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>
31 #include <linux/security.h>
35 #endif /* CONFIG_NFSD_V3 */
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms *p_next;
61 struct file_ra_state p_ra;
62 unsigned int p_hindex;
65 struct raparm_hbucket {
66 struct raparms *pb_head;
68 } ____cacheline_aligned_in_smp;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
87 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
91 err = follow_down(&path);
95 exp2 = rqst_exp_get_by_name(rqstp, &path);
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
112 /* successfully crossed mount point */
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
120 path.dentry = dentry;
130 static void follow_to_parent(struct path *path)
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
136 dp = dget_parent(path->dentry);
141 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
147 follow_to_parent(&path);
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
154 return PTR_ERR(exp2);
156 *dentryp = dget(path.dentry);
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
168 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
170 if (d_mountpoint(dentry))
172 if (nfsd4_is_junction(dentry))
174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
176 return dentry->d_inode != NULL;
180 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
181 const char *name, unsigned int len,
182 struct svc_export **exp_ret, struct dentry **dentry_ret)
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
191 dparent = fhp->fh_dentry;
192 exp = fhp->fh_export;
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name, len)) {
198 dentry = dget(dparent);
199 else if (dparent != exp->ex_path.dentry)
200 dentry = dget_parent(dparent);
201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
202 dentry = dget(dparent); /* .. == . just like at / */
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
211 dentry = lookup_one_len(name, dparent, len);
212 host_err = PTR_ERR(dentry);
216 * check if we have crossed a mount point ...
218 if (nfsd_mountpoint(dentry, exp)) {
219 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
225 *dentry_ret = dentry;
231 return nfserrno(host_err);
235 * Look up one component of a pathname.
236 * N.B. After this call _both_ fhp and resfh need an fh_put
238 * If the lookup would cross a mountpoint, and the mounted filesystem
239 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
240 * accepted as it stands and the mounted directory is
241 * returned. Otherwise the covered directory is returned.
242 * NOTE: this mountpoint crossing is not supported properly by all
243 * clients and is explicitly disallowed for NFSv3
244 * NeilBrown <neilb@cse.unsw.edu.au>
247 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
248 unsigned int len, struct svc_fh *resfh)
250 struct svc_export *exp;
251 struct dentry *dentry;
254 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
257 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
260 err = check_nfsd_access(exp, rqstp);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err = fh_compose(resfh, exp, dentry, fhp);
268 if (!err && !dentry->d_inode)
276 static int nfsd_break_lease(struct inode *inode)
278 if (!S_ISREG(inode->i_mode))
280 return break_lease(inode, O_WRONLY | O_NONBLOCK);
284 * Commit metadata changes to stable storage.
287 commit_metadata(struct svc_fh *fhp)
289 struct inode *inode = fhp->fh_dentry->d_inode;
290 const struct export_operations *export_ops = inode->i_sb->s_export_op;
292 if (!EX_ISSYNC(fhp->fh_export))
295 if (export_ops->commit_metadata)
296 return export_ops->commit_metadata(inode);
297 return sync_inode_metadata(inode, 1);
301 * Go over the attributes and take care of the small differences between
302 * NFS semantics and what Linux expects.
305 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
308 * NFSv2 does not differentiate between "set-[ac]time-to-now"
309 * which only requires access, and "set-[ac]time-to-X" which
310 * requires ownership.
311 * So if it looks like it might be "set both to the same time which
312 * is close to now", and if inode_change_ok fails, then we
313 * convert to "set to now" instead of "set to explicit time"
315 * We only call inode_change_ok as the last test as technically
316 * it is not an interface that we should be using.
318 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
319 #define MAX_TOUCH_TIME_ERROR (30*60)
320 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
321 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
325 * Now just make sure time is in the right ballpark.
326 * Solaris, at least, doesn't seem to care what the time
327 * request is. We require it be within 30 minutes of now.
329 time_t delta = iap->ia_atime.tv_sec - get_seconds();
332 if (delta < MAX_TOUCH_TIME_ERROR &&
333 inode_change_ok(inode, iap) != 0) {
335 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
336 * This will cause notify_change to set these times
339 iap->ia_valid &= ~BOTH_TIME_SET;
343 /* sanitize the mode change */
344 if (iap->ia_valid & ATTR_MODE) {
345 iap->ia_mode &= S_IALLUGO;
346 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
349 /* Revoke setuid/setgid on chown */
350 if (!S_ISDIR(inode->i_mode) &&
351 (((iap->ia_valid & ATTR_UID) && !uid_eq(iap->ia_uid, inode->i_uid)) ||
352 ((iap->ia_valid & ATTR_GID) && !gid_eq(iap->ia_gid, inode->i_gid)))) {
353 iap->ia_valid |= ATTR_KILL_PRIV;
354 if (iap->ia_valid & ATTR_MODE) {
355 /* we're setting mode too, just clear the s*id bits */
356 iap->ia_mode &= ~S_ISUID;
357 if (iap->ia_mode & S_IXGRP)
358 iap->ia_mode &= ~S_ISGID;
360 /* set ATTR_KILL_* bits and let VFS handle it */
361 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
367 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
370 struct inode *inode = fhp->fh_dentry->d_inode;
373 if (iap->ia_size < inode->i_size) {
376 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
377 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
382 host_err = get_write_access(inode);
386 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
388 goto out_put_write_access;
391 out_put_write_access:
392 put_write_access(inode);
394 return nfserrno(host_err);
398 * Set various file attributes. After this call fhp needs an fh_put.
