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/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms *p_next;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
64 struct raparm_hbucket {
65 struct raparms *pb_head;
67 } ____cacheline_aligned_in_smp;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
90 err = follow_down(&path);
94 exp2 = rqst_exp_get_by_name(rqstp, &path);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
104 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
109 if (nfsd_v4client(rqstp) ||
110 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
119 path.dentry = dentry;
129 static void follow_to_parent(struct path *path)
133 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 dp = dget_parent(path->dentry);
140 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142 struct svc_export *exp2;
143 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
144 .dentry = dget(dparent)};
146 follow_to_parent(&path);
148 exp2 = rqst_exp_parent(rqstp, &path);
149 if (PTR_ERR(exp2) == -ENOENT) {
150 *dentryp = dget(dparent);
151 } else if (IS_ERR(exp2)) {
153 return PTR_ERR(exp2);
155 *dentryp = dget(path.dentry);
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
169 if (d_mountpoint(dentry))
171 if (nfsd4_is_junction(dentry))
173 if (!(exp->ex_flags & NFSEXP_V4ROOT))
175 return dentry->d_inode != NULL;
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180 const char *name, unsigned int len,
181 struct svc_export **exp_ret, struct dentry **dentry_ret)
183 struct svc_export *exp;
184 struct dentry *dparent;
185 struct dentry *dentry;
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
190 dparent = fhp->fh_dentry;
191 exp = fhp->fh_export;
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
197 dentry = dget(dparent);
198 else if (dparent != exp->ex_path.dentry)
199 dentry = dget_parent(dparent);
200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
201 dentry = dget(dparent); /* .. == . just like at / */
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
210 dentry = lookup_one_len(name, dparent, len);
211 host_err = PTR_ERR(dentry);
215 * check if we have crossed a mount point ...
217 if (nfsd_mountpoint(dentry, exp)) {
218 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
224 *dentry_ret = dentry;
230 return nfserrno(host_err);
234 * Look up one component of a pathname.
235 * N.B. After this call _both_ fhp and resfh need an fh_put
237 * If the lookup would cross a mountpoint, and the mounted filesystem
238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239 * accepted as it stands and the mounted directory is
240 * returned. Otherwise the covered directory is returned.
241 * NOTE: this mountpoint crossing is not supported properly by all
242 * clients and is explicitly disallowed for NFSv3
243 * NeilBrown <neilb@cse.unsw.edu.au>
246 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
247 unsigned int len, struct svc_fh *resfh)
249 struct svc_export *exp;
250 struct dentry *dentry;
253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
259 err = check_nfsd_access(exp, rqstp);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
275 static int nfsd_break_lease(struct inode *inode)
277 if (!S_ISREG(inode->i_mode))
279 return break_lease(inode, O_WRONLY | O_NONBLOCK);
283 * Commit metadata changes to stable storage.
286 commit_metadata(struct svc_fh *fhp)
288 struct inode *inode = fhp->fh_dentry->d_inode;
289 const struct export_operations *export_ops = inode->i_sb->s_export_op;
291 if (!EX_ISSYNC(fhp->fh_export))
294 if (export_ops->commit_metadata)
295 return export_ops->commit_metadata(inode);
296 return sync_inode_metadata(inode, 1);
300 * Go over the attributes and take care of the small differences between
301 * NFS semantics and what Linux expects.
304 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
307 * NFSv2 does not differentiate between "set-[ac]time-to-now"
308 * which only requires access, and "set-[ac]time-to-X" which
309 * requires ownership.
310 * So if it looks like it might be "set both to the same time which
311 * is close to now", and if inode_change_ok fails, then we
312 * convert to "set to now" instead of "set to explicit time"
314 * We only call inode_change_ok as the last test as technically
315 * it is not an interface that we should be using.
317 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
318 #define MAX_TOUCH_TIME_ERROR (30*60)
319 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
320 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
324 * Now just make sure time is in the right ballpark.
325 * Solaris, at least, doesn't seem to care what the time
326 * request is. We require it be within 30 minutes of now.
328 time_t delta = iap->ia_atime.tv_sec - get_seconds();
331 if (delta < MAX_TOUCH_TIME_ERROR &&
332 inode_change_ok(inode, iap) != 0) {
334 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
335 * This will cause notify_change to set these times
338 iap->ia_valid &= ~BOTH_TIME_SET;
342 /* sanitize the mode change */
343 if (iap->ia_valid & ATTR_MODE) {
344 iap->ia_mode &= S_IALLUGO;
345 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
348 /* Revoke setuid/setgid on chown */
349 if (!S_ISDIR(inode->i_mode) &&
350 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
351 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
352 iap->ia_valid |= ATTR_KILL_PRIV;
353 if (iap->ia_valid & ATTR_MODE) {
354 /* we're setting mode too, just clear the s*id bits */
355 iap->ia_mode &= ~S_ISUID;
356 if (iap->ia_mode & S_IXGRP)
357 iap->ia_mode &= ~S_ISGID;
359 /* set ATTR_KILL_* bits and let VFS handle it */
360 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
366 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
369 struct inode *inode = fhp->fh_dentry->d_inode;
372 if (iap->ia_size < inode->i_size) {
375 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
376 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
381 host_err = get_write_access(inode);
385 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
387 goto out_put_write_access;
390 out_put_write_access:
391 put_write_access(inode);
393 return nfserrno(host_err);
397 * Set various file attributes. After this call fhp needs an fh_put.
400 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
401 int check_guard, time_t guardtime)
403 struct dentry *dentry;
405 int accmode = NFSD_MAY_SATTR;
409 bool get_write_count;
412 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
413 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
414 if (iap->ia_valid & ATTR_SIZE)
417 /* Callers that do fh_verify should do the fh_want_write: */
418 get_write_count = !fhp->fh_dentry;
421 err = fh_verify(rqstp, fhp, ftype, accmode);
424 if (get_write_count) {
425 host_err = fh_want_write(fhp);
427 return nfserrno(host_err);
430 dentry = fhp->fh_dentry;
431 inode = dentry->d_inode;
433 /* Ignore any mode updates on symlinks */
434 if (S_ISLNK(inode->i_mode))
435 iap->ia_valid &= ~ATTR_MODE;
440 nfsd_sanitize_attrs(inode, iap);
443 * The size case is special, it changes the file in addition to the
446 if (iap->ia_valid & ATTR_SIZE) {
447 err = nfsd_get_write_access(rqstp, fhp, iap);
453 iap->ia_valid |= ATTR_CTIME;
455 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
456 err = nfserr_notsync;
457 goto out_put_write_access;
460 host_err = nfsd_break_lease(inode);
462 goto out_put_write_access_nfserror;
465 host_err = notify_change(dentry, iap);
468 out_put_write_access_nfserror:
469 err = nfserrno(host_err);
470 out_put_write_access:
472 put_write_access(inode);
474 commit_metadata(fhp);
479 #if defined(CONFIG_NFSD_V2_ACL) || \
480 defined(CONFIG_NFSD_V3_ACL) || \
481 defined(CONFIG_NFSD_V4)
482 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
487 buflen = vfs_getxattr(dentry, key, NULL, 0);
491 *buf = kmalloc(buflen, GFP_KERNEL);
495 ret = vfs_getxattr(dentry, key, *buf, buflen);
502 #if defined(CONFIG_NFSD_V4)
504 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
512 return vfs_setxattr(dentry, key, NULL, 0, 0);
514 buflen = posix_acl_xattr_size(pacl->a_count);
515 buf = kmalloc(buflen, GFP_KERNEL);
520 len = posix_acl_to_xattr(pacl, buf, buflen);
526 error = vfs_setxattr(dentry, key, buf, len, 0);
533 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
534 struct nfs4_acl *acl)
538 struct dentry *dentry;
540 struct posix_acl *pacl = NULL, *dpacl = NULL;
541 unsigned int flags = 0;
544 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
548 dentry = fhp->fh_dentry;
549 inode = dentry->d_inode;
550 if (S_ISDIR(inode->i_mode))
551 flags = NFS4_ACL_DIR;
553 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
554 if (host_error == -EINVAL) {
555 return nfserr_attrnotsupp;
556 } else if (host_error < 0)
559 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
563 if (S_ISDIR(inode->i_mode))
564 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
567 posix_acl_release(pacl);
568 posix_acl_release(dpacl);
570 if (host_error == -EOPNOTSUPP)
571 return nfserr_attrnotsupp;
573 return nfserrno(host_error);
576 static struct posix_acl *
577 _get_posix_acl(struct dentry *dentry, char *key)
580 struct posix_acl *pacl = NULL;
583 buflen = nfsd_getxattr(dentry, key, &buf);
587 return ERR_PTR(buflen);
589 pacl = posix_acl_from_xattr(buf, buflen);
595 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
597 struct inode *inode = dentry->d_inode;
599 struct posix_acl *pacl = NULL, *dpacl = NULL;
600 unsigned int flags = 0;
602 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
603 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
604 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
606 error = PTR_ERR(pacl);
611 if (S_ISDIR(inode->i_mode)) {
612 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
613 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
615 else if (IS_ERR(dpacl)) {
616 error = PTR_ERR(dpacl);
620 flags = NFS4_ACL_DIR;
623 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
625 error = PTR_ERR(*acl);
629 posix_acl_release(pacl);
630 posix_acl_release(dpacl);
634 #define NFSD_XATTR_JUNCTION_PREFIX XATTR_TRUSTED_PREFIX "junction."
635 #define NFSD_XATTR_JUNCTION_TYPE NFSD_XATTR_JUNCTION_PREFIX "type"
636 int nfsd4_is_junction(struct dentry *dentry)
638 struct inode *inode = dentry->d_inode;
642 if (inode->i_mode & S_IXUGO)
644 if (!(inode->i_mode & S_ISVTX))
646 if (vfs_getxattr(dentry, NFSD_XATTR_JUNCTION_TYPE, NULL, 0) <= 0)
650 #endif /* defined(CONFIG_NFSD_V4) */
652 #ifdef CONFIG_NFSD_V3
654 * Check server access rights to a file system object
660 static struct accessmap nfs3_regaccess[] = {
661 { NFS3_ACCESS_READ, NFSD_MAY_READ },
662 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
663 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
664 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
669 static struct accessmap nfs3_diraccess[] = {
670 { NFS3_ACCESS_READ, NFSD_MAY_READ },
671 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
672 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
673 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
674 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
679 static struct accessmap nfs3_anyaccess[] = {
680 /* Some clients - Solaris 2.6 at least, make an access call
681 * to the server to check for access for things like /dev/null
682 * (which really, the server doesn't care about). So
683 * We provide simple access checking for them, looking
684 * mainly at mode bits, and we make sure to ignore read-only
687 { NFS3_ACCESS_READ, NFSD_MAY_READ },
688 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
689 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
690 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
696 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
698 struct accessmap *map;
699 struct svc_export *export;
700 struct dentry *dentry;
701 u32 query, result = 0, sresult = 0;
704 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
708 export = fhp->fh_export;
709 dentry = fhp->fh_dentry;
711 if (S_ISREG(dentry->d_inode->i_mode))
712 map = nfs3_regaccess;
713 else if (S_ISDIR(dentry->d_inode->i_mode))
714 map = nfs3_diraccess;
716 map = nfs3_anyaccess;
720 for (; map->access; map++) {
721 if (map->access & query) {
724 sresult |= map->access;
726 err2 = nfsd_permission(rqstp, export, dentry, map->how);
729 result |= map->access;
732 /* the following error codes just mean the access was not allowed,
733 * rather than an error occurred */
737 /* simply don't "or" in the access bit. */
747 *supported = sresult;
752 #endif /* CONFIG_NFSD_V3 */
754 static int nfsd_open_break_lease(struct inode *inode, int access)
758 if (access & NFSD_MAY_NOT_BREAK_LEASE)
760 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
761 return break_lease(inode, mode | O_NONBLOCK);
765 * Open an existing file or directory.
766 * The may_flags argument indicates the type of open (read/write/lock)
767 * and additional flags.
768 * N.B. After this call fhp needs an fh_put
771 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
772 int may_flags, struct file **filp)
774 struct dentry *dentry;
776 int flags = O_RDONLY|O_LARGEFILE;
780 validate_process_creds();
783 * If we get here, then the client has already done an "open",
784 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
785 * in case a chmod has now revoked permission.
787 err = fh_verify(rqstp, fhp, type, may_flags | NFSD_MAY_OWNER_OVERRIDE);
791 dentry = fhp->fh_dentry;
792 inode = dentry->d_inode;
794 /* Disallow write access to files with the append-only bit set
795 * or any access when mandatory locking enabled
798 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
801 * We must ignore files (but only files) which might have mandatory
802 * locks on them because there is no way to know if the accesser has
805 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
811 host_err = nfsd_open_break_lease(inode, may_flags);
812 if (host_err) /* NOMEM or WOULDBLOCK */
815 if (may_flags & NFSD_MAY_WRITE) {
816 if (may_flags & NFSD_MAY_READ)
817 flags = O_RDWR|O_LARGEFILE;
819 flags = O_WRONLY|O_LARGEFILE;
821 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
822 flags, current_cred());
824 host_err = PTR_ERR(*filp);
827 host_err = ima_file_check(*filp, may_flags);
829 if (may_flags & NFSD_MAY_64BIT_COOKIE)
830 (*filp)->f_mode |= FMODE_64BITHASH;
832 (*filp)->f_mode |= FMODE_32BITHASH;
836 err = nfserrno(host_err);
838 validate_process_creds();
846 nfsd_close(struct file *filp)
852 * Obtain the readahead parameters for the file
853 * specified by (dev, ino).
856 static inline struct raparms *
857 nfsd_get_raparms(dev_t dev, ino_t ino)
859 struct raparms *ra, **rap, **frap = NULL;
862 struct raparm_hbucket *rab;
864 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
865 rab = &raparm_hash[hash];
867 spin_lock(&rab->pb_lock);
868 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
869 if (ra->p_ino == ino && ra->p_dev == dev)
872 if (ra->p_count == 0)
875 depth = nfsdstats.ra_size;
877 spin_unlock(&rab->pb_lock);
887 if (rap != &rab->pb_head) {
889 ra->p_next = rab->pb_head;
893 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
894 spin_unlock(&rab->pb_lock);
899 * Grab and keep cached pages associated with a file in the svc_rqst
900 * so that they can be passed to the network sendmsg/sendpage routines
901 * directly. They will be released after the sending has completed.
904 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
905 struct splice_desc *sd)
907 struct svc_rqst *rqstp = sd->u.data;
908 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
909 struct page *page = buf->page;
914 if (rqstp->rq_res.page_len == 0) {
919 rqstp->rq_res.page_base = buf->offset;
920 rqstp->rq_res.page_len = size;
921 } else if (page != pp[-1]) {
927 rqstp->rq_res.page_len += size;
929 rqstp->rq_res.page_len += size;
934 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
935 struct splice_desc *sd)
937 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
941 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
942 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
950 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
951 struct splice_desc sd = {
958 rqstp->rq_resused = 1;
959 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
963 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
968 nfsdstats.io_read += host_err;
971 fsnotify_access(file);
973 err = nfserrno(host_err);
977 static void kill_suid(struct dentry *dentry)
980 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
982 mutex_lock(&dentry->d_inode->i_mutex);
983 notify_change(dentry, &ia);
984 mutex_unlock(&dentry->d_inode->i_mutex);
988 * Gathered writes: If another process is currently writing to the file,
989 * there's a high chance this is another nfsd (triggered by a bulk write
990 * from a client's biod). Rather than syncing the file with each write
991 * request, we sleep for 10 msec.
993 * I don't know if this roughly approximates C. Juszak's idea of
994 * gathered writes, but it's a nice and simple solution (IMHO), and it
997 * Note: we do this only in the NFSv2 case, since v3 and higher have a
998 * better tool (separate unstable writes and commits) for solving this
1001 static int wait_for_concurrent_writes(struct file *file)
1003 struct inode *inode = file->f_path.dentry->d_inode;
1004 static ino_t last_ino;
1005 static dev_t last_dev;
1008 if (atomic_read(&inode->i_writecount) > 1
1009 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1010 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1012 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1015 if (inode->i_state & I_DIRTY) {
1016 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1017 err = vfs_fsync(file, 0);
1019 last_ino = inode->i_ino;
1020 last_dev = inode->i_sb->s_dev;
1025 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1026 loff_t offset, struct kvec *vec, int vlen,
1027 unsigned long *cnt, int *stablep)
1029 struct svc_export *exp;
1030 struct dentry *dentry;
1031 struct inode *inode;
1035 int stable = *stablep;
1038 dentry = file->f_path.dentry;
1039 inode = dentry->d_inode;
1040 exp = fhp->fh_export;
1043 * Request sync writes if
1044 * - the sync export option has been set, or
1045 * - the client requested O_SYNC behavior (NFSv3 feature).
1046 * - The file system doesn't support fsync().
1047 * When NFSv2 gathered writes have been configured for this volume,
1048 * flushing the data to disk is handled separately below.
1050 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1052 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1054 *stablep = 2; /* FILE_SYNC */
1057 if (!EX_ISSYNC(exp))
1059 if (stable && !use_wgather) {
1060 spin_lock(&file->f_lock);
1061 file->f_flags |= O_SYNC;
1062 spin_unlock(&file->f_lock);
1065 /* Write the data. */
1066 oldfs = get_fs(); set_fs(KERNEL_DS);
1067 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1072 nfsdstats.io_write += host_err;
1073 fsnotify_modify(file);
1075 /* clear setuid/setgid flag after write */
1076 if (inode->i_mode & (S_ISUID | S_ISGID))
1079 if (stable && use_wgather)
1080 host_err = wait_for_concurrent_writes(file);
1083 dprintk("nfsd: write complete host_err=%d\n", host_err);
1087 err = nfserrno(host_err);
1092 * Read data from a file. count must contain the requested read count
1093 * on entry. On return, *count contains the number of bytes actually read.
1094 * N.B. After this call fhp needs an fh_put
1096 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1097 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1100 struct inode *inode;
1104 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1108 inode = file->f_path.dentry->d_inode;
1110 /* Get readahead parameters */
1111 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1113 if (ra && ra->p_set)
1114 file->f_ra = ra->p_ra;
1116 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1118 /* Write back readahead params */
1120 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1121 spin_lock(&rab->pb_lock);
1122 ra->p_ra = file->f_ra;
1125 spin_unlock(&rab->pb_lock);
1132 /* As above, but use the provided file descriptor. */
1134 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1135 loff_t offset, struct kvec *vec, int vlen,
1136 unsigned long *count)
1141 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1142 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1145 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1146 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1147 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1153 * Write data to a file.
1154 * The stable flag requests synchronous writes.
1155 * N.B. After this call fhp needs an fh_put
1158 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1159 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1165 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1166 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1169 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1172 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1177 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1185 #ifdef CONFIG_NFSD_V3
1187 * Commit all pending writes to stable storage.
1189 * Note: we only guarantee that data that lies within the range specified
1190 * by the 'offset' and 'count' parameters will be synced.
1192 * Unfortunately we cannot lock the file to make sure we return full WCC
1193 * data to the client, as locking happens lower down in the filesystem.
1196 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1197 loff_t offset, unsigned long count)
1200 loff_t end = LLONG_MAX;
1201 __be32 err = nfserr_inval;
1206 end = offset + (loff_t)count - 1;
1211 err = nfsd_open(rqstp, fhp, S_IFREG,
1212 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1215 if (EX_ISSYNC(fhp->fh_export)) {
1216 int err2 = vfs_fsync_range(file, offset, end, 0);
1218 if (err2 != -EINVAL)
1219 err = nfserrno(err2);
1221 err = nfserr_notsupp;
1228 #endif /* CONFIG_NFSD_V3 */
1231 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1235 * Mode has already been set earlier in create:
1237 iap->ia_valid &= ~ATTR_MODE;
1239 * Setting uid/gid works only for root. Irix appears to
1240 * send along the gid on create when it tries to implement
1241 * setgid directories via NFS:
1243 if (current_fsuid() != 0)
1244 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1246 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1250 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1251 * setting size to 0 may fail for some specific file systems by the permission
1252 * checking which requires WRITE permission but the mode is 000.
1253 * we ignore the resizing(to 0) on the just new created file, since the size is
1254 * 0 after file created.
1256 * call this only after vfs_create() is called.
1259 nfsd_check_ignore_resizing(struct iattr *iap)
1261 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1262 iap->ia_valid &= ~ATTR_SIZE;
1266 * Create a file (regular, directory, device, fifo); UNIX sockets
1267 * not yet implemented.
1268 * If the response fh has been verified, the parent directory should
1269 * already be locked. Note that the parent directory is left locked.
1271 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1274 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1275 char *fname, int flen, struct iattr *iap,
1276 int type, dev_t rdev, struct svc_fh *resfhp)
1278 struct dentry *dentry, *dchild = NULL;
1288 if (isdotent(fname, flen))
1291 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1295 dentry = fhp->fh_dentry;
1296 dirp = dentry->d_inode;
1298 err = nfserr_notdir;
1299 if (!dirp->i_op->lookup)
1302 * Check whether the response file handle has been verified yet.
1303 * If it has, the parent directory should already be locked.
1305 if (!resfhp->fh_dentry) {
1306 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1307 fh_lock_nested(fhp, I_MUTEX_PARENT);
1308 dchild = lookup_one_len(fname, dentry, flen);
1309 host_err = PTR_ERR(dchild);
1312 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1316 /* called from nfsd_proc_create */
1317 dchild = dget(resfhp->fh_dentry);
1318 if (!fhp->fh_locked) {
1319 /* not actually possible */
1321 "nfsd_create: parent %s/%s not locked!\n",
1322 dentry->d_parent->d_name.name,
1323 dentry->d_name.name);
1329 * Make sure the child dentry is still negative ...
1332 if (dchild->d_inode) {
1333 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1334 dentry->d_name.name, dchild->d_name.name);
1338 if (!(iap->ia_valid & ATTR_MODE))
1340 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1343 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1344 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1349 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1354 * Get the dir op function pointer.
1359 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1361 nfsd_check_ignore_resizing(iap);
1364 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1370 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1374 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1378 err = nfsd_create_setattr(rqstp, resfhp, iap);
1381 * nfsd_setattr already committed the child. Transactional filesystems
1382 * had a chance to commit changes for both parent and child
1383 * simultaneously making the following commit_metadata a noop.
1385 err2 = nfserrno(commit_metadata(fhp));
1388 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1390 * Update the file handle to get the new inode info.
1393 err = fh_update(resfhp);
1395 if (dchild && !IS_ERR(dchild))
1400 err = nfserrno(host_err);
1404 #ifdef CONFIG_NFSD_V3
1406 static inline int nfsd_create_is_exclusive(int createmode)
1408 return createmode == NFS3_CREATE_EXCLUSIVE
1409 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1413 * NFSv3 and NFSv4 version of nfsd_create
1416 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1417 char *fname, int flen, struct iattr *iap,
1418 struct svc_fh *resfhp, int createmode, u32 *verifier,
1419 bool *truncp, bool *created)
1421 struct dentry *dentry, *dchild = NULL;
1425 __u32 v_mtime=0, v_atime=0;
1431 if (isdotent(fname, flen))
1433 if (!(iap->ia_valid & ATTR_MODE))
1435 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1439 dentry = fhp->fh_dentry;
1440 dirp = dentry->d_inode;
1442 /* Get all the sanity checks out of the way before
1443 * we lock the parent. */
1444 err = nfserr_notdir;
1445 if (!dirp->i_op->lookup)
1447 fh_lock_nested(fhp, I_MUTEX_PARENT);
1450 * Compose the response file handle.
1452 dchild = lookup_one_len(fname, dentry, flen);
1453 host_err = PTR_ERR(dchild);
1457 /* If file doesn't exist, check for permissions to create one */
1458 if (!dchild->d_inode) {
1459 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1464 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1468 if (nfsd_create_is_exclusive(createmode)) {
1469 /* solaris7 gets confused (bugid 4218508) if these have
1470 * the high bit set, so just clear the high bits. If this is
1471 * ever changed to use different attrs for storing the
1472 * verifier, then do_open_lookup() will also need to be fixed
1475 v_mtime = verifier[0]&0x7fffffff;
1476 v_atime = verifier[1]&0x7fffffff;
1479 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1482 if (dchild->d_inode) {
1485 switch (createmode) {
1486 case NFS3_CREATE_UNCHECKED:
1487 if (! S_ISREG(dchild->d_inode->i_mode))
1490 /* in nfsv4, we need to treat this case a little
1491 * differently. we don't want to truncate the
1492 * file now; this would be wrong if the OPEN
1493 * fails for some other reason. furthermore,
1494 * if the size is nonzero, we should ignore it
1495 * according to spec!
1497 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1500 iap->ia_valid &= ATTR_SIZE;
1504 case NFS3_CREATE_EXCLUSIVE:
1505 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1506 && dchild->d_inode->i_atime.tv_sec == v_atime
1507 && dchild->d_inode->i_size == 0 ) {
1512 case NFS4_CREATE_EXCLUSIVE4_1:
1513 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1514 && dchild->d_inode->i_atime.tv_sec == v_atime
1515 && dchild->d_inode->i_size == 0 ) {
1521 case NFS3_CREATE_GUARDED:
1524 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1528 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1530 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1536 nfsd_check_ignore_resizing(iap);
1538 if (nfsd_create_is_exclusive(createmode)) {
1539 /* Cram the verifier into atime/mtime */
1540 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1541 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1542 /* XXX someone who knows this better please fix it for nsec */
1543 iap->ia_mtime.tv_sec = v_mtime;
1544 iap->ia_atime.tv_sec = v_atime;
1545 iap->ia_mtime.tv_nsec = 0;
1546 iap->ia_atime.tv_nsec = 0;
1550 err = nfsd_create_setattr(rqstp, resfhp, iap);
1553 * nfsd_setattr already committed the child (and possibly also the parent).
1556 err = nfserrno(commit_metadata(fhp));
1558 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1560 * Update the filehandle to get the new inode info.
1563 err = fh_update(resfhp);
1567 if (dchild && !IS_ERR(dchild))
1572 err = nfserrno(host_err);
1575 #endif /* CONFIG_NFSD_V3 */
1578 * Read a symlink. On entry, *lenp must contain the maximum path length that
1579 * fits into the buffer. On return, it contains the true length.
1580 * N.B. After this call fhp needs an fh_put
1583 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1585 struct dentry *dentry;
1586 struct inode *inode;
1591 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1595 dentry = fhp->fh_dentry;
1596 inode = dentry->d_inode;
1599 if (!inode->i_op->readlink)
1602 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1603 /* N.B. Why does this call need a get_fs()??
1604 * Remove the set_fs and watch the fireworks:-) --okir
1607 oldfs = get_fs(); set_fs(KERNEL_DS);
1608 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1619 err = nfserrno(host_err);
1624 * Create a symlink and look up its inode
1625 * N.B. After this call _both_ fhp and resfhp need an fh_put
1628 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1629 char *fname, int flen,
1630 char *path, int plen,
1631 struct svc_fh *resfhp,
1634 struct dentry *dentry, *dnew;
1642 if (isdotent(fname, flen))
1645 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1649 dentry = fhp->fh_dentry;
1650 dnew = lookup_one_len(fname, dentry, flen);
1651 host_err = PTR_ERR(dnew);
1655 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1659 if (unlikely(path[plen] != 0)) {
1660 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1661 if (path_alloced == NULL)
1664 strncpy(path_alloced, path, plen);
1665 path_alloced[plen] = 0;
1666 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1667 kfree(path_alloced);
1670 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1671 err = nfserrno(host_err);
1673 err = nfserrno(commit_metadata(fhp));
1676 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1678 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1680 if (err==0) err = cerr;
1685 err = nfserrno(host_err);
1691 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1694 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1695 char *name, int len, struct svc_fh *tfhp)
1697 struct dentry *ddir, *dnew, *dold;
1702 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1705 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1709 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1715 if (isdotent(name, len))
1718 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1719 ddir = ffhp->fh_dentry;
1720 dirp = ddir->d_inode;
1722 dnew = lookup_one_len(name, ddir, len);
1723 host_err = PTR_ERR(dnew);
1727 dold = tfhp->fh_dentry;
1729 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1731 err = nfserrno(host_err);
1736 goto out_drop_write;
1737 host_err = nfsd_break_lease(dold->d_inode);
1739 err = nfserrno(host_err);
1740 goto out_drop_write;
1742 host_err = vfs_link(dold, dirp, dnew);
1744 err = nfserrno(commit_metadata(ffhp));
1746 err = nfserrno(commit_metadata(tfhp));
1748 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1751 err = nfserrno(host_err);
1754 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1763 err = nfserrno(host_err);
1769 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1772 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1773 struct svc_fh *tfhp, char *tname, int tlen)
1775 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1776 struct inode *fdir, *tdir;
1780 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1783 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1787 fdentry = ffhp->fh_dentry;
1788 fdir = fdentry->d_inode;
1790 tdentry = tfhp->fh_dentry;
1791 tdir = tdentry->d_inode;
1793 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1794 if (ffhp->fh_export != tfhp->fh_export)
1798 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1801 /* cannot use fh_lock as we need deadlock protective ordering
1802 * so do it by hand */
1803 trap = lock_rename(tdentry, fdentry);
1804 ffhp->fh_locked = tfhp->fh_locked = 1;
1808 odentry = lookup_one_len(fname, fdentry, flen);
1809 host_err = PTR_ERR(odentry);
1810 if (IS_ERR(odentry))
1814 if (!odentry->d_inode)
1817 if (odentry == trap)
1820 ndentry = lookup_one_len(tname, tdentry, tlen);
1821 host_err = PTR_ERR(ndentry);
1822 if (IS_ERR(ndentry))
1824 host_err = -ENOTEMPTY;
1825 if (ndentry == trap)
1829 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1831 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1835 host_err = nfsd_break_lease(odentry->d_inode);
1837 goto out_drop_write;
1838 if (ndentry->d_inode) {
1839 host_err = nfsd_break_lease(ndentry->d_inode);
1841 goto out_drop_write;
1843 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1845 host_err = commit_metadata(tfhp);
1847 host_err = commit_metadata(ffhp);
1850 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1856 err = nfserrno(host_err);
1858 /* we cannot reply on fh_unlock on the two filehandles,
1859 * as that would do the wrong thing if the two directories
1860 * were the same, so again we do it by hand
1862 fill_post_wcc(ffhp);
1863 fill_post_wcc(tfhp);
1864 unlock_rename(tdentry, fdentry);
1865 ffhp->fh_locked = tfhp->fh_locked = 0;
1872 * Unlink a file or directory
1873 * N.B. After this call fhp needs an fh_put
1876 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1877 char *fname, int flen)
1879 struct dentry *dentry, *rdentry;
1885 if (!flen || isdotent(fname, flen))
1887 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1891 fh_lock_nested(fhp, I_MUTEX_PARENT);
1892 dentry = fhp->fh_dentry;
1893 dirp = dentry->d_inode;
1895 rdentry = lookup_one_len(fname, dentry, flen);
1896 host_err = PTR_ERR(rdentry);
1897 if (IS_ERR(rdentry))
1900 if (!rdentry->d_inode) {
1907 type = rdentry->d_inode->i_mode & S_IFMT;
1909 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1913 host_err = nfsd_break_lease(rdentry->d_inode);
1915 goto out_drop_write;
1916 if (type != S_IFDIR)
1917 host_err = vfs_unlink(dirp, rdentry);
1919 host_err = vfs_rmdir(dirp, rdentry);
1921 host_err = commit_metadata(fhp);
1923 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1928 err = nfserrno(host_err);
1934 * We do this buffering because we must not call back into the file
1935 * system's ->lookup() method from the filldir callback. That may well
1936 * deadlock a number of file systems.
1938 * This is based heavily on the implementation of same in XFS.
1940 struct buffered_dirent {
1944 unsigned int d_type;
1948 struct readdir_data {
1954 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1955 loff_t offset, u64 ino, unsigned int d_type)
1957 struct readdir_data *buf = __buf;
1958 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1959 unsigned int reclen;
1961 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1962 if (buf->used + reclen > PAGE_SIZE) {
1967 de->namlen = namlen;
1968 de->offset = offset;
1970 de->d_type = d_type;
1971 memcpy(de->name, name, namlen);
1972 buf->used += reclen;
1977 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1978 struct readdir_cd *cdp, loff_t *offsetp)
1980 struct readdir_data buf;
1981 struct buffered_dirent *de;
1986 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1988 return nfserrno(-ENOMEM);
1993 struct inode *dir_inode = file->f_path.dentry->d_inode;
1994 unsigned int reclen;
1996 cdp->err = nfserr_eof; /* will be cleared on successful read */
2000 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2013 * Various filldir functions may end up calling back into
2014 * lookup_one_len() and the file system's ->lookup() method.
2015 * These expect i_mutex to be held, as it would within readdir.
2017 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2021 de = (struct buffered_dirent *)buf.dirent;
2023 offset = de->offset;
2025 if (func(cdp, de->name, de->namlen, de->offset,
2026 de->ino, de->d_type))
2029 if (cdp->err != nfs_ok)
2032 reclen = ALIGN(sizeof(*de) + de->namlen,
2035 de = (struct buffered_dirent *)((char *)de + reclen);
2037 mutex_unlock(&dir_inode->i_mutex);
2038 if (size > 0) /* We bailed out early */
2041 offset = vfs_llseek(file, 0, SEEK_CUR);
2044 free_page((unsigned long)(buf.dirent));
2047 return nfserrno(host_err);
2054 * Read entries from a directory.
2055 * The NFSv3/4 verifier we ignore for now.
2058 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2059 struct readdir_cd *cdp, filldir_t func)
2063 loff_t offset = *offsetp;
2064 int may_flags = NFSD_MAY_READ;
2066 /* NFSv2 only supports 32 bit cookies */
2067 if (rqstp->rq_vers > 2)
2068 may_flags |= NFSD_MAY_64BIT_COOKIE;
2070 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2074 offset = vfs_llseek(file, offset, 0);
2076 err = nfserrno((int)offset);
2080 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2082 if (err == nfserr_eof || err == nfserr_toosmall)
2083 err = nfs_ok; /* can still be found in ->err */
2091 * Get file system stats
2092 * N.B. After this call fhp needs an fh_put
2095 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2099 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2101 struct path path = {
2102 .mnt = fhp->fh_export->ex_path.mnt,
2103 .dentry = fhp->fh_dentry,
2105 if (vfs_statfs(&path, stat))
2111 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2113 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2117 * Check for a user's access permissions to this inode.
2120 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2121 struct dentry *dentry, int acc)
2123 struct inode *inode = dentry->d_inode;
2126 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2129 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2131 (acc & NFSD_MAY_READ)? " read" : "",
2132 (acc & NFSD_MAY_WRITE)? " write" : "",
2133 (acc & NFSD_MAY_EXEC)? " exec" : "",
2134 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2135 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2136 (acc & NFSD_MAY_LOCK)? " lock" : "",
2137 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2139 IS_IMMUTABLE(inode)? " immut" : "",
2140 IS_APPEND(inode)? " append" : "",
2141 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2142 dprintk(" owner %d/%d user %d/%d\n",
2143 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2146 /* Normally we reject any write/sattr etc access on a read-only file
2147 * system. But if it is IRIX doing check on write-access for a
2148 * device special file, we ignore rofs.
2150 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2151 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2152 if (exp_rdonly(rqstp, exp) ||
2153 __mnt_is_readonly(exp->ex_path.mnt))
2155 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2158 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2161 if (acc & NFSD_MAY_LOCK) {
2162 /* If we cannot rely on authentication in NLM requests,
2163 * just allow locks, otherwise require read permission, or
2166 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2169 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2172 * The file owner always gets access permission for accesses that
2173 * would normally be checked at open time. This is to make
2174 * file access work even when the client has done a fchmod(fd, 0).
2176 * However, `cp foo bar' should fail nevertheless when bar is
2177 * readonly. A sensible way to do this might be to reject all
2178 * attempts to truncate a read-only file, because a creat() call
2179 * always implies file truncation.
2180 * ... but this isn't really fair. A process may reasonably call
2181 * ftruncate on an open file descriptor on a file with perm 000.
2182 * We must trust the client to do permission checking - using "ACCESS"
2185 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2186 inode->i_uid == current_fsuid())
2189 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2190 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2192 /* Allow read access to binaries even when mode 111 */
2193 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2194 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2195 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2196 err = inode_permission(inode, MAY_EXEC);
2198 return err? nfserrno(err) : 0;
2202 nfsd_racache_shutdown(void)
2204 struct raparms *raparm, *last_raparm;
2207 dprintk("nfsd: freeing readahead buffers.\n");
2209 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2210 raparm = raparm_hash[i].pb_head;
2212 last_raparm = raparm;
2213 raparm = raparm->p_next;
2216 raparm_hash[i].pb_head = NULL;
2220 * Initialize readahead param cache
2223 nfsd_racache_init(int cache_size)
2228 struct raparms **raparm = NULL;
2231 if (raparm_hash[0].pb_head)
2233 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2236 cache_size = nperbucket * RAPARM_HASH_SIZE;
2238 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2240 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2241 spin_lock_init(&raparm_hash[i].pb_lock);
2243 raparm = &raparm_hash[i].pb_head;
2244 for (j = 0; j < nperbucket; j++) {
2245 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2248 raparm = &(*raparm)->p_next;
2253 nfsdstats.ra_size = cache_size;
2257 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2258 nfsd_racache_shutdown();
2262 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2264 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2266 struct inode *inode = fhp->fh_dentry->d_inode;
2270 struct posix_acl *acl;
2272 if (!IS_POSIXACL(inode))
2273 return ERR_PTR(-EOPNOTSUPP);
2276 case ACL_TYPE_ACCESS:
2277 name = POSIX_ACL_XATTR_ACCESS;
2279 case ACL_TYPE_DEFAULT:
2280 name = POSIX_ACL_XATTR_DEFAULT;
2283 return ERR_PTR(-EOPNOTSUPP);
2286 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2288 return ERR_PTR(size);
2290 acl = posix_acl_from_xattr(value, size);
2296 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2298 struct inode *inode = fhp->fh_dentry->d_inode;
2304 if (!IS_POSIXACL(inode) ||
2305 !inode->i_op->setxattr || !inode->i_op->removexattr)
2308 case ACL_TYPE_ACCESS:
2309 name = POSIX_ACL_XATTR_ACCESS;
2311 case ACL_TYPE_DEFAULT:
2312 name = POSIX_ACL_XATTR_DEFAULT;
2318 if (acl && acl->a_count) {
2319 size = posix_acl_xattr_size(acl->a_count);
2320 value = kmalloc(size, GFP_KERNEL);
2323 error = posix_acl_to_xattr(acl, value, size);
2330 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2334 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2336 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2339 error = vfs_removexattr(fhp->fh_dentry, name);
2340 if (error == -ENODATA)
2344 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2350 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */