
378 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
379 int check_guard, time_t guardtime)
381 struct dentry *dentry;
383 int accmode = NFSD_MAY_SATTR;
387 bool get_write_count;
388 bool size_change = (iap->ia_valid & ATTR_SIZE);
390 if (iap->ia_valid & ATTR_SIZE) {
391 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
396 * If utimes(2) and friends are called with times not NULL, we should
397 * not set NFSD_MAY_WRITE bit. Otherwise fh_verify->nfsd_permission
398 * will return EACCESS, when the caller's effective UID does not match
399 * the owner of the file, and the caller is not privileged. In this
400 * situation, we should return EPERM(notify_change will return this).
402 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME)) {
403 accmode |= NFSD_MAY_OWNER_OVERRIDE;
404 if (!(iap->ia_valid & (ATTR_ATIME_SET | ATTR_MTIME_SET)))
405 accmode |= NFSD_MAY_WRITE;
408 /* Callers that do fh_verify should do the fh_want_write: */
409 get_write_count = !fhp->fh_dentry;
412 err = fh_verify(rqstp, fhp, ftype, accmode);
415 if (get_write_count) {
416 host_err = fh_want_write(fhp);
421 dentry = fhp->fh_dentry;
422 inode = d_inode(dentry);
424 /* Ignore any mode updates on symlinks */
425 if (S_ISLNK(inode->i_mode))
426 iap->ia_valid &= ~ATTR_MODE;
431 nfsd_sanitize_attrs(inode, iap);
433 if (check_guard && guardtime != inode->i_ctime.tv_sec)
434 return nfserr_notsync;
437 * The size case is special, it changes the file in addition to the
438 * attributes, and file systems don't expect it to be mixed with
439 * "random" attribute changes. We thus split out the size change
440 * into a separate call to ->setattr, and do the rest as a separate
444 err = nfsd_get_write_access(rqstp, fhp, iap);
452 * RFC5661, Section 18.30.4:
453 * Changing the size of a file with SETATTR indirectly
454 * changes the time_modify and change attributes.
456 * (and similar for the older RFCs)
458 struct iattr size_attr = {
459 .ia_valid = ATTR_SIZE | ATTR_CTIME | ATTR_MTIME,
460 .ia_size = iap->ia_size,
463 host_err = notify_change(dentry, &size_attr, NULL);
466 iap->ia_valid &= ~ATTR_SIZE;
469 * Avoid the additional setattr call below if the only other
470 * attribute that the client sends is the mtime, as we update
471 * it as part of the size change above.
473 if ((iap->ia_valid & ~ATTR_MTIME) == 0)
477 iap->ia_valid |= ATTR_CTIME;
478 host_err = notify_change(dentry, iap, NULL);
483 put_write_access(inode);
486 host_err = commit_metadata(fhp);
487 return nfserrno(host_err);
490 #if defined(CONFIG_NFSD_V4)
492 * NFS junction information is stored in an extended attribute.
494 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
497 * nfsd4_is_junction - Test if an object could be an NFS junction
499 * @dentry: object to test
501 * Returns 1 if "dentry" appears to contain NFS junction information.
502 * Otherwise 0 is returned.
504 int nfsd4_is_junction(struct dentry *dentry)
506 struct inode *inode = d_inode(dentry);
510 if (inode->i_mode & S_IXUGO)
512 if (!(inode->i_mode & S_ISVTX))
514 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
518 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
519 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
520 struct xdr_netobj *label)
524 struct dentry *dentry;
526 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
530 dentry = fhp->fh_dentry;
532 mutex_lock(&d_inode(dentry)->i_mutex);
533 host_error = security_inode_setsecctx(dentry, label->data, label->len);
534 mutex_unlock(&d_inode(dentry)->i_mutex);
535 return nfserrno(host_error);
538 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
539 struct xdr_netobj *label)
541 return nfserr_notsupp;
545 __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp,
546 struct file *file, loff_t offset, loff_t len,
551 if (!S_ISREG(file_inode(file)->i_mode))
554 error = vfs_fallocate(file, flags, offset, len);
556 error = commit_metadata(fhp);
558 return nfserrno(error);
560 #endif /* defined(CONFIG_NFSD_V4) */
562 #ifdef CONFIG_NFSD_V3
564 * Check server access rights to a file system object
570 static struct accessmap nfs3_regaccess[] = {
571 { NFS3_ACCESS_READ, NFSD_MAY_READ },
572 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
573 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
574 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
579 static struct accessmap nfs3_diraccess[] = {
580 { NFS3_ACCESS_READ, NFSD_MAY_READ },
581 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
582 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
583 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
584 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
589 static struct accessmap nfs3_anyaccess[] = {
590 /* Some clients - Solaris 2.6 at least, make an access call
591 * to the server to check for access for things like /dev/null
592 * (which really, the server doesn't care about). So
593 * We provide simple access checking for them, looking
594 * mainly at mode bits, and we make sure to ignore read-only
597 { NFS3_ACCESS_READ, NFSD_MAY_READ },
598 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
599 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
600 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
606 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
608 struct accessmap *map;
609 struct svc_export *export;
610 struct dentry *dentry;
611 u32 query, result = 0, sresult = 0;
614 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
618 export = fhp->fh_export;
619 dentry = fhp->fh_dentry;
621 if (d_is_reg(dentry))
622 map = nfs3_regaccess;
623 else if (d_is_dir(dentry))
624 map = nfs3_diraccess;
626 map = nfs3_anyaccess;
630 for (; map->access; map++) {
631 if (map->access & query) {
634 sresult |= map->access;
636 err2 = nfsd_permission(rqstp, export, dentry, map->how);
639 result |= map->access;
642 /* the following error codes just mean the access was not allowed,
643 * rather than an error occurred */
647 /* simply don't "or" in the access bit. */
657 *supported = sresult;
662 #endif /* CONFIG_NFSD_V3 */
664 static int nfsd_open_break_lease(struct inode *inode, int access)
668 if (access & NFSD_MAY_NOT_BREAK_LEASE)
670 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
671 return break_lease(inode, mode | O_NONBLOCK);
675 * Open an existing file or directory.
676 * The may_flags argument indicates the type of open (read/write/lock)
677 * and additional flags.
678 * N.B. After this call fhp needs an fh_put
681 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
682 int may_flags, struct file **filp)
687 int flags = O_RDONLY|O_LARGEFILE;
691 validate_process_creds();
694 * If we get here, then the client has already done an "open",
695 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
696 * in case a chmod has now revoked permission.
698 * Arguably we should also allow the owner override for
699 * directories, but we never have and it doesn't seem to have
700 * caused anyone a problem. If we were to change this, note
701 * also that our filldir callbacks would need a variant of
702 * lookup_one_len that doesn't check permissions.
705 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
706 err = fh_verify(rqstp, fhp, type, may_flags);
710 path.mnt = fhp->fh_export->ex_path.mnt;
711 path.dentry = fhp->fh_dentry;
712 inode = d_inode(path.dentry);
714 /* Disallow write access to files with the append-only bit set
715 * or any access when mandatory locking enabled
718 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
721 * We must ignore files (but only files) which might have mandatory
722 * locks on them because there is no way to know if the accesser has
725 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
731 host_err = nfsd_open_break_lease(inode, may_flags);
732 if (host_err) /* NOMEM or WOULDBLOCK */
735 if (may_flags & NFSD_MAY_WRITE) {
736 if (may_flags & NFSD_MAY_READ)
737 flags = O_RDWR|O_LARGEFILE;
739 flags = O_WRONLY|O_LARGEFILE;
742 file = dentry_open(&path, flags, current_cred());
744 host_err = PTR_ERR(file);
748 host_err = ima_file_check(file, may_flags, 0);
754 if (may_flags & NFSD_MAY_64BIT_COOKIE)
755 file->f_mode |= FMODE_64BITHASH;
757 file->f_mode |= FMODE_32BITHASH;
761 err = nfserrno(host_err);
763 validate_process_creds();
768 nfsd_init_raparms(struct file *file)
770 struct inode *inode = file_inode(file);
771 dev_t dev = inode->i_sb->s_dev;
772 ino_t ino = inode->i_ino;
773 struct raparms *ra, **rap, **frap = NULL;
776 struct raparm_hbucket *rab;
778 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
779 rab = &raparm_hash[hash];
781 spin_lock(&rab->pb_lock);
782 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
783 if (ra->p_ino == ino && ra->p_dev == dev)
786 if (ra->p_count == 0)
789 depth = nfsdstats.ra_size;
791 spin_unlock(&rab->pb_lock);
801 if (rap != &rab->pb_head) {
803 ra->p_next = rab->pb_head;
807 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
808 spin_unlock(&rab->pb_lock);
811 file->f_ra = ra->p_ra;
815 void nfsd_put_raparams(struct file *file, struct raparms *ra)
817 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
819 spin_lock(&rab->pb_lock);
820 ra->p_ra = file->f_ra;
823 spin_unlock(&rab->pb_lock);
827 * Grab and keep cached pages associated with a file in the svc_rqst
828 * so that they can be passed to the network sendmsg/sendpage routines
829 * directly. They will be released after the sending has completed.
832 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
833 struct splice_desc *sd)
835 struct svc_rqst *rqstp = sd->u.data;
836 struct page **pp = rqstp->rq_next_page;
837 struct page *page = buf->page;
842 if (rqstp->rq_res.page_len == 0) {
844 put_page(*rqstp->rq_next_page);
845 *(rqstp->rq_next_page++) = page;
846 rqstp->rq_res.page_base = buf->offset;
847 rqstp->rq_res.page_len = size;
848 } else if (page != pp[-1]) {
850 if (*rqstp->rq_next_page)
851 put_page(*rqstp->rq_next_page);
852 *(rqstp->rq_next_page++) = page;
853 rqstp->rq_res.page_len += size;
855 rqstp->rq_res.page_len += size;
860 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
861 struct splice_desc *sd)
863 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
867 nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
870 nfsdstats.io_read += host_err;
872 fsnotify_access(file);
875 return nfserrno(host_err);
878 __be32 nfsd_splice_read(struct svc_rqst *rqstp,
879 struct file *file, loff_t offset, unsigned long *count)
881 struct splice_desc sd = {
889 rqstp->rq_next_page = rqstp->rq_respages + 1;
890 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
891 return nfsd_finish_read(file, count, host_err);
894 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
895 unsigned long *count)
902 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
904 return nfsd_finish_read(file, count, host_err);
908 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
909 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
911 if (file->f_op->splice_read && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags))
912 return nfsd_splice_read(rqstp, file, offset, count);
914 return nfsd_readv(file, offset, vec, vlen, count);
918 * Gathered writes: If another process is currently writing to the file,
919 * there's a high chance this is another nfsd (triggered by a bulk write
920 * from a client's biod). Rather than syncing the file with each write
921 * request, we sleep for 10 msec.
923 * I don't know if this roughly approximates C. Juszak's idea of
924 * gathered writes, but it's a nice and simple solution (IMHO), and it
927 * Note: we do this only in the NFSv2 case, since v3 and higher have a
928 * better tool (separate unstable writes and commits) for solving this
931 static int wait_for_concurrent_writes(struct file *file)
933 struct inode *inode = file_inode(file);
934 static ino_t last_ino;
935 static dev_t last_dev;
938 if (atomic_read(&inode->i_writecount) > 1
939 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
940 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
942 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
945 if (inode->i_state & I_DIRTY) {
946 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
947 err = vfs_fsync(file, 0);
949 last_ino = inode->i_ino;
950 last_dev = inode->i_sb->s_dev;
955 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
956 loff_t offset, struct kvec *vec, int vlen,
957 unsigned long *cnt, int *stablep)
959 struct svc_export *exp;
964 int stable = *stablep;
967 loff_t end = LLONG_MAX;
968 unsigned int pflags = current->flags;
970 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
972 * We want less throttling in balance_dirty_pages()
973 * and shrink_inactive_list() so that nfs to
974 * localhost doesn't cause nfsd to lock up due to all
975 * the client's dirty pages or its congested queue.
977 current->flags |= PF_LESS_THROTTLE;
979 inode = file_inode(file);
980 exp = fhp->fh_export;
982 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
987 /* Write the data. */
988 oldfs = get_fs(); set_fs(KERNEL_DS);
989 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
994 nfsdstats.io_write += host_err;
995 fsnotify_modify(file);
999 host_err = wait_for_concurrent_writes(file);
1002 end = offset + *cnt - 1;
1003 host_err = vfs_fsync_range(file, offset, end, 0);
1008 dprintk("nfsd: write complete host_err=%d\n", host_err);
1012 err = nfserrno(host_err);
1013 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
1014 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
1019 * Read data from a file. count must contain the requested read count
1020 * on entry. On return, *count contains the number of bytes actually read.
1021 * N.B. After this call fhp needs an fh_put
1023 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1024 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1030 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1034 ra = nfsd_init_raparms(file);
1035 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1037 nfsd_put_raparams(file, ra);
1044 * Write data to a file.
1045 * The stable flag requests synchronous writes.
1046 * N.B. After this call fhp needs an fh_put
1049 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1050 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1056 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1057 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1060 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1063 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1068 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1076 #ifdef CONFIG_NFSD_V3
1078 * Commit all pending writes to stable storage.
1080 * Note: we only guarantee that data that lies within the range specified
1081 * by the 'offset' and 'count' parameters will be synced.
1083 * Unfortunately we cannot lock the file to make sure we return full WCC
1084 * data to the client, as locking happens lower down in the filesystem.
1087 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1088 loff_t offset, unsigned long count)
1091 loff_t end = LLONG_MAX;
1092 __be32 err = nfserr_inval;
1097 end = offset + (loff_t)count - 1;
1102 err = nfsd_open(rqstp, fhp, S_IFREG,
1103 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1106 if (EX_ISSYNC(fhp->fh_export)) {
1107 int err2 = vfs_fsync_range(file, offset, end, 0);
1109 if (err2 != -EINVAL)
1110 err = nfserrno(err2);
1112 err = nfserr_notsupp;
1119 #endif /* CONFIG_NFSD_V3 */
1122 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1126 * Mode has already been set earlier in create:
1128 iap->ia_valid &= ~ATTR_MODE;
1130 * Setting uid/gid works only for root. Irix appears to
1131 * send along the gid on create when it tries to implement
1132 * setgid directories via NFS:
1134 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1135 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1137 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1138 /* Callers expect file metadata to be committed here */
1139 return nfserrno(commit_metadata(resfhp));
1142 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1143 * setting size to 0 may fail for some specific file systems by the permission
1144 * checking which requires WRITE permission but the mode is 000.
1145 * we ignore the resizing(to 0) on the just new created file, since the size is
1146 * 0 after file created.
1148 * call this only after vfs_create() is called.
1151 nfsd_check_ignore_resizing(struct iattr *iap)
1153 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1154 iap->ia_valid &= ~ATTR_SIZE;
1158 * Create a file (regular, directory, device, fifo); UNIX sockets
1159 * not yet implemented.
1160 * If the response fh has been verified, the parent directory should
1161 * already be locked. Note that the parent directory is left locked.
1163 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1166 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1167 char *fname, int flen, struct iattr *iap,
1168 int type, dev_t rdev, struct svc_fh *resfhp)
1170 struct dentry *dentry, *dchild = NULL;
1180 if (isdotent(fname, flen))
1183 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1187 dentry = fhp->fh_dentry;
1188 dirp = d_inode(dentry);
1190 err = nfserr_notdir;
1191 if (!dirp->i_op->lookup)
1194 * Check whether the response file handle has been verified yet.
1195 * If it has, the parent directory should already be locked.
1197 if (!resfhp->fh_dentry) {
1198 host_err = fh_want_write(fhp);
1202 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1203 fh_lock_nested(fhp, I_MUTEX_PARENT);
1204 dchild = lookup_one_len(fname, dentry, flen);
1205 host_err = PTR_ERR(dchild);
1208 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1212 /* called from nfsd_proc_create */
1213 dchild = dget(resfhp->fh_dentry);
1214 if (!fhp->fh_locked) {
1215 /* not actually possible */
1217 "nfsd_create: parent %pd2 not locked!\n",
1224 * Make sure the child dentry is still negative ...
1227 if (d_really_is_positive(dchild)) {
1228 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1233 if (!(iap->ia_valid & ATTR_MODE))
1235 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1238 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1239 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1245 * Get the dir op function pointer.
1251 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1253 nfsd_check_ignore_resizing(iap);
1256 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1262 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1268 err = nfsd_create_setattr(rqstp, resfhp, iap);
1271 * nfsd_create_setattr already committed the child. Transactional
1272 * filesystems had a chance to commit changes for both parent and
1273 * child * simultaneously making the following commit_metadata a
1276 err2 = nfserrno(commit_metadata(fhp));
1280 * Update the file handle to get the new inode info.
1283 err = fh_update(resfhp);
1285 if (dchild && !IS_ERR(dchild))
1290 err = nfserrno(host_err);
1294 #ifdef CONFIG_NFSD_V3
1297 * NFSv3 and NFSv4 version of nfsd_create
1300 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1301 char *fname, int flen, struct iattr *iap,
1302 struct svc_fh *resfhp, int createmode, u32 *verifier,
1303 bool *truncp, bool *created)
1305 struct dentry *dentry, *dchild = NULL;
1309 __u32 v_mtime=0, v_atime=0;
1315 if (isdotent(fname, flen))
1317 if (!(iap->ia_valid & ATTR_MODE))
1319 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1323 dentry = fhp->fh_dentry;
1324 dirp = d_inode(dentry);
1326 /* Get all the sanity checks out of the way before
1327 * we lock the parent. */
1328 err = nfserr_notdir;
1329 if (!dirp->i_op->lookup)
1332 host_err = fh_want_write(fhp);
1336 fh_lock_nested(fhp, I_MUTEX_PARENT);
1339 * Compose the response file handle.
1341 dchild = lookup_one_len(fname, dentry, flen);
1342 host_err = PTR_ERR(dchild);
1346 /* If file doesn't exist, check for permissions to create one */
1347 if (d_really_is_negative(dchild)) {
1348 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1353 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1357 if (nfsd_create_is_exclusive(createmode)) {
1358 /* solaris7 gets confused (bugid 4218508) if these have
1359 * the high bit set, so just clear the high bits. If this is
1360 * ever changed to use different attrs for storing the
1361 * verifier, then do_open_lookup() will also need to be fixed
1364 v_mtime = verifier[0]&0x7fffffff;
1365 v_atime = verifier[1]&0x7fffffff;
1368 if (d_really_is_positive(dchild)) {
1371 switch (createmode) {
1372 case NFS3_CREATE_UNCHECKED:
1373 if (! d_is_reg(dchild))
1376 /* in nfsv4, we need to treat this case a little
1377 * differently. we don't want to truncate the
1378 * file now; this would be wrong if the OPEN
1379 * fails for some other reason. furthermore,
1380 * if the size is nonzero, we should ignore it
1381 * according to spec!
1383 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1386 iap->ia_valid &= ATTR_SIZE;
1390 case NFS3_CREATE_EXCLUSIVE:
1391 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1392 && d_inode(dchild)->i_atime.tv_sec == v_atime
1393 && d_inode(dchild)->i_size == 0 ) {
1398 case NFS4_CREATE_EXCLUSIVE4_1:
1399 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1400 && d_inode(dchild)->i_atime.tv_sec == v_atime
1401 && d_inode(dchild)->i_size == 0 ) {
1407 case NFS3_CREATE_GUARDED:
1414 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1422 nfsd_check_ignore_resizing(iap);
1424 if (nfsd_create_is_exclusive(createmode)) {
1425 /* Cram the verifier into atime/mtime */
1426 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1427 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1428 /* XXX someone who knows this better please fix it for nsec */
1429 iap->ia_mtime.tv_sec = v_mtime;
1430 iap->ia_atime.tv_sec = v_atime;
1431 iap->ia_mtime.tv_nsec = 0;
1432 iap->ia_atime.tv_nsec = 0;
1436 err = nfsd_create_setattr(rqstp, resfhp, iap);
1439 * nfsd_create_setattr already committed the child
1440 * (and possibly also the parent).
1443 err = nfserrno(commit_metadata(fhp));
1446 * Update the filehandle to get the new inode info.
1449 err = fh_update(resfhp);
1453 if (dchild && !IS_ERR(dchild))
1459 err = nfserrno(host_err);
1462 #endif /* CONFIG_NFSD_V3 */
1465 * Read a symlink. On entry, *lenp must contain the maximum path length that
1466 * fits into the buffer. On return, it contains the true length.
1467 * N.B. After this call fhp needs an fh_put
1470 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1472 struct inode *inode;
1478 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1482 path.mnt = fhp->fh_export->ex_path.mnt;
1483 path.dentry = fhp->fh_dentry;
1484 inode = d_inode(path.dentry);
1487 if (!inode->i_op->readlink)
1491 /* N.B. Why does this call need a get_fs()??
1492 * Remove the set_fs and watch the fireworks:-) --okir
1495 oldfs = get_fs(); set_fs(KERNEL_DS);
1496 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1507 err = nfserrno(host_err);
1512 * Create a symlink and look up its inode
1513 * N.B. After this call _both_ fhp and resfhp need an fh_put
1516 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1517 char *fname, int flen,
1519 struct svc_fh *resfhp)
1521 struct dentry *dentry, *dnew;
1526 if (!flen || path[0] == '\0')
1529 if (isdotent(fname, flen))
1532 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1536 host_err = fh_want_write(fhp);
1541 dentry = fhp->fh_dentry;
1542 dnew = lookup_one_len(fname, dentry, flen);
1543 host_err = PTR_ERR(dnew);
1547 host_err = vfs_symlink(d_inode(dentry), dnew, path);
1548 err = nfserrno(host_err);
1550 err = nfserrno(commit_metadata(fhp));
1555 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1557 if (err==0) err = cerr;
1562 err = nfserrno(host_err);
1568 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1571 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1572 char *name, int len, struct svc_fh *tfhp)
1574 struct dentry *ddir, *dnew, *dold;
1579 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1582 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1586 if (d_is_dir(tfhp->fh_dentry))
1592 if (isdotent(name, len))
1595 host_err = fh_want_write(tfhp);
1597 err = nfserrno(host_err);
1601 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1602 ddir = ffhp->fh_dentry;
1603 dirp = d_inode(ddir);
1605 dnew = lookup_one_len(name, ddir, len);
1606 host_err = PTR_ERR(dnew);
1610 dold = tfhp->fh_dentry;
1613 if (d_really_is_negative(dold))
1615 host_err = vfs_link(dold, dirp, dnew, NULL);
1617 err = nfserrno(commit_metadata(ffhp));
1619 err = nfserrno(commit_metadata(tfhp));
1621 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1624 err = nfserrno(host_err);
1630 fh_drop_write(tfhp);
1635 err = nfserrno(host_err);
1641 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1644 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1645 struct svc_fh *tfhp, char *tname, int tlen)
1647 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1648 struct inode *fdir, *tdir;
1652 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1655 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1659 fdentry = ffhp->fh_dentry;
1660 fdir = d_inode(fdentry);
1662 tdentry = tfhp->fh_dentry;
1663 tdir = d_inode(tdentry);
1666 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1669 host_err = fh_want_write(ffhp);
1671 err = nfserrno(host_err);
1675 /* cannot use fh_lock as we need deadlock protective ordering
1676 * so do it by hand */
1677 trap = lock_rename(tdentry, fdentry);
1678 ffhp->fh_locked = tfhp->fh_locked = true;
1682 odentry = lookup_one_len(fname, fdentry, flen);
1683 host_err = PTR_ERR(odentry);
1684 if (IS_ERR(odentry))
1688 if (d_really_is_negative(odentry))
1691 if (odentry == trap)
1694 ndentry = lookup_one_len(tname, tdentry, tlen);
1695 host_err = PTR_ERR(ndentry);
1696 if (IS_ERR(ndentry))
1698 host_err = -ENOTEMPTY;
1699 if (ndentry == trap)
1703 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1705 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1708 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
1710 host_err = commit_metadata(tfhp);
1712 host_err = commit_metadata(ffhp);
1719 err = nfserrno(host_err);
1721 * We cannot rely on fh_unlock on the two filehandles,
1722 * as that would do the wrong thing if the two directories
1723 * were the same, so again we do it by hand.
1725 fill_post_wcc(ffhp);
1726 fill_post_wcc(tfhp);
1727 unlock_rename(tdentry, fdentry);
1728 ffhp->fh_locked = tfhp->fh_locked = false;
1729 fh_drop_write(ffhp);
1736 * Unlink a file or directory
1737 * N.B. After this call fhp needs an fh_put
1740 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1741 char *fname, int flen)
1743 struct dentry *dentry, *rdentry;
1749 if (!flen || isdotent(fname, flen))
1751 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1755 host_err = fh_want_write(fhp);
1759 fh_lock_nested(fhp, I_MUTEX_PARENT);
1760 dentry = fhp->fh_dentry;
1761 dirp = d_inode(dentry);
1763 rdentry = lookup_one_len(fname, dentry, flen);
1764 host_err = PTR_ERR(rdentry);
1765 if (IS_ERR(rdentry))
1768 if (d_really_is_negative(rdentry)) {
1775 type = d_inode(rdentry)->i_mode & S_IFMT;
1777 if (type != S_IFDIR)
1778 host_err = vfs_unlink(dirp, rdentry, NULL);
1780 host_err = vfs_rmdir(dirp, rdentry);
1782 host_err = commit_metadata(fhp);
1786 err = nfserrno(host_err);
1792 * We do this buffering because we must not call back into the file
1793 * system's ->lookup() method from the filldir callback. That may well
1794 * deadlock a number of file systems.
1796 * This is based heavily on the implementation of same in XFS.
1798 struct buffered_dirent {
1802 unsigned int d_type;
1806 struct readdir_data {
1807 struct dir_context ctx;
1813 static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name,
1814 int namlen, loff_t offset, u64 ino,
1815 unsigned int d_type)
1817 struct readdir_data *buf =
1818 container_of(ctx, struct readdir_data, ctx);
1819 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1820 unsigned int reclen;
1822 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1823 if (buf->used + reclen > PAGE_SIZE) {
1828 de->namlen = namlen;
1829 de->offset = offset;
1831 de->d_type = d_type;
1832 memcpy(de->name, name, namlen);
1833 buf->used += reclen;
1838 static __be32 nfsd_buffered_readdir(struct file *file, nfsd_filldir_t func,
1839 struct readdir_cd *cdp, loff_t *offsetp)
1841 struct buffered_dirent *de;
1845 struct readdir_data buf = {
1846 .ctx.actor = nfsd_buffered_filldir,
1847 .dirent = (void *)__get_free_page(GFP_KERNEL)
1851 return nfserrno(-ENOMEM);
1856 struct inode *dir_inode = file_inode(file);
1857 unsigned int reclen;
1859 cdp->err = nfserr_eof; /* will be cleared on successful read */
1863 host_err = iterate_dir(file, &buf.ctx);
1876 * Various filldir functions may end up calling back into
1877 * lookup_one_len() and the file system's ->lookup() method.
1878 * These expect i_mutex to be held, as it would within readdir.
1880 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1884 de = (struct buffered_dirent *)buf.dirent;
1886 offset = de->offset;
1888 if (func(cdp, de->name, de->namlen, de->offset,
1889 de->ino, de->d_type))
1892 if (cdp->err != nfs_ok)
1895 reclen = ALIGN(sizeof(*de) + de->namlen,
1898 de = (struct buffered_dirent *)((char *)de + reclen);
1900 mutex_unlock(&dir_inode->i_mutex);
1901 if (size > 0) /* We bailed out early */
1904 offset = vfs_llseek(file, 0, SEEK_CUR);
1907 free_page((unsigned long)(buf.dirent));
1910 return nfserrno(host_err);
1917 * Read entries from a directory.
1918 * The NFSv3/4 verifier we ignore for now.
1921 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1922 struct readdir_cd *cdp, nfsd_filldir_t func)
1926 loff_t offset = *offsetp;
1927 int may_flags = NFSD_MAY_READ;
1929 /* NFSv2 only supports 32 bit cookies */
1930 if (rqstp->rq_vers > 2)
1931 may_flags |= NFSD_MAY_64BIT_COOKIE;
1933 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1937 offset = vfs_llseek(file, offset, SEEK_SET);
1939 err = nfserrno((int)offset);
1943 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1945 if (err == nfserr_eof || err == nfserr_toosmall)
1946 err = nfs_ok; /* can still be found in ->err */
1954 * Get file system stats
1955 * N.B. After this call fhp needs an fh_put
1958 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1962 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1964 struct path path = {
1965 .mnt = fhp->fh_export->ex_path.mnt,
1966 .dentry = fhp->fh_dentry,
1968 if (vfs_statfs(&path, stat))
1974 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1976 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1980 * Check for a user's access permissions to this inode.
1983 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1984 struct dentry *dentry, int acc)
1986 struct inode *inode = d_inode(dentry);
1989 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
1992 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1994 (acc & NFSD_MAY_READ)? " read" : "",
1995 (acc & NFSD_MAY_WRITE)? " write" : "",
1996 (acc & NFSD_MAY_EXEC)? " exec" : "",
1997 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1998 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1999 (acc & NFSD_MAY_LOCK)? " lock" : "",
2000 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2002 IS_IMMUTABLE(inode)? " immut" : "",
2003 IS_APPEND(inode)? " append" : "",
2004 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2005 dprintk(" owner %d/%d user %d/%d\n",
2006 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2009 /* Normally we reject any write/sattr etc access on a read-only file
2010 * system. But if it is IRIX doing check on write-access for a
2011 * device special file, we ignore rofs.
2013 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2014 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2015 if (exp_rdonly(rqstp, exp) ||
2016 __mnt_is_readonly(exp->ex_path.mnt))
2018 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2021 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2024 if (acc & NFSD_MAY_LOCK) {
2025 /* If we cannot rely on authentication in NLM requests,
2026 * just allow locks, otherwise require read permission, or
2029 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2032 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2035 * The file owner always gets access permission for accesses that
2036 * would normally be checked at open time. This is to make
2037 * file access work even when the client has done a fchmod(fd, 0).
2039 * However, `cp foo bar' should fail nevertheless when bar is
2040 * readonly. A sensible way to do this might be to reject all
2041 * attempts to truncate a read-only file, because a creat() call
2042 * always implies file truncation.
2043 * ... but this isn't really fair. A process may reasonably call
2044 * ftruncate on an open file descriptor on a file with perm 000.
2045 * We must trust the client to do permission checking - using "ACCESS"
2048 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2049 uid_eq(inode->i_uid, current_fsuid()))
2052 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2053 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2055 /* Allow read access to binaries even when mode 111 */
2056 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2057 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2058 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2059 err = inode_permission(inode, MAY_EXEC);
2061 return err? nfserrno(err) : 0;
2065 nfsd_racache_shutdown(void)
2067 struct raparms *raparm, *last_raparm;
2070 dprintk("nfsd: freeing readahead buffers.\n");
2072 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2073 raparm = raparm_hash[i].pb_head;
2075 last_raparm = raparm;
2076 raparm = raparm->p_next;
2079 raparm_hash[i].pb_head = NULL;
2083 * Initialize readahead param cache
2086 nfsd_racache_init(int cache_size)
2091 struct raparms **raparm = NULL;
2094 if (raparm_hash[0].pb_head)
2096 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2097 nperbucket = max(2, nperbucket);
2098 cache_size = nperbucket * RAPARM_HASH_SIZE;
2100 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2102 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2103 spin_lock_init(&raparm_hash[i].pb_lock);
2105 raparm = &raparm_hash[i].pb_head;
2106 for (j = 0; j < nperbucket; j++) {
2107 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2110 raparm = &(*raparm)->p_next;
2115 nfsdstats.ra_size = cache_size;
2119 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2120 nfsd_racache_shutdown();