401 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
402 int check_guard, time_t guardtime)
404 struct dentry *dentry;
406 int accmode = NFSD_MAY_SATTR;
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)
418 err = fh_verify(rqstp, fhp, ftype, accmode);
422 dentry = fhp->fh_dentry;
423 inode = dentry->d_inode;
425 /* Ignore any mode updates on symlinks */
426 if (S_ISLNK(inode->i_mode))
427 iap->ia_valid &= ~ATTR_MODE;
432 nfsd_sanitize_attrs(inode, iap);
435 * The size case is special, it changes the file in addition to the
438 if (iap->ia_valid & ATTR_SIZE) {
439 err = nfsd_get_write_access(rqstp, fhp, iap);
445 iap->ia_valid |= ATTR_CTIME;
447 err = nfserr_notsync;
448 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
449 host_err = nfsd_break_lease(inode);
454 host_err = notify_change(dentry, iap, NULL);
455 err = nfserrno(host_err);
459 put_write_access(inode);
461 commit_metadata(fhp);
466 err = nfserrno(host_err);
470 #if defined(CONFIG_NFSD_V2_ACL) || \
471 defined(CONFIG_NFSD_V3_ACL) || \
472 defined(CONFIG_NFSD_V4)
473 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
478 buflen = vfs_getxattr(dentry, key, NULL, 0);
482 *buf = kmalloc(buflen, GFP_KERNEL);
486 ret = vfs_getxattr(dentry, key, *buf, buflen);
493 #if defined(CONFIG_NFSD_V4)
495 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
502 buflen = posix_acl_xattr_size(pacl->a_count);
503 buf = kmalloc(buflen, GFP_KERNEL);
508 len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
514 error = vfs_setxattr(dentry, key, buf, len, 0);
521 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
522 struct nfs4_acl *acl)
526 struct dentry *dentry;
528 struct posix_acl *pacl = NULL, *dpacl = NULL;
529 unsigned int flags = 0;
532 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
536 dentry = fhp->fh_dentry;
537 inode = dentry->d_inode;
538 if (S_ISDIR(inode->i_mode))
539 flags = NFS4_ACL_DIR;
541 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
542 if (host_error == -EINVAL) {
543 return nfserr_attrnotsupp;
544 } else if (host_error < 0)
547 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
551 if (S_ISDIR(inode->i_mode))
552 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
555 posix_acl_release(pacl);
556 posix_acl_release(dpacl);
558 if (host_error == -EOPNOTSUPP)
559 return nfserr_attrnotsupp;
561 return nfserrno(host_error);
564 static struct posix_acl *
565 _get_posix_acl(struct dentry *dentry, char *key)
568 struct posix_acl *pacl = NULL;
571 buflen = nfsd_getxattr(dentry, key, &buf);
575 return ERR_PTR(buflen);
577 pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
583 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
585 struct inode *inode = dentry->d_inode;
587 struct posix_acl *pacl = NULL, *dpacl = NULL;
588 unsigned int flags = 0;
590 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
591 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
592 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
594 error = PTR_ERR(pacl);
599 if (S_ISDIR(inode->i_mode)) {
600 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
601 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
603 else if (IS_ERR(dpacl)) {
604 error = PTR_ERR(dpacl);
608 flags = NFS4_ACL_DIR;
611 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
613 error = PTR_ERR(*acl);
617 posix_acl_release(pacl);
618 posix_acl_release(dpacl);
623 * NFS junction information is stored in an extended attribute.
625 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
628 * nfsd4_is_junction - Test if an object could be an NFS junction
630 * @dentry: object to test
632 * Returns 1 if "dentry" appears to contain NFS junction information.
633 * Otherwise 0 is returned.
635 int nfsd4_is_junction(struct dentry *dentry)
637 struct inode *inode = dentry->d_inode;
641 if (inode->i_mode & S_IXUGO)
643 if (!(inode->i_mode & S_ISVTX))
645 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
649 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
650 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
651 struct xdr_netobj *label)
655 struct dentry *dentry;
657 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
661 dentry = fhp->fh_dentry;
663 mutex_lock(&dentry->d_inode->i_mutex);
664 host_error = security_inode_setsecctx(dentry, label->data, label->len);
665 mutex_unlock(&dentry->d_inode->i_mutex);
666 return nfserrno(host_error);
669 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
670 struct xdr_netobj *label)
672 return nfserr_notsupp;
676 #endif /* defined(CONFIG_NFSD_V4) */
678 #ifdef CONFIG_NFSD_V3
680 * Check server access rights to a file system object
686 static struct accessmap nfs3_regaccess[] = {
687 { NFS3_ACCESS_READ, NFSD_MAY_READ },
688 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
689 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
690 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
695 static struct accessmap nfs3_diraccess[] = {
696 { NFS3_ACCESS_READ, NFSD_MAY_READ },
697 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
698 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
699 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
700 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
705 static struct accessmap nfs3_anyaccess[] = {
706 /* Some clients - Solaris 2.6 at least, make an access call
707 * to the server to check for access for things like /dev/null
708 * (which really, the server doesn't care about). So
709 * We provide simple access checking for them, looking
710 * mainly at mode bits, and we make sure to ignore read-only
713 { NFS3_ACCESS_READ, NFSD_MAY_READ },
714 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
715 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
716 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
722 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
724 struct accessmap *map;
725 struct svc_export *export;
726 struct dentry *dentry;
727 u32 query, result = 0, sresult = 0;
730 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
734 export = fhp->fh_export;
735 dentry = fhp->fh_dentry;
737 if (S_ISREG(dentry->d_inode->i_mode))
738 map = nfs3_regaccess;
739 else if (S_ISDIR(dentry->d_inode->i_mode))
740 map = nfs3_diraccess;
742 map = nfs3_anyaccess;
746 for (; map->access; map++) {
747 if (map->access & query) {
750 sresult |= map->access;
752 err2 = nfsd_permission(rqstp, export, dentry, map->how);
755 result |= map->access;
758 /* the following error codes just mean the access was not allowed,
759 * rather than an error occurred */
763 /* simply don't "or" in the access bit. */
773 *supported = sresult;
778 #endif /* CONFIG_NFSD_V3 */
780 static int nfsd_open_break_lease(struct inode *inode, int access)
784 if (access & NFSD_MAY_NOT_BREAK_LEASE)
786 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
787 return break_lease(inode, mode | O_NONBLOCK);
791 * Open an existing file or directory.
792 * The may_flags argument indicates the type of open (read/write/lock)
793 * and additional flags.
794 * N.B. After this call fhp needs an fh_put
797 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
798 int may_flags, struct file **filp)
802 int flags = O_RDONLY|O_LARGEFILE;
806 validate_process_creds();
809 * If we get here, then the client has already done an "open",
810 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
811 * in case a chmod has now revoked permission.
813 * Arguably we should also allow the owner override for
814 * directories, but we never have and it doesn't seem to have
815 * caused anyone a problem. If we were to change this, note
816 * also that our filldir callbacks would need a variant of
817 * lookup_one_len that doesn't check permissions.
820 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
821 err = fh_verify(rqstp, fhp, type, may_flags);
825 path.mnt = fhp->fh_export->ex_path.mnt;
826 path.dentry = fhp->fh_dentry;
827 inode = path.dentry->d_inode;
829 /* Disallow write access to files with the append-only bit set
830 * or any access when mandatory locking enabled
833 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
836 * We must ignore files (but only files) which might have mandatory
837 * locks on them because there is no way to know if the accesser has
840 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
846 host_err = nfsd_open_break_lease(inode, may_flags);
847 if (host_err) /* NOMEM or WOULDBLOCK */
850 if (may_flags & NFSD_MAY_WRITE) {
851 if (may_flags & NFSD_MAY_READ)
852 flags = O_RDWR|O_LARGEFILE;
854 flags = O_WRONLY|O_LARGEFILE;
856 *filp = dentry_open(&path, flags, current_cred());
858 host_err = PTR_ERR(*filp);
861 host_err = ima_file_check(*filp, may_flags);
863 if (may_flags & NFSD_MAY_64BIT_COOKIE)
864 (*filp)->f_mode |= FMODE_64BITHASH;
866 (*filp)->f_mode |= FMODE_32BITHASH;
870 err = nfserrno(host_err);
872 validate_process_creds();
880 nfsd_close(struct file *filp)
886 * Obtain the readahead parameters for the file
887 * specified by (dev, ino).
890 static inline struct raparms *
891 nfsd_get_raparms(dev_t dev, ino_t ino)
893 struct raparms *ra, **rap, **frap = NULL;
896 struct raparm_hbucket *rab;
898 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
899 rab = &raparm_hash[hash];
901 spin_lock(&rab->pb_lock);
902 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
903 if (ra->p_ino == ino && ra->p_dev == dev)
906 if (ra->p_count == 0)
909 depth = nfsdstats.ra_size;
911 spin_unlock(&rab->pb_lock);
921 if (rap != &rab->pb_head) {
923 ra->p_next = rab->pb_head;
927 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
928 spin_unlock(&rab->pb_lock);
933 * Grab and keep cached pages associated with a file in the svc_rqst
934 * so that they can be passed to the network sendmsg/sendpage routines
935 * directly. They will be released after the sending has completed.
938 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
939 struct splice_desc *sd)
941 struct svc_rqst *rqstp = sd->u.data;
942 struct page **pp = rqstp->rq_next_page;
943 struct page *page = buf->page;
948 if (rqstp->rq_res.page_len == 0) {
950 put_page(*rqstp->rq_next_page);
951 *(rqstp->rq_next_page++) = page;
952 rqstp->rq_res.page_base = buf->offset;
953 rqstp->rq_res.page_len = size;
954 } else if (page != pp[-1]) {
956 if (*rqstp->rq_next_page)
957 put_page(*rqstp->rq_next_page);
958 *(rqstp->rq_next_page++) = page;
959 rqstp->rq_res.page_len += size;
961 rqstp->rq_res.page_len += size;
966 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
967 struct splice_desc *sd)
969 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
973 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
974 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
982 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
983 struct splice_desc sd = {
990 rqstp->rq_next_page = rqstp->rq_respages + 1;
991 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
995 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
1000 nfsdstats.io_read += host_err;
1003 fsnotify_access(file);
1005 err = nfserrno(host_err);
1009 static void kill_suid(struct dentry *dentry)
1012 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
1014 mutex_lock(&dentry->d_inode->i_mutex);
1016 * Note we call this on write, so notify_change will not
1017 * encounter any conflicting delegations:
1019 notify_change(dentry, &ia, NULL);
1020 mutex_unlock(&dentry->d_inode->i_mutex);
1024 * Gathered writes: If another process is currently writing to the file,
1025 * there's a high chance this is another nfsd (triggered by a bulk write
1026 * from a client's biod). Rather than syncing the file with each write
1027 * request, we sleep for 10 msec.
1029 * I don't know if this roughly approximates C. Juszak's idea of
1030 * gathered writes, but it's a nice and simple solution (IMHO), and it
1033 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1034 * better tool (separate unstable writes and commits) for solving this
1037 static int wait_for_concurrent_writes(struct file *file)
1039 struct inode *inode = file_inode(file);
1040 static ino_t last_ino;
1041 static dev_t last_dev;
1044 if (atomic_read(&inode->i_writecount) > 1
1045 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1046 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1048 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1051 if (inode->i_state & I_DIRTY) {
1052 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1053 err = vfs_fsync(file, 0);
1055 last_ino = inode->i_ino;
1056 last_dev = inode->i_sb->s_dev;
1061 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1062 loff_t offset, struct kvec *vec, int vlen,
1063 unsigned long *cnt, int *stablep)
1065 struct svc_export *exp;
1066 struct dentry *dentry;
1067 struct inode *inode;
1071 int stable = *stablep;
1073 loff_t pos = offset;
1075 dentry = file->f_path.dentry;
1076 inode = dentry->d_inode;
1077 exp = fhp->fh_export;
1079 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1081 if (!EX_ISSYNC(exp))
1084 /* Write the data. */
1085 oldfs = get_fs(); set_fs(KERNEL_DS);
1086 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1091 nfsdstats.io_write += host_err;
1092 fsnotify_modify(file);
1094 /* clear setuid/setgid flag after write */
1095 if (inode->i_mode & (S_ISUID | S_ISGID))
1100 host_err = wait_for_concurrent_writes(file);
1102 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1106 dprintk("nfsd: write complete host_err=%d\n", host_err);
1110 err = nfserrno(host_err);
1115 * Read data from a file. count must contain the requested read count
1116 * on entry. On return, *count contains the number of bytes actually read.
1117 * N.B. After this call fhp needs an fh_put
1119 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1120 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1123 struct inode *inode;
1127 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1131 inode = file_inode(file);
1133 /* Get readahead parameters */
1134 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1136 if (ra && ra->p_set)
1137 file->f_ra = ra->p_ra;
1139 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1141 /* Write back readahead params */
1143 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1144 spin_lock(&rab->pb_lock);
1145 ra->p_ra = file->f_ra;
1148 spin_unlock(&rab->pb_lock);
1155 /* As above, but use the provided file descriptor. */
1157 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1158 loff_t offset, struct kvec *vec, int vlen,
1159 unsigned long *count)
1164 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1165 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1168 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1169 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1170 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1176 * Write data to a file.
1177 * The stable flag requests synchronous writes.
1178 * N.B. After this call fhp needs an fh_put
1181 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1182 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1188 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1189 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1192 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1195 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1200 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1208 #ifdef CONFIG_NFSD_V3
1210 * Commit all pending writes to stable storage.
1212 * Note: we only guarantee that data that lies within the range specified
1213 * by the 'offset' and 'count' parameters will be synced.
1215 * Unfortunately we cannot lock the file to make sure we return full WCC
1216 * data to the client, as locking happens lower down in the filesystem.
1219 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1220 loff_t offset, unsigned long count)
1223 loff_t end = LLONG_MAX;
1224 __be32 err = nfserr_inval;
1229 end = offset + (loff_t)count - 1;
1234 err = nfsd_open(rqstp, fhp, S_IFREG,
1235 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1238 if (EX_ISSYNC(fhp->fh_export)) {
1239 int err2 = vfs_fsync_range(file, offset, end, 0);
1241 if (err2 != -EINVAL)
1242 err = nfserrno(err2);
1244 err = nfserr_notsupp;
1251 #endif /* CONFIG_NFSD_V3 */
1254 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1258 * Mode has already been set earlier in create:
1260 iap->ia_valid &= ~ATTR_MODE;
1262 * Setting uid/gid works only for root. Irix appears to
1263 * send along the gid on create when it tries to implement
1264 * setgid directories via NFS:
1266 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1267 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1269 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1273 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1274 * setting size to 0 may fail for some specific file systems by the permission
1275 * checking which requires WRITE permission but the mode is 000.
1276 * we ignore the resizing(to 0) on the just new created file, since the size is
1277 * 0 after file created.
1279 * call this only after vfs_create() is called.
1282 nfsd_check_ignore_resizing(struct iattr *iap)
1284 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1285 iap->ia_valid &= ~ATTR_SIZE;
1289 * Create a file (regular, directory, device, fifo); UNIX sockets
1290 * not yet implemented.
1291 * If the response fh has been verified, the parent directory should
1292 * already be locked. Note that the parent directory is left locked.
1294 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1297 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1298 char *fname, int flen, struct iattr *iap,
1299 int type, dev_t rdev, struct svc_fh *resfhp)
1301 struct dentry *dentry, *dchild = NULL;
1311 if (isdotent(fname, flen))
1314 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1318 dentry = fhp->fh_dentry;
1319 dirp = dentry->d_inode;
1321 err = nfserr_notdir;
1322 if (!dirp->i_op->lookup)
1325 * Check whether the response file handle has been verified yet.
1326 * If it has, the parent directory should already be locked.
1328 if (!resfhp->fh_dentry) {
1329 host_err = fh_want_write(fhp);
1333 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1334 fh_lock_nested(fhp, I_MUTEX_PARENT);
1335 dchild = lookup_one_len(fname, dentry, flen);
1336 host_err = PTR_ERR(dchild);
1339 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1343 /* called from nfsd_proc_create */
1344 dchild = dget(resfhp->fh_dentry);
1345 if (!fhp->fh_locked) {
1346 /* not actually possible */
1348 "nfsd_create: parent %pd2 not locked!\n",
1355 * Make sure the child dentry is still negative ...
1358 if (dchild->d_inode) {
1359 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1364 if (!(iap->ia_valid & ATTR_MODE))
1366 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1369 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1370 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1376 * Get the dir op function pointer.
1382 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1384 nfsd_check_ignore_resizing(iap);
1387 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1393 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1399 err = nfsd_create_setattr(rqstp, resfhp, iap);
1402 * nfsd_setattr already committed the child. Transactional filesystems
1403 * had a chance to commit changes for both parent and child
1404 * simultaneously making the following commit_metadata a noop.
1406 err2 = nfserrno(commit_metadata(fhp));
1410 * Update the file handle to get the new inode info.
1413 err = fh_update(resfhp);
1415 if (dchild && !IS_ERR(dchild))
1420 err = nfserrno(host_err);
1424 #ifdef CONFIG_NFSD_V3
1426 static inline int nfsd_create_is_exclusive(int createmode)
1428 return createmode == NFS3_CREATE_EXCLUSIVE
1429 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1433 * NFSv3 and NFSv4 version of nfsd_create
1436 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1437 char *fname, int flen, struct iattr *iap,
1438 struct svc_fh *resfhp, int createmode, u32 *verifier,
1439 bool *truncp, bool *created)
1441 struct dentry *dentry, *dchild = NULL;
1445 __u32 v_mtime=0, v_atime=0;
1451 if (isdotent(fname, flen))
1453 if (!(iap->ia_valid & ATTR_MODE))
1455 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1459 dentry = fhp->fh_dentry;
1460 dirp = dentry->d_inode;
1462 /* Get all the sanity checks out of the way before
1463 * we lock the parent. */
1464 err = nfserr_notdir;
1465 if (!dirp->i_op->lookup)
1468 host_err = fh_want_write(fhp);
1472 fh_lock_nested(fhp, I_MUTEX_PARENT);
1475 * Compose the response file handle.
1477 dchild = lookup_one_len(fname, dentry, flen);
1478 host_err = PTR_ERR(dchild);
1482 /* If file doesn't exist, check for permissions to create one */
1483 if (!dchild->d_inode) {
1484 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1489 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1493 if (nfsd_create_is_exclusive(createmode)) {
1494 /* solaris7 gets confused (bugid 4218508) if these have
1495 * the high bit set, so just clear the high bits. If this is
1496 * ever changed to use different attrs for storing the
1497 * verifier, then do_open_lookup() will also need to be fixed
1500 v_mtime = verifier[0]&0x7fffffff;
1501 v_atime = verifier[1]&0x7fffffff;
1504 if (dchild->d_inode) {
1507 switch (createmode) {
1508 case NFS3_CREATE_UNCHECKED:
1509 if (! S_ISREG(dchild->d_inode->i_mode))
1512 /* in nfsv4, we need to treat this case a little
1513 * differently. we don't want to truncate the
1514 * file now; this would be wrong if the OPEN
1515 * fails for some other reason. furthermore,
1516 * if the size is nonzero, we should ignore it
1517 * according to spec!
1519 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1522 iap->ia_valid &= ATTR_SIZE;
1526 case NFS3_CREATE_EXCLUSIVE:
1527 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1528 && dchild->d_inode->i_atime.tv_sec == v_atime
1529 && dchild->d_inode->i_size == 0 ) {
1534 case NFS4_CREATE_EXCLUSIVE4_1:
1535 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1536 && dchild->d_inode->i_atime.tv_sec == v_atime
1537 && dchild->d_inode->i_size == 0 ) {
1543 case NFS3_CREATE_GUARDED:
1550 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1558 nfsd_check_ignore_resizing(iap);
1560 if (nfsd_create_is_exclusive(createmode)) {
1561 /* Cram the verifier into atime/mtime */
1562 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1563 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1564 /* XXX someone who knows this better please fix it for nsec */
1565 iap->ia_mtime.tv_sec = v_mtime;
1566 iap->ia_atime.tv_sec = v_atime;
1567 iap->ia_mtime.tv_nsec = 0;
1568 iap->ia_atime.tv_nsec = 0;
1572 err = nfsd_create_setattr(rqstp, resfhp, iap);
1575 * nfsd_setattr already committed the child (and possibly also the parent).
1578 err = nfserrno(commit_metadata(fhp));
1581 * Update the filehandle to get the new inode info.
1584 err = fh_update(resfhp);
1588 if (dchild && !IS_ERR(dchild))
1594 err = nfserrno(host_err);
1597 #endif /* CONFIG_NFSD_V3 */
1600 * Read a symlink. On entry, *lenp must contain the maximum path length that
1601 * fits into the buffer. On return, it contains the true length.
1602 * N.B. After this call fhp needs an fh_put
1605 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1607 struct inode *inode;
1613 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1617 path.mnt = fhp->fh_export->ex_path.mnt;
1618 path.dentry = fhp->fh_dentry;
1619 inode = path.dentry->d_inode;
1622 if (!inode->i_op->readlink)
1626 /* N.B. Why does this call need a get_fs()??
1627 * Remove the set_fs and watch the fireworks:-) --okir
1630 oldfs = get_fs(); set_fs(KERNEL_DS);
1631 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1642 err = nfserrno(host_err);
1647 * Create a symlink and look up its inode
1648 * N.B. After this call _both_ fhp and resfhp need an fh_put
1651 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1652 char *fname, int flen,
1653 char *path, int plen,
1654 struct svc_fh *resfhp,
1657 struct dentry *dentry, *dnew;
1665 if (isdotent(fname, flen))
1668 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1672 host_err = fh_want_write(fhp);
1677 dentry = fhp->fh_dentry;
1678 dnew = lookup_one_len(fname, dentry, flen);
1679 host_err = PTR_ERR(dnew);
1683 if (unlikely(path[plen] != 0)) {
1684 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1685 if (path_alloced == NULL)
1688 strncpy(path_alloced, path, plen);
1689 path_alloced[plen] = 0;
1690 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1691 kfree(path_alloced);
1694 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1695 err = nfserrno(host_err);
1697 err = nfserrno(commit_metadata(fhp));
1702 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1704 if (err==0) err = cerr;
1709 err = nfserrno(host_err);
1715 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1718 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1719 char *name, int len, struct svc_fh *tfhp)
1721 struct dentry *ddir, *dnew, *dold;
1726 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1729 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1733 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1739 if (isdotent(name, len))
1742 host_err = fh_want_write(tfhp);
1744 err = nfserrno(host_err);
1748 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1749 ddir = ffhp->fh_dentry;
1750 dirp = ddir->d_inode;
1752 dnew = lookup_one_len(name, ddir, len);
1753 host_err = PTR_ERR(dnew);
1757 dold = tfhp->fh_dentry;
1762 host_err = nfsd_break_lease(dold->d_inode);
1764 err = nfserrno(host_err);
1767 host_err = vfs_link(dold, dirp, dnew, NULL);
1769 err = nfserrno(commit_metadata(ffhp));
1771 err = nfserrno(commit_metadata(tfhp));
1773 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1776 err = nfserrno(host_err);
1782 fh_drop_write(tfhp);
1787 err = nfserrno(host_err);
1793 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1796 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1797 struct svc_fh *tfhp, char *tname, int tlen)
1799 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1800 struct inode *fdir, *tdir;
1804 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1807 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1811 fdentry = ffhp->fh_dentry;
1812 fdir = fdentry->d_inode;
1814 tdentry = tfhp->fh_dentry;
1815 tdir = tdentry->d_inode;
1818 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1821 host_err = fh_want_write(ffhp);
1823 err = nfserrno(host_err);
1827 /* cannot use fh_lock as we need deadlock protective ordering
1828 * so do it by hand */
1829 trap = lock_rename(tdentry, fdentry);
1830 ffhp->fh_locked = tfhp->fh_locked = 1;
1834 odentry = lookup_one_len(fname, fdentry, flen);
1835 host_err = PTR_ERR(odentry);
1836 if (IS_ERR(odentry))
1840 if (!odentry->d_inode)
1843 if (odentry == trap)
1846 ndentry = lookup_one_len(tname, tdentry, tlen);
1847 host_err = PTR_ERR(ndentry);
1848 if (IS_ERR(ndentry))
1850 host_err = -ENOTEMPTY;
1851 if (ndentry == trap)
1855 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1857 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1860 host_err = nfsd_break_lease(odentry->d_inode);
1863 if (ndentry->d_inode) {
1864 host_err = nfsd_break_lease(ndentry->d_inode);
1868 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL);
1870 host_err = commit_metadata(tfhp);
1872 host_err = commit_metadata(ffhp);
1879 err = nfserrno(host_err);
1881 /* we cannot reply on fh_unlock on the two filehandles,
1882 * as that would do the wrong thing if the two directories
1883 * were the same, so again we do it by hand
1885 fill_post_wcc(ffhp);
1886 fill_post_wcc(tfhp);
1887 unlock_rename(tdentry, fdentry);
1888 ffhp->fh_locked = tfhp->fh_locked = 0;
1889 fh_drop_write(ffhp);
1896 * Unlink a file or directory
1897 * N.B. After this call fhp needs an fh_put
1900 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1901 char *fname, int flen)
1903 struct dentry *dentry, *rdentry;
1909 if (!flen || isdotent(fname, flen))
1911 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1915 host_err = fh_want_write(fhp);
1919 fh_lock_nested(fhp, I_MUTEX_PARENT);
1920 dentry = fhp->fh_dentry;
1921 dirp = dentry->d_inode;
1923 rdentry = lookup_one_len(fname, dentry, flen);
1924 host_err = PTR_ERR(rdentry);
1925 if (IS_ERR(rdentry))
1928 if (!rdentry->d_inode) {
1935 type = rdentry->d_inode->i_mode & S_IFMT;
1937 host_err = nfsd_break_lease(rdentry->d_inode);
1940 if (type != S_IFDIR)
1941 host_err = vfs_unlink(dirp, rdentry, NULL);
1943 host_err = vfs_rmdir(dirp, rdentry);
1945 host_err = commit_metadata(fhp);
1950 err = nfserrno(host_err);
1956 * We do this buffering because we must not call back into the file
1957 * system's ->lookup() method from the filldir callback. That may well
1958 * deadlock a number of file systems.
1960 * This is based heavily on the implementation of same in XFS.
1962 struct buffered_dirent {
1966 unsigned int d_type;
1970 struct readdir_data {
1971 struct dir_context ctx;
1977 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1978 loff_t offset, u64 ino, unsigned int d_type)
1980 struct readdir_data *buf = __buf;
1981 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1982 unsigned int reclen;
1984 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1985 if (buf->used + reclen > PAGE_SIZE) {
1990 de->namlen = namlen;
1991 de->offset = offset;
1993 de->d_type = d_type;
1994 memcpy(de->name, name, namlen);
1995 buf->used += reclen;
2000 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
2001 struct readdir_cd *cdp, loff_t *offsetp)
2003 struct buffered_dirent *de;
2007 struct readdir_data buf = {
2008 .ctx.actor = nfsd_buffered_filldir,
2009 .dirent = (void *)__get_free_page(GFP_KERNEL)
2013 return nfserrno(-ENOMEM);
2018 struct inode *dir_inode = file_inode(file);
2019 unsigned int reclen;
2021 cdp->err = nfserr_eof; /* will be cleared on successful read */
2025 host_err = iterate_dir(file, &buf.ctx);
2038 * Various filldir functions may end up calling back into
2039 * lookup_one_len() and the file system's ->lookup() method.
2040 * These expect i_mutex to be held, as it would within readdir.
2042 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2046 de = (struct buffered_dirent *)buf.dirent;
2048 offset = de->offset;
2050 if (func(cdp, de->name, de->namlen, de->offset,
2051 de->ino, de->d_type))
2054 if (cdp->err != nfs_ok)
2057 reclen = ALIGN(sizeof(*de) + de->namlen,
2060 de = (struct buffered_dirent *)((char *)de + reclen);
2062 mutex_unlock(&dir_inode->i_mutex);
2063 if (size > 0) /* We bailed out early */
2066 offset = vfs_llseek(file, 0, SEEK_CUR);
2069 free_page((unsigned long)(buf.dirent));
2072 return nfserrno(host_err);
2079 * Read entries from a directory.
2080 * The NFSv3/4 verifier we ignore for now.
2083 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2084 struct readdir_cd *cdp, filldir_t func)
2088 loff_t offset = *offsetp;
2089 int may_flags = NFSD_MAY_READ;
2091 /* NFSv2 only supports 32 bit cookies */
2092 if (rqstp->rq_vers > 2)
2093 may_flags |= NFSD_MAY_64BIT_COOKIE;
2095 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2099 offset = vfs_llseek(file, offset, SEEK_SET);
2101 err = nfserrno((int)offset);
2105 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2107 if (err == nfserr_eof || err == nfserr_toosmall)
2108 err = nfs_ok; /* can still be found in ->err */
2116 * Get file system stats
2117 * N.B. After this call fhp needs an fh_put
2120 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2124 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2126 struct path path = {
2127 .mnt = fhp->fh_export->ex_path.mnt,
2128 .dentry = fhp->fh_dentry,
2130 if (vfs_statfs(&path, stat))
2136 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2138 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2142 * Check for a user's access permissions to this inode.
2145 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2146 struct dentry *dentry, int acc)
2148 struct inode *inode = dentry->d_inode;
2151 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2154 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2156 (acc & NFSD_MAY_READ)? " read" : "",
2157 (acc & NFSD_MAY_WRITE)? " write" : "",
2158 (acc & NFSD_MAY_EXEC)? " exec" : "",
2159 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2160 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2161 (acc & NFSD_MAY_LOCK)? " lock" : "",
2162 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2164 IS_IMMUTABLE(inode)? " immut" : "",
2165 IS_APPEND(inode)? " append" : "",
2166 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2167 dprintk(" owner %d/%d user %d/%d\n",
2168 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2171 /* Normally we reject any write/sattr etc access on a read-only file
2172 * system. But if it is IRIX doing check on write-access for a
2173 * device special file, we ignore rofs.
2175 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2176 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2177 if (exp_rdonly(rqstp, exp) ||
2178 __mnt_is_readonly(exp->ex_path.mnt))
2180 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2183 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2186 if (acc & NFSD_MAY_LOCK) {
2187 /* If we cannot rely on authentication in NLM requests,
2188 * just allow locks, otherwise require read permission, or
2191 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2194 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2197 * The file owner always gets access permission for accesses that
2198 * would normally be checked at open time. This is to make
2199 * file access work even when the client has done a fchmod(fd, 0).
2201 * However, `cp foo bar' should fail nevertheless when bar is
2202 * readonly. A sensible way to do this might be to reject all
2203 * attempts to truncate a read-only file, because a creat() call
2204 * always implies file truncation.
2205 * ... but this isn't really fair. A process may reasonably call
2206 * ftruncate on an open file descriptor on a file with perm 000.
2207 * We must trust the client to do permission checking - using "ACCESS"
2210 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2211 uid_eq(inode->i_uid, current_fsuid()))
2214 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2215 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2217 /* Allow read access to binaries even when mode 111 */
2218 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2219 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2220 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2221 err = inode_permission(inode, MAY_EXEC);
2223 return err? nfserrno(err) : 0;
2227 nfsd_racache_shutdown(void)
2229 struct raparms *raparm, *last_raparm;
2232 dprintk("nfsd: freeing readahead buffers.\n");
2234 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2235 raparm = raparm_hash[i].pb_head;
2237 last_raparm = raparm;
2238 raparm = raparm->p_next;
2241 raparm_hash[i].pb_head = NULL;
2245 * Initialize readahead param cache
2248 nfsd_racache_init(int cache_size)
2253 struct raparms **raparm = NULL;
2256 if (raparm_hash[0].pb_head)
2258 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2261 cache_size = nperbucket * RAPARM_HASH_SIZE;
2263 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2265 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2266 spin_lock_init(&raparm_hash[i].pb_lock);
2268 raparm = &raparm_hash[i].pb_head;
2269 for (j = 0; j < nperbucket; j++) {
2270 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2273 raparm = &(*raparm)->p_next;
2278 nfsdstats.ra_size = cache_size;
2282 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2283 nfsd_racache_shutdown();
2287 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2289 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2291 struct inode *inode = fhp->fh_dentry->d_inode;
2295 struct posix_acl *acl;
2297 if (!IS_POSIXACL(inode))
2298 return ERR_PTR(-EOPNOTSUPP);
2301 case ACL_TYPE_ACCESS:
2302 name = POSIX_ACL_XATTR_ACCESS;
2304 case ACL_TYPE_DEFAULT:
2305 name = POSIX_ACL_XATTR_DEFAULT;
2308 return ERR_PTR(-EOPNOTSUPP);
2311 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2313 return ERR_PTR(size);
2315 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2321 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2323 struct inode *inode = fhp->fh_dentry->d_inode;
2329 if (!IS_POSIXACL(inode) ||
2330 !inode->i_op->setxattr || !inode->i_op->removexattr)
2333 case ACL_TYPE_ACCESS:
2334 name = POSIX_ACL_XATTR_ACCESS;
2336 case ACL_TYPE_DEFAULT:
2337 name = POSIX_ACL_XATTR_DEFAULT;
2343 if (acl && acl->a_count) {
2344 size = posix_acl_xattr_size(acl->a_count);
2345 value = kmalloc(size, GFP_KERNEL);
2348 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2355 error = fh_want_write(fhp);
2359 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2361 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2364 error = vfs_removexattr(fhp->fh_dentry, name);
2365 if (error == -ENODATA)
2375 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */