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) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
351 ((iap->ia_valid & ATTR_GID) && !gid_eq(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)
511 buflen = posix_acl_xattr_size(pacl->a_count);
512 buf = kmalloc(buflen, GFP_KERNEL);
517 len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
523 error = vfs_setxattr(dentry, key, buf, len, 0);
530 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
531 struct nfs4_acl *acl)
535 struct dentry *dentry;
537 struct posix_acl *pacl = NULL, *dpacl = NULL;
538 unsigned int flags = 0;
541 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
545 dentry = fhp->fh_dentry;
546 inode = dentry->d_inode;
547 if (S_ISDIR(inode->i_mode))
548 flags = NFS4_ACL_DIR;
550 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
551 if (host_error == -EINVAL) {
552 return nfserr_attrnotsupp;
553 } else if (host_error < 0)
556 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
560 if (S_ISDIR(inode->i_mode))
561 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
564 posix_acl_release(pacl);
565 posix_acl_release(dpacl);
567 if (host_error == -EOPNOTSUPP)
568 return nfserr_attrnotsupp;
570 return nfserrno(host_error);
573 static struct posix_acl *
574 _get_posix_acl(struct dentry *dentry, char *key)
577 struct posix_acl *pacl = NULL;
580 buflen = nfsd_getxattr(dentry, key, &buf);
584 return ERR_PTR(buflen);
586 pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
592 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
594 struct inode *inode = dentry->d_inode;
596 struct posix_acl *pacl = NULL, *dpacl = NULL;
597 unsigned int flags = 0;
599 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
600 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
601 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
603 error = PTR_ERR(pacl);
608 if (S_ISDIR(inode->i_mode)) {
609 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
610 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
612 else if (IS_ERR(dpacl)) {
613 error = PTR_ERR(dpacl);
617 flags = NFS4_ACL_DIR;
620 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
622 error = PTR_ERR(*acl);
626 posix_acl_release(pacl);
627 posix_acl_release(dpacl);
632 * NFS junction information is stored in an extended attribute.
634 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
637 * nfsd4_is_junction - Test if an object could be an NFS junction
639 * @dentry: object to test
641 * Returns 1 if "dentry" appears to contain NFS junction information.
642 * Otherwise 0 is returned.
644 int nfsd4_is_junction(struct dentry *dentry)
646 struct inode *inode = dentry->d_inode;
650 if (inode->i_mode & S_IXUGO)
652 if (!(inode->i_mode & S_ISVTX))
654 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
658 #endif /* defined(CONFIG_NFSD_V4) */
660 #ifdef CONFIG_NFSD_V3
662 * Check server access rights to a file system object
668 static struct accessmap nfs3_regaccess[] = {
669 { NFS3_ACCESS_READ, NFSD_MAY_READ },
670 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
671 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
672 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
677 static struct accessmap nfs3_diraccess[] = {
678 { NFS3_ACCESS_READ, NFSD_MAY_READ },
679 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
680 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
681 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
682 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
687 static struct accessmap nfs3_anyaccess[] = {
688 /* Some clients - Solaris 2.6 at least, make an access call
689 * to the server to check for access for things like /dev/null
690 * (which really, the server doesn't care about). So
691 * We provide simple access checking for them, looking
692 * mainly at mode bits, and we make sure to ignore read-only
695 { NFS3_ACCESS_READ, NFSD_MAY_READ },
696 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
697 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
698 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
704 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
706 struct accessmap *map;
707 struct svc_export *export;
708 struct dentry *dentry;
709 u32 query, result = 0, sresult = 0;
712 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
716 export = fhp->fh_export;
717 dentry = fhp->fh_dentry;
719 if (S_ISREG(dentry->d_inode->i_mode))
720 map = nfs3_regaccess;
721 else if (S_ISDIR(dentry->d_inode->i_mode))
722 map = nfs3_diraccess;
724 map = nfs3_anyaccess;
728 for (; map->access; map++) {
729 if (map->access & query) {
732 sresult |= map->access;
734 err2 = nfsd_permission(rqstp, export, dentry, map->how);
737 result |= map->access;
740 /* the following error codes just mean the access was not allowed,
741 * rather than an error occurred */
745 /* simply don't "or" in the access bit. */
755 *supported = sresult;
760 #endif /* CONFIG_NFSD_V3 */
762 static int nfsd_open_break_lease(struct inode *inode, int access)
766 if (access & NFSD_MAY_NOT_BREAK_LEASE)
768 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
769 return break_lease(inode, mode | O_NONBLOCK);
773 * Open an existing file or directory.
774 * The may_flags argument indicates the type of open (read/write/lock)
775 * and additional flags.
776 * N.B. After this call fhp needs an fh_put
779 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
780 int may_flags, struct file **filp)
784 int flags = O_RDONLY|O_LARGEFILE;
788 validate_process_creds();
791 * If we get here, then the client has already done an "open",
792 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
793 * in case a chmod has now revoked permission.
795 * Arguably we should also allow the owner override for
796 * directories, but we never have and it doesn't seem to have
797 * caused anyone a problem. If we were to change this, note
798 * also that our filldir callbacks would need a variant of
799 * lookup_one_len that doesn't check permissions.
802 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
803 err = fh_verify(rqstp, fhp, type, may_flags);
807 path.mnt = fhp->fh_export->ex_path.mnt;
808 path.dentry = fhp->fh_dentry;
809 inode = path.dentry->d_inode;
811 /* Disallow write access to files with the append-only bit set
812 * or any access when mandatory locking enabled
815 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
818 * We must ignore files (but only files) which might have mandatory
819 * locks on them because there is no way to know if the accesser has
822 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
828 host_err = nfsd_open_break_lease(inode, may_flags);
829 if (host_err) /* NOMEM or WOULDBLOCK */
832 if (may_flags & NFSD_MAY_WRITE) {
833 if (may_flags & NFSD_MAY_READ)
834 flags = O_RDWR|O_LARGEFILE;
836 flags = O_WRONLY|O_LARGEFILE;
838 *filp = dentry_open(&path, flags, current_cred());
840 host_err = PTR_ERR(*filp);
843 host_err = ima_file_check(*filp, may_flags);
845 if (may_flags & NFSD_MAY_64BIT_COOKIE)
846 (*filp)->f_mode |= FMODE_64BITHASH;
848 (*filp)->f_mode |= FMODE_32BITHASH;
852 err = nfserrno(host_err);
854 validate_process_creds();
862 nfsd_close(struct file *filp)
868 * Obtain the readahead parameters for the file
869 * specified by (dev, ino).
872 static inline struct raparms *
873 nfsd_get_raparms(dev_t dev, ino_t ino)
875 struct raparms *ra, **rap, **frap = NULL;
878 struct raparm_hbucket *rab;
880 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
881 rab = &raparm_hash[hash];
883 spin_lock(&rab->pb_lock);
884 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
885 if (ra->p_ino == ino && ra->p_dev == dev)
888 if (ra->p_count == 0)
891 depth = nfsdstats.ra_size;
893 spin_unlock(&rab->pb_lock);
903 if (rap != &rab->pb_head) {
905 ra->p_next = rab->pb_head;
909 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
910 spin_unlock(&rab->pb_lock);
915 * Grab and keep cached pages associated with a file in the svc_rqst
916 * so that they can be passed to the network sendmsg/sendpage routines
917 * directly. They will be released after the sending has completed.
920 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
921 struct splice_desc *sd)
923 struct svc_rqst *rqstp = sd->u.data;
924 struct page **pp = rqstp->rq_next_page;
925 struct page *page = buf->page;
930 if (rqstp->rq_res.page_len == 0) {
932 put_page(*rqstp->rq_next_page);
933 *(rqstp->rq_next_page++) = page;
934 rqstp->rq_res.page_base = buf->offset;
935 rqstp->rq_res.page_len = size;
936 } else if (page != pp[-1]) {
938 if (*rqstp->rq_next_page)
939 put_page(*rqstp->rq_next_page);
940 *(rqstp->rq_next_page++) = page;
941 rqstp->rq_res.page_len += size;
943 rqstp->rq_res.page_len += size;
948 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
949 struct splice_desc *sd)
951 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
955 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
956 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
964 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
965 struct splice_desc sd = {
972 rqstp->rq_next_page = rqstp->rq_respages + 1;
973 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
977 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
982 nfsdstats.io_read += host_err;
985 fsnotify_access(file);
987 err = nfserrno(host_err);
991 static void kill_suid(struct dentry *dentry)
994 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
996 mutex_lock(&dentry->d_inode->i_mutex);
997 notify_change(dentry, &ia);
998 mutex_unlock(&dentry->d_inode->i_mutex);
1002 * Gathered writes: If another process is currently writing to the file,
1003 * there's a high chance this is another nfsd (triggered by a bulk write
1004 * from a client's biod). Rather than syncing the file with each write
1005 * request, we sleep for 10 msec.
1007 * I don't know if this roughly approximates C. Juszak's idea of
1008 * gathered writes, but it's a nice and simple solution (IMHO), and it
1011 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1012 * better tool (separate unstable writes and commits) for solving this
1015 static int wait_for_concurrent_writes(struct file *file)
1017 struct inode *inode = file_inode(file);
1018 static ino_t last_ino;
1019 static dev_t last_dev;
1022 if (atomic_read(&inode->i_writecount) > 1
1023 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1024 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1026 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1029 if (inode->i_state & I_DIRTY) {
1030 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1031 err = vfs_fsync(file, 0);
1033 last_ino = inode->i_ino;
1034 last_dev = inode->i_sb->s_dev;
1039 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1040 loff_t offset, struct kvec *vec, int vlen,
1041 unsigned long *cnt, int *stablep)
1043 struct svc_export *exp;
1044 struct dentry *dentry;
1045 struct inode *inode;
1049 int stable = *stablep;
1051 loff_t pos = offset;
1053 dentry = file->f_path.dentry;
1054 inode = dentry->d_inode;
1055 exp = fhp->fh_export;
1057 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1059 if (!EX_ISSYNC(exp))
1062 /* Write the data. */
1063 oldfs = get_fs(); set_fs(KERNEL_DS);
1064 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1069 nfsdstats.io_write += host_err;
1070 fsnotify_modify(file);
1072 /* clear setuid/setgid flag after write */
1073 if (inode->i_mode & (S_ISUID | S_ISGID))
1078 host_err = wait_for_concurrent_writes(file);
1080 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1084 dprintk("nfsd: write complete host_err=%d\n", host_err);
1088 err = nfserrno(host_err);
1093 * Read data from a file. count must contain the requested read count
1094 * on entry. On return, *count contains the number of bytes actually read.
1095 * N.B. After this call fhp needs an fh_put
1097 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1098 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1101 struct inode *inode;
1105 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1109 inode = file_inode(file);
1111 /* Get readahead parameters */
1112 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1114 if (ra && ra->p_set)
1115 file->f_ra = ra->p_ra;
1117 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1119 /* Write back readahead params */
1121 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1122 spin_lock(&rab->pb_lock);
1123 ra->p_ra = file->f_ra;
1126 spin_unlock(&rab->pb_lock);
1133 /* As above, but use the provided file descriptor. */
1135 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1136 loff_t offset, struct kvec *vec, int vlen,
1137 unsigned long *count)
1142 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1143 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1146 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1147 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1148 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1154 * Write data to a file.
1155 * The stable flag requests synchronous writes.
1156 * N.B. After this call fhp needs an fh_put
1159 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1160 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1166 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1167 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1170 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1173 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1178 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1186 #ifdef CONFIG_NFSD_V3
1188 * Commit all pending writes to stable storage.
1190 * Note: we only guarantee that data that lies within the range specified
1191 * by the 'offset' and 'count' parameters will be synced.
1193 * Unfortunately we cannot lock the file to make sure we return full WCC
1194 * data to the client, as locking happens lower down in the filesystem.
1197 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1198 loff_t offset, unsigned long count)
1201 loff_t end = LLONG_MAX;
1202 __be32 err = nfserr_inval;
1207 end = offset + (loff_t)count - 1;
1212 err = nfsd_open(rqstp, fhp, S_IFREG,
1213 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1216 if (EX_ISSYNC(fhp->fh_export)) {
1217 int err2 = vfs_fsync_range(file, offset, end, 0);
1219 if (err2 != -EINVAL)
1220 err = nfserrno(err2);
1222 err = nfserr_notsupp;
1229 #endif /* CONFIG_NFSD_V3 */
1232 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1236 * Mode has already been set earlier in create:
1238 iap->ia_valid &= ~ATTR_MODE;
1240 * Setting uid/gid works only for root. Irix appears to
1241 * send along the gid on create when it tries to implement
1242 * setgid directories via NFS:
1244 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1245 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1247 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1251 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1252 * setting size to 0 may fail for some specific file systems by the permission
1253 * checking which requires WRITE permission but the mode is 000.
1254 * we ignore the resizing(to 0) on the just new created file, since the size is
1255 * 0 after file created.
1257 * call this only after vfs_create() is called.
1260 nfsd_check_ignore_resizing(struct iattr *iap)
1262 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1263 iap->ia_valid &= ~ATTR_SIZE;
1267 * Create a file (regular, directory, device, fifo); UNIX sockets
1268 * not yet implemented.
1269 * If the response fh has been verified, the parent directory should
1270 * already be locked. Note that the parent directory is left locked.
1272 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1275 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1276 char *fname, int flen, struct iattr *iap,
1277 int type, dev_t rdev, struct svc_fh *resfhp)
1279 struct dentry *dentry, *dchild = NULL;
1289 if (isdotent(fname, flen))
1292 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1296 dentry = fhp->fh_dentry;
1297 dirp = dentry->d_inode;
1299 err = nfserr_notdir;
1300 if (!dirp->i_op->lookup)
1303 * Check whether the response file handle has been verified yet.
1304 * If it has, the parent directory should already be locked.
1306 if (!resfhp->fh_dentry) {
1307 host_err = fh_want_write(fhp);
1311 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1312 fh_lock_nested(fhp, I_MUTEX_PARENT);
1313 dchild = lookup_one_len(fname, dentry, flen);
1314 host_err = PTR_ERR(dchild);
1317 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1321 /* called from nfsd_proc_create */
1322 dchild = dget(resfhp->fh_dentry);
1323 if (!fhp->fh_locked) {
1324 /* not actually possible */
1326 "nfsd_create: parent %s/%s not locked!\n",
1327 dentry->d_parent->d_name.name,
1328 dentry->d_name.name);
1334 * Make sure the child dentry is still negative ...
1337 if (dchild->d_inode) {
1338 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1339 dentry->d_name.name, dchild->d_name.name);
1343 if (!(iap->ia_valid & ATTR_MODE))
1345 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1348 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1349 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1355 * Get the dir op function pointer.
1361 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1363 nfsd_check_ignore_resizing(iap);
1366 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1372 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
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));
1389 * Update the file handle to get the new inode info.
1392 err = fh_update(resfhp);
1394 if (dchild && !IS_ERR(dchild))
1399 err = nfserrno(host_err);
1403 #ifdef CONFIG_NFSD_V3
1405 static inline int nfsd_create_is_exclusive(int createmode)
1407 return createmode == NFS3_CREATE_EXCLUSIVE
1408 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1412 * NFSv3 and NFSv4 version of nfsd_create
1415 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1416 char *fname, int flen, struct iattr *iap,
1417 struct svc_fh *resfhp, int createmode, u32 *verifier,
1418 bool *truncp, bool *created)
1420 struct dentry *dentry, *dchild = NULL;
1424 __u32 v_mtime=0, v_atime=0;
1430 if (isdotent(fname, flen))
1432 if (!(iap->ia_valid & ATTR_MODE))
1434 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1438 dentry = fhp->fh_dentry;
1439 dirp = dentry->d_inode;
1441 /* Get all the sanity checks out of the way before
1442 * we lock the parent. */
1443 err = nfserr_notdir;
1444 if (!dirp->i_op->lookup)
1447 host_err = fh_want_write(fhp);
1451 fh_lock_nested(fhp, I_MUTEX_PARENT);
1454 * Compose the response file handle.
1456 dchild = lookup_one_len(fname, dentry, flen);
1457 host_err = PTR_ERR(dchild);
1461 /* If file doesn't exist, check for permissions to create one */
1462 if (!dchild->d_inode) {
1463 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1468 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1472 if (nfsd_create_is_exclusive(createmode)) {
1473 /* solaris7 gets confused (bugid 4218508) if these have
1474 * the high bit set, so just clear the high bits. If this is
1475 * ever changed to use different attrs for storing the
1476 * verifier, then do_open_lookup() will also need to be fixed
1479 v_mtime = verifier[0]&0x7fffffff;
1480 v_atime = verifier[1]&0x7fffffff;
1483 if (dchild->d_inode) {
1486 switch (createmode) {
1487 case NFS3_CREATE_UNCHECKED:
1488 if (! S_ISREG(dchild->d_inode->i_mode))
1491 /* in nfsv4, we need to treat this case a little
1492 * differently. we don't want to truncate the
1493 * file now; this would be wrong if the OPEN
1494 * fails for some other reason. furthermore,
1495 * if the size is nonzero, we should ignore it
1496 * according to spec!
1498 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1501 iap->ia_valid &= ATTR_SIZE;
1505 case NFS3_CREATE_EXCLUSIVE:
1506 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1507 && dchild->d_inode->i_atime.tv_sec == v_atime
1508 && dchild->d_inode->i_size == 0 ) {
1513 case NFS4_CREATE_EXCLUSIVE4_1:
1514 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1515 && dchild->d_inode->i_atime.tv_sec == v_atime
1516 && dchild->d_inode->i_size == 0 ) {
1522 case NFS3_CREATE_GUARDED:
1529 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1537 nfsd_check_ignore_resizing(iap);
1539 if (nfsd_create_is_exclusive(createmode)) {
1540 /* Cram the verifier into atime/mtime */
1541 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1542 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1543 /* XXX someone who knows this better please fix it for nsec */
1544 iap->ia_mtime.tv_sec = v_mtime;
1545 iap->ia_atime.tv_sec = v_atime;
1546 iap->ia_mtime.tv_nsec = 0;
1547 iap->ia_atime.tv_nsec = 0;
1551 err = nfsd_create_setattr(rqstp, resfhp, iap);
1554 * nfsd_setattr already committed the child (and possibly also the parent).
1557 err = nfserrno(commit_metadata(fhp));
1560 * Update the filehandle to get the new inode info.
1563 err = fh_update(resfhp);
1567 if (dchild && !IS_ERR(dchild))
1573 err = nfserrno(host_err);
1576 #endif /* CONFIG_NFSD_V3 */
1579 * Read a symlink. On entry, *lenp must contain the maximum path length that
1580 * fits into the buffer. On return, it contains the true length.
1581 * N.B. After this call fhp needs an fh_put
1584 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1586 struct inode *inode;
1592 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1596 path.mnt = fhp->fh_export->ex_path.mnt;
1597 path.dentry = fhp->fh_dentry;
1598 inode = path.dentry->d_inode;
1601 if (!inode->i_op->readlink)
1605 /* N.B. Why does this call need a get_fs()??
1606 * Remove the set_fs and watch the fireworks:-) --okir
1609 oldfs = get_fs(); set_fs(KERNEL_DS);
1610 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1621 err = nfserrno(host_err);
1626 * Create a symlink and look up its inode
1627 * N.B. After this call _both_ fhp and resfhp need an fh_put
1630 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1631 char *fname, int flen,
1632 char *path, int plen,
1633 struct svc_fh *resfhp,
1636 struct dentry *dentry, *dnew;
1644 if (isdotent(fname, flen))
1647 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1651 host_err = fh_want_write(fhp);
1656 dentry = fhp->fh_dentry;
1657 dnew = lookup_one_len(fname, dentry, flen);
1658 host_err = PTR_ERR(dnew);
1662 if (unlikely(path[plen] != 0)) {
1663 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1664 if (path_alloced == NULL)
1667 strncpy(path_alloced, path, plen);
1668 path_alloced[plen] = 0;
1669 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1670 kfree(path_alloced);
1673 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1674 err = nfserrno(host_err);
1676 err = nfserrno(commit_metadata(fhp));
1681 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1683 if (err==0) err = cerr;
1688 err = nfserrno(host_err);
1694 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1697 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1698 char *name, int len, struct svc_fh *tfhp)
1700 struct dentry *ddir, *dnew, *dold;
1705 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1708 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1712 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1718 if (isdotent(name, len))
1721 host_err = fh_want_write(tfhp);
1723 err = nfserrno(host_err);
1727 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1728 ddir = ffhp->fh_dentry;
1729 dirp = ddir->d_inode;
1731 dnew = lookup_one_len(name, ddir, len);
1732 host_err = PTR_ERR(dnew);
1736 dold = tfhp->fh_dentry;
1741 host_err = nfsd_break_lease(dold->d_inode);
1743 err = nfserrno(host_err);
1746 host_err = vfs_link(dold, dirp, dnew);
1748 err = nfserrno(commit_metadata(ffhp));
1750 err = nfserrno(commit_metadata(tfhp));
1752 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1755 err = nfserrno(host_err);
1761 fh_drop_write(tfhp);
1766 err = nfserrno(host_err);
1772 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1775 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1776 struct svc_fh *tfhp, char *tname, int tlen)
1778 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1779 struct inode *fdir, *tdir;
1783 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1786 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1790 fdentry = ffhp->fh_dentry;
1791 fdir = fdentry->d_inode;
1793 tdentry = tfhp->fh_dentry;
1794 tdir = tdentry->d_inode;
1797 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1800 host_err = fh_want_write(ffhp);
1802 err = nfserrno(host_err);
1806 /* cannot use fh_lock as we need deadlock protective ordering
1807 * so do it by hand */
1808 trap = lock_rename(tdentry, fdentry);
1809 ffhp->fh_locked = tfhp->fh_locked = 1;
1813 odentry = lookup_one_len(fname, fdentry, flen);
1814 host_err = PTR_ERR(odentry);
1815 if (IS_ERR(odentry))
1819 if (!odentry->d_inode)
1822 if (odentry == trap)
1825 ndentry = lookup_one_len(tname, tdentry, tlen);
1826 host_err = PTR_ERR(ndentry);
1827 if (IS_ERR(ndentry))
1829 host_err = -ENOTEMPTY;
1830 if (ndentry == trap)
1834 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1836 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1839 host_err = nfsd_break_lease(odentry->d_inode);
1842 if (ndentry->d_inode) {
1843 host_err = nfsd_break_lease(ndentry->d_inode);
1847 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1849 host_err = commit_metadata(tfhp);
1851 host_err = commit_metadata(ffhp);
1858 err = nfserrno(host_err);
1860 /* we cannot reply on fh_unlock on the two filehandles,
1861 * as that would do the wrong thing if the two directories
1862 * were the same, so again we do it by hand
1864 fill_post_wcc(ffhp);
1865 fill_post_wcc(tfhp);
1866 unlock_rename(tdentry, fdentry);
1867 ffhp->fh_locked = tfhp->fh_locked = 0;
1868 fh_drop_write(ffhp);
1875 * Unlink a file or directory
1876 * N.B. After this call fhp needs an fh_put
1879 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1880 char *fname, int flen)
1882 struct dentry *dentry, *rdentry;
1888 if (!flen || isdotent(fname, flen))
1890 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1894 host_err = fh_want_write(fhp);
1898 fh_lock_nested(fhp, I_MUTEX_PARENT);
1899 dentry = fhp->fh_dentry;
1900 dirp = dentry->d_inode;
1902 rdentry = lookup_one_len(fname, dentry, flen);
1903 host_err = PTR_ERR(rdentry);
1904 if (IS_ERR(rdentry))
1907 if (!rdentry->d_inode) {
1914 type = rdentry->d_inode->i_mode & S_IFMT;
1916 host_err = nfsd_break_lease(rdentry->d_inode);
1919 if (type != S_IFDIR)
1920 host_err = vfs_unlink(dirp, rdentry);
1922 host_err = vfs_rmdir(dirp, rdentry);
1924 host_err = commit_metadata(fhp);
1929 err = nfserrno(host_err);
1935 * We do this buffering because we must not call back into the file
1936 * system's ->lookup() method from the filldir callback. That may well
1937 * deadlock a number of file systems.
1939 * This is based heavily on the implementation of same in XFS.
1941 struct buffered_dirent {
1945 unsigned int d_type;
1949 struct readdir_data {
1955 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1956 loff_t offset, u64 ino, unsigned int d_type)
1958 struct readdir_data *buf = __buf;
1959 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1960 unsigned int reclen;
1962 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1963 if (buf->used + reclen > PAGE_SIZE) {
1968 de->namlen = namlen;
1969 de->offset = offset;
1971 de->d_type = d_type;
1972 memcpy(de->name, name, namlen);
1973 buf->used += reclen;
1978 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1979 struct readdir_cd *cdp, loff_t *offsetp)
1981 struct readdir_data buf;
1982 struct buffered_dirent *de;
1987 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1989 return nfserrno(-ENOMEM);
1994 struct inode *dir_inode = file_inode(file);
1995 unsigned int reclen;
1997 cdp->err = nfserr_eof; /* will be cleared on successful read */
2001 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
2014 * Various filldir functions may end up calling back into
2015 * lookup_one_len() and the file system's ->lookup() method.
2016 * These expect i_mutex to be held, as it would within readdir.
2018 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2022 de = (struct buffered_dirent *)buf.dirent;
2024 offset = de->offset;
2026 if (func(cdp, de->name, de->namlen, de->offset,
2027 de->ino, de->d_type))
2030 if (cdp->err != nfs_ok)
2033 reclen = ALIGN(sizeof(*de) + de->namlen,
2036 de = (struct buffered_dirent *)((char *)de + reclen);
2038 mutex_unlock(&dir_inode->i_mutex);
2039 if (size > 0) /* We bailed out early */
2042 offset = vfs_llseek(file, 0, SEEK_CUR);
2045 free_page((unsigned long)(buf.dirent));
2048 return nfserrno(host_err);
2055 * Read entries from a directory.
2056 * The NFSv3/4 verifier we ignore for now.
2059 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2060 struct readdir_cd *cdp, filldir_t func)
2064 loff_t offset = *offsetp;
2065 int may_flags = NFSD_MAY_READ;
2067 /* NFSv2 only supports 32 bit cookies */
2068 if (rqstp->rq_vers > 2)
2069 may_flags |= NFSD_MAY_64BIT_COOKIE;
2071 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2075 offset = vfs_llseek(file, offset, SEEK_SET);
2077 err = nfserrno((int)offset);
2081 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2083 if (err == nfserr_eof || err == nfserr_toosmall)
2084 err = nfs_ok; /* can still be found in ->err */
2092 * Get file system stats
2093 * N.B. After this call fhp needs an fh_put
2096 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2100 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2102 struct path path = {
2103 .mnt = fhp->fh_export->ex_path.mnt,
2104 .dentry = fhp->fh_dentry,
2106 if (vfs_statfs(&path, stat))
2112 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2114 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2118 * Check for a user's access permissions to this inode.
2121 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2122 struct dentry *dentry, int acc)
2124 struct inode *inode = dentry->d_inode;
2127 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2130 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2132 (acc & NFSD_MAY_READ)? " read" : "",
2133 (acc & NFSD_MAY_WRITE)? " write" : "",
2134 (acc & NFSD_MAY_EXEC)? " exec" : "",
2135 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2136 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2137 (acc & NFSD_MAY_LOCK)? " lock" : "",
2138 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2140 IS_IMMUTABLE(inode)? " immut" : "",
2141 IS_APPEND(inode)? " append" : "",
2142 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2143 dprintk(" owner %d/%d user %d/%d\n",
2144 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2147 /* Normally we reject any write/sattr etc access on a read-only file
2148 * system. But if it is IRIX doing check on write-access for a
2149 * device special file, we ignore rofs.
2151 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2152 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2153 if (exp_rdonly(rqstp, exp) ||
2154 __mnt_is_readonly(exp->ex_path.mnt))
2156 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2159 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2162 if (acc & NFSD_MAY_LOCK) {
2163 /* If we cannot rely on authentication in NLM requests,
2164 * just allow locks, otherwise require read permission, or
2167 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2170 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2173 * The file owner always gets access permission for accesses that
2174 * would normally be checked at open time. This is to make
2175 * file access work even when the client has done a fchmod(fd, 0).
2177 * However, `cp foo bar' should fail nevertheless when bar is
2178 * readonly. A sensible way to do this might be to reject all
2179 * attempts to truncate a read-only file, because a creat() call
2180 * always implies file truncation.
2181 * ... but this isn't really fair. A process may reasonably call
2182 * ftruncate on an open file descriptor on a file with perm 000.
2183 * We must trust the client to do permission checking - using "ACCESS"
2186 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2187 uid_eq(inode->i_uid, current_fsuid()))
2190 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2191 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2193 /* Allow read access to binaries even when mode 111 */
2194 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2195 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2196 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2197 err = inode_permission(inode, MAY_EXEC);
2199 return err? nfserrno(err) : 0;
2203 nfsd_racache_shutdown(void)
2205 struct raparms *raparm, *last_raparm;
2208 dprintk("nfsd: freeing readahead buffers.\n");
2210 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2211 raparm = raparm_hash[i].pb_head;
2213 last_raparm = raparm;
2214 raparm = raparm->p_next;
2217 raparm_hash[i].pb_head = NULL;
2221 * Initialize readahead param cache
2224 nfsd_racache_init(int cache_size)
2229 struct raparms **raparm = NULL;
2232 if (raparm_hash[0].pb_head)
2234 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2237 cache_size = nperbucket * RAPARM_HASH_SIZE;
2239 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2241 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2242 spin_lock_init(&raparm_hash[i].pb_lock);
2244 raparm = &raparm_hash[i].pb_head;
2245 for (j = 0; j < nperbucket; j++) {
2246 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2249 raparm = &(*raparm)->p_next;
2254 nfsdstats.ra_size = cache_size;
2258 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2259 nfsd_racache_shutdown();
2263 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2265 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2267 struct inode *inode = fhp->fh_dentry->d_inode;
2271 struct posix_acl *acl;
2273 if (!IS_POSIXACL(inode))
2274 return ERR_PTR(-EOPNOTSUPP);
2277 case ACL_TYPE_ACCESS:
2278 name = POSIX_ACL_XATTR_ACCESS;
2280 case ACL_TYPE_DEFAULT:
2281 name = POSIX_ACL_XATTR_DEFAULT;
2284 return ERR_PTR(-EOPNOTSUPP);
2287 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2289 return ERR_PTR(size);
2291 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2297 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2299 struct inode *inode = fhp->fh_dentry->d_inode;
2305 if (!IS_POSIXACL(inode) ||
2306 !inode->i_op->setxattr || !inode->i_op->removexattr)
2309 case ACL_TYPE_ACCESS:
2310 name = POSIX_ACL_XATTR_ACCESS;
2312 case ACL_TYPE_DEFAULT:
2313 name = POSIX_ACL_XATTR_DEFAULT;
2319 if (acl && acl->a_count) {
2320 size = posix_acl_xattr_size(acl->a_count);
2321 value = kmalloc(size, GFP_KERNEL);
2324 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2331 error = fh_want_write(fhp);
2335 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2337 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2340 error = vfs_removexattr(fhp->fh_dentry, name);
2341 if (error == -ENODATA)
2351 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */