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 * In the nfsd4_open() case, this may be held across
212 * subsequent open and delegation acquisition which may
213 * need to take the child's i_mutex:
215 fh_lock_nested(fhp, I_MUTEX_PARENT);
216 dentry = lookup_one_len(name, dparent, len);
217 host_err = PTR_ERR(dentry);
221 * check if we have crossed a mount point ...
223 if (nfsd_mountpoint(dentry, exp)) {
224 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
230 *dentry_ret = dentry;
236 return nfserrno(host_err);
240 * Look up one component of a pathname.
241 * N.B. After this call _both_ fhp and resfh need an fh_put
243 * If the lookup would cross a mountpoint, and the mounted filesystem
244 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
245 * accepted as it stands and the mounted directory is
246 * returned. Otherwise the covered directory is returned.
247 * NOTE: this mountpoint crossing is not supported properly by all
248 * clients and is explicitly disallowed for NFSv3
249 * NeilBrown <neilb@cse.unsw.edu.au>
252 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
253 unsigned int len, struct svc_fh *resfh)
255 struct svc_export *exp;
256 struct dentry *dentry;
259 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
262 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
265 err = check_nfsd_access(exp, rqstp);
269 * Note: we compose the file handle now, but as the
270 * dentry may be negative, it may need to be updated.
272 err = fh_compose(resfh, exp, dentry, fhp);
273 if (!err && !dentry->d_inode)
282 * Commit metadata changes to stable storage.
285 commit_metadata(struct svc_fh *fhp)
287 struct inode *inode = fhp->fh_dentry->d_inode;
288 const struct export_operations *export_ops = inode->i_sb->s_export_op;
290 if (!EX_ISSYNC(fhp->fh_export))
293 if (export_ops->commit_metadata)
294 return export_ops->commit_metadata(inode);
295 return sync_inode_metadata(inode, 1);
299 * Go over the attributes and take care of the small differences between
300 * NFS semantics and what Linux expects.
303 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
306 * NFSv2 does not differentiate between "set-[ac]time-to-now"
307 * which only requires access, and "set-[ac]time-to-X" which
308 * requires ownership.
309 * So if it looks like it might be "set both to the same time which
310 * is close to now", and if inode_change_ok fails, then we
311 * convert to "set to now" instead of "set to explicit time"
313 * We only call inode_change_ok as the last test as technically
314 * it is not an interface that we should be using.
316 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
317 #define MAX_TOUCH_TIME_ERROR (30*60)
318 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
319 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
323 * Now just make sure time is in the right ballpark.
324 * Solaris, at least, doesn't seem to care what the time
325 * request is. We require it be within 30 minutes of now.
327 time_t delta = iap->ia_atime.tv_sec - get_seconds();
330 if (delta < MAX_TOUCH_TIME_ERROR &&
331 inode_change_ok(inode, iap) != 0) {
333 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
334 * This will cause notify_change to set these times
337 iap->ia_valid &= ~BOTH_TIME_SET;
341 /* sanitize the mode change */
342 if (iap->ia_valid & ATTR_MODE) {
343 iap->ia_mode &= S_IALLUGO;
344 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
347 /* Revoke setuid/setgid on chown */
348 if (!S_ISDIR(inode->i_mode) &&
349 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
350 iap->ia_valid |= ATTR_KILL_PRIV;
351 if (iap->ia_valid & ATTR_MODE) {
352 /* we're setting mode too, just clear the s*id bits */
353 iap->ia_mode &= ~S_ISUID;
354 if (iap->ia_mode & S_IXGRP)
355 iap->ia_mode &= ~S_ISGID;
357 /* set ATTR_KILL_* bits and let VFS handle it */
358 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
364 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
367 struct inode *inode = fhp->fh_dentry->d_inode;
370 if (iap->ia_size < inode->i_size) {
373 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
374 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
379 host_err = get_write_access(inode);
383 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
385 goto out_put_write_access;
388 out_put_write_access:
389 put_write_access(inode);
391 return nfserrno(host_err);
395 * Set various file attributes. After this call fhp needs an fh_put.
398 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
399 int check_guard, time_t guardtime)
401 struct dentry *dentry;
403 int accmode = NFSD_MAY_SATTR;
409 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
410 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
411 if (iap->ia_valid & ATTR_SIZE)
415 err = fh_verify(rqstp, fhp, ftype, accmode);
419 dentry = fhp->fh_dentry;
420 inode = dentry->d_inode;
422 /* Ignore any mode updates on symlinks */
423 if (S_ISLNK(inode->i_mode))
424 iap->ia_valid &= ~ATTR_MODE;
429 nfsd_sanitize_attrs(inode, iap);
432 * The size case is special, it changes the file in addition to the
435 if (iap->ia_valid & ATTR_SIZE) {
436 err = nfsd_get_write_access(rqstp, fhp, iap);
442 iap->ia_valid |= ATTR_CTIME;
444 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
445 err = nfserr_notsync;
446 goto out_put_write_access;
450 host_err = notify_change(dentry, iap, NULL);
453 out_put_write_access:
455 put_write_access(inode);
457 commit_metadata(fhp);
462 #if defined(CONFIG_NFSD_V2_ACL) || \
463 defined(CONFIG_NFSD_V3_ACL) || \
464 defined(CONFIG_NFSD_V4)
465 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
470 buflen = vfs_getxattr(dentry, key, NULL, 0);
474 *buf = kmalloc(buflen, GFP_KERNEL);
478 ret = vfs_getxattr(dentry, key, *buf, buflen);
485 #if defined(CONFIG_NFSD_V4)
487 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
494 buflen = posix_acl_xattr_size(pacl->a_count);
495 buf = kmalloc(buflen, GFP_KERNEL);
500 len = posix_acl_to_xattr(&init_user_ns, pacl, buf, buflen);
506 error = vfs_setxattr(dentry, key, buf, len, 0);
513 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
514 struct nfs4_acl *acl)
518 struct dentry *dentry;
520 struct posix_acl *pacl = NULL, *dpacl = NULL;
521 unsigned int flags = 0;
524 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_SATTR);
528 dentry = fhp->fh_dentry;
529 inode = dentry->d_inode;
530 if (S_ISDIR(inode->i_mode))
531 flags = NFS4_ACL_DIR;
533 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
534 if (host_error == -EINVAL) {
535 return nfserr_attrnotsupp;
536 } else if (host_error < 0)
539 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
543 if (S_ISDIR(inode->i_mode))
544 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
547 posix_acl_release(pacl);
548 posix_acl_release(dpacl);
550 if (host_error == -EOPNOTSUPP)
551 return nfserr_attrnotsupp;
553 return nfserrno(host_error);
556 static struct posix_acl *
557 _get_posix_acl(struct dentry *dentry, char *key)
560 struct posix_acl *pacl = NULL;
563 buflen = nfsd_getxattr(dentry, key, &buf);
567 return ERR_PTR(buflen);
569 pacl = posix_acl_from_xattr(&init_user_ns, buf, buflen);
575 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
577 struct inode *inode = dentry->d_inode;
579 struct posix_acl *pacl = NULL, *dpacl = NULL;
580 unsigned int flags = 0;
582 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
583 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
584 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
586 error = PTR_ERR(pacl);
591 if (S_ISDIR(inode->i_mode)) {
592 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
593 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
595 else if (IS_ERR(dpacl)) {
596 error = PTR_ERR(dpacl);
600 flags = NFS4_ACL_DIR;
603 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
605 error = PTR_ERR(*acl);
609 posix_acl_release(pacl);
610 posix_acl_release(dpacl);
615 * NFS junction information is stored in an extended attribute.
617 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
620 * nfsd4_is_junction - Test if an object could be an NFS junction
622 * @dentry: object to test
624 * Returns 1 if "dentry" appears to contain NFS junction information.
625 * Otherwise 0 is returned.
627 int nfsd4_is_junction(struct dentry *dentry)
629 struct inode *inode = dentry->d_inode;
633 if (inode->i_mode & S_IXUGO)
635 if (!(inode->i_mode & S_ISVTX))
637 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
641 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
642 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
643 struct xdr_netobj *label)
647 struct dentry *dentry;
649 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
653 dentry = fhp->fh_dentry;
655 mutex_lock(&dentry->d_inode->i_mutex);
656 host_error = security_inode_setsecctx(dentry, label->data, label->len);
657 mutex_unlock(&dentry->d_inode->i_mutex);
658 return nfserrno(host_error);
661 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
662 struct xdr_netobj *label)
664 return nfserr_notsupp;
668 #endif /* defined(CONFIG_NFSD_V4) */
670 #ifdef CONFIG_NFSD_V3
672 * Check server access rights to a file system object
678 static struct accessmap nfs3_regaccess[] = {
679 { NFS3_ACCESS_READ, NFSD_MAY_READ },
680 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
681 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
682 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
687 static struct accessmap nfs3_diraccess[] = {
688 { NFS3_ACCESS_READ, NFSD_MAY_READ },
689 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
690 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
691 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
692 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
697 static struct accessmap nfs3_anyaccess[] = {
698 /* Some clients - Solaris 2.6 at least, make an access call
699 * to the server to check for access for things like /dev/null
700 * (which really, the server doesn't care about). So
701 * We provide simple access checking for them, looking
702 * mainly at mode bits, and we make sure to ignore read-only
705 { NFS3_ACCESS_READ, NFSD_MAY_READ },
706 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
707 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
708 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
714 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
716 struct accessmap *map;
717 struct svc_export *export;
718 struct dentry *dentry;
719 u32 query, result = 0, sresult = 0;
722 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
726 export = fhp->fh_export;
727 dentry = fhp->fh_dentry;
729 if (S_ISREG(dentry->d_inode->i_mode))
730 map = nfs3_regaccess;
731 else if (S_ISDIR(dentry->d_inode->i_mode))
732 map = nfs3_diraccess;
734 map = nfs3_anyaccess;
738 for (; map->access; map++) {
739 if (map->access & query) {
742 sresult |= map->access;
744 err2 = nfsd_permission(rqstp, export, dentry, map->how);
747 result |= map->access;
750 /* the following error codes just mean the access was not allowed,
751 * rather than an error occurred */
755 /* simply don't "or" in the access bit. */
765 *supported = sresult;
770 #endif /* CONFIG_NFSD_V3 */
772 static int nfsd_open_break_lease(struct inode *inode, int access)
776 if (access & NFSD_MAY_NOT_BREAK_LEASE)
778 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
779 return break_lease(inode, mode | O_NONBLOCK);
783 * Open an existing file or directory.
784 * The may_flags argument indicates the type of open (read/write/lock)
785 * and additional flags.
786 * N.B. After this call fhp needs an fh_put
789 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
790 int may_flags, struct file **filp)
794 int flags = O_RDONLY|O_LARGEFILE;
798 validate_process_creds();
801 * If we get here, then the client has already done an "open",
802 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
803 * in case a chmod has now revoked permission.
805 * Arguably we should also allow the owner override for
806 * directories, but we never have and it doesn't seem to have
807 * caused anyone a problem. If we were to change this, note
808 * also that our filldir callbacks would need a variant of
809 * lookup_one_len that doesn't check permissions.
812 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
813 err = fh_verify(rqstp, fhp, type, may_flags);
817 path.mnt = fhp->fh_export->ex_path.mnt;
818 path.dentry = fhp->fh_dentry;
819 inode = path.dentry->d_inode;
821 /* Disallow write access to files with the append-only bit set
822 * or any access when mandatory locking enabled
825 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
828 * We must ignore files (but only files) which might have mandatory
829 * locks on them because there is no way to know if the accesser has
832 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
838 host_err = nfsd_open_break_lease(inode, may_flags);
839 if (host_err) /* NOMEM or WOULDBLOCK */
842 if (may_flags & NFSD_MAY_WRITE) {
843 if (may_flags & NFSD_MAY_READ)
844 flags = O_RDWR|O_LARGEFILE;
846 flags = O_WRONLY|O_LARGEFILE;
848 *filp = dentry_open(&path, flags, current_cred());
850 host_err = PTR_ERR(*filp);
853 host_err = ima_file_check(*filp, may_flags);
855 if (may_flags & NFSD_MAY_64BIT_COOKIE)
856 (*filp)->f_mode |= FMODE_64BITHASH;
858 (*filp)->f_mode |= FMODE_32BITHASH;
862 err = nfserrno(host_err);
864 validate_process_creds();
872 nfsd_close(struct file *filp)
878 * Obtain the readahead parameters for the file
879 * specified by (dev, ino).
882 static inline struct raparms *
883 nfsd_get_raparms(dev_t dev, ino_t ino)
885 struct raparms *ra, **rap, **frap = NULL;
888 struct raparm_hbucket *rab;
890 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
891 rab = &raparm_hash[hash];
893 spin_lock(&rab->pb_lock);
894 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
895 if (ra->p_ino == ino && ra->p_dev == dev)
898 if (ra->p_count == 0)
901 depth = nfsdstats.ra_size;
903 spin_unlock(&rab->pb_lock);
913 if (rap != &rab->pb_head) {
915 ra->p_next = rab->pb_head;
919 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
920 spin_unlock(&rab->pb_lock);
925 * Grab and keep cached pages associated with a file in the svc_rqst
926 * so that they can be passed to the network sendmsg/sendpage routines
927 * directly. They will be released after the sending has completed.
930 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
931 struct splice_desc *sd)
933 struct svc_rqst *rqstp = sd->u.data;
934 struct page **pp = rqstp->rq_next_page;
935 struct page *page = buf->page;
940 if (rqstp->rq_res.page_len == 0) {
942 put_page(*rqstp->rq_next_page);
943 *(rqstp->rq_next_page++) = page;
944 rqstp->rq_res.page_base = buf->offset;
945 rqstp->rq_res.page_len = size;
946 } else if (page != pp[-1]) {
948 if (*rqstp->rq_next_page)
949 put_page(*rqstp->rq_next_page);
950 *(rqstp->rq_next_page++) = page;
951 rqstp->rq_res.page_len += size;
953 rqstp->rq_res.page_len += size;
958 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
959 struct splice_desc *sd)
961 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
965 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
966 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
974 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
975 struct splice_desc sd = {
982 rqstp->rq_next_page = rqstp->rq_respages + 1;
983 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
987 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
992 nfsdstats.io_read += host_err;
995 fsnotify_access(file);
997 err = nfserrno(host_err);
1001 static void kill_suid(struct dentry *dentry)
1004 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
1006 mutex_lock(&dentry->d_inode->i_mutex);
1008 * Note we call this on write, so notify_change will not
1009 * encounter any conflicting delegations:
1011 notify_change(dentry, &ia, NULL);
1012 mutex_unlock(&dentry->d_inode->i_mutex);
1016 * Gathered writes: If another process is currently writing to the file,
1017 * there's a high chance this is another nfsd (triggered by a bulk write
1018 * from a client's biod). Rather than syncing the file with each write
1019 * request, we sleep for 10 msec.
1021 * I don't know if this roughly approximates C. Juszak's idea of
1022 * gathered writes, but it's a nice and simple solution (IMHO), and it
1025 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1026 * better tool (separate unstable writes and commits) for solving this
1029 static int wait_for_concurrent_writes(struct file *file)
1031 struct inode *inode = file_inode(file);
1032 static ino_t last_ino;
1033 static dev_t last_dev;
1036 if (atomic_read(&inode->i_writecount) > 1
1037 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1038 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1040 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1043 if (inode->i_state & I_DIRTY) {
1044 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1045 err = vfs_fsync(file, 0);
1047 last_ino = inode->i_ino;
1048 last_dev = inode->i_sb->s_dev;
1053 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1054 loff_t offset, struct kvec *vec, int vlen,
1055 unsigned long *cnt, int *stablep)
1057 struct svc_export *exp;
1058 struct dentry *dentry;
1059 struct inode *inode;
1063 int stable = *stablep;
1065 loff_t pos = offset;
1067 dentry = file->f_path.dentry;
1068 inode = dentry->d_inode;
1069 exp = fhp->fh_export;
1071 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1073 if (!EX_ISSYNC(exp))
1076 /* Write the data. */
1077 oldfs = get_fs(); set_fs(KERNEL_DS);
1078 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
1083 nfsdstats.io_write += host_err;
1084 fsnotify_modify(file);
1086 /* clear setuid/setgid flag after write */
1087 if (inode->i_mode & (S_ISUID | S_ISGID))
1092 host_err = wait_for_concurrent_writes(file);
1094 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
1098 dprintk("nfsd: write complete host_err=%d\n", host_err);
1102 err = nfserrno(host_err);
1107 * Read data from a file. count must contain the requested read count
1108 * on entry. On return, *count contains the number of bytes actually read.
1109 * N.B. After this call fhp needs an fh_put
1111 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1112 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1115 struct inode *inode;
1119 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1123 inode = file_inode(file);
1125 /* Get readahead parameters */
1126 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
1128 if (ra && ra->p_set)
1129 file->f_ra = ra->p_ra;
1131 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1133 /* Write back readahead params */
1135 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
1136 spin_lock(&rab->pb_lock);
1137 ra->p_ra = file->f_ra;
1140 spin_unlock(&rab->pb_lock);
1147 /* As above, but use the provided file descriptor. */
1149 nfsd_read_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1150 loff_t offset, struct kvec *vec, int vlen,
1151 unsigned long *count)
1156 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1157 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1160 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1161 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1162 err = nfsd_read(rqstp, fhp, offset, vec, vlen, count);
1168 * Write data to a file.
1169 * The stable flag requests synchronous writes.
1170 * N.B. After this call fhp needs an fh_put
1173 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1174 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1180 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1181 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1184 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1187 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1192 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1200 #ifdef CONFIG_NFSD_V3
1202 * Commit all pending writes to stable storage.
1204 * Note: we only guarantee that data that lies within the range specified
1205 * by the 'offset' and 'count' parameters will be synced.
1207 * Unfortunately we cannot lock the file to make sure we return full WCC
1208 * data to the client, as locking happens lower down in the filesystem.
1211 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1212 loff_t offset, unsigned long count)
1215 loff_t end = LLONG_MAX;
1216 __be32 err = nfserr_inval;
1221 end = offset + (loff_t)count - 1;
1226 err = nfsd_open(rqstp, fhp, S_IFREG,
1227 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1230 if (EX_ISSYNC(fhp->fh_export)) {
1231 int err2 = vfs_fsync_range(file, offset, end, 0);
1233 if (err2 != -EINVAL)
1234 err = nfserrno(err2);
1236 err = nfserr_notsupp;
1243 #endif /* CONFIG_NFSD_V3 */
1246 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1250 * Mode has already been set earlier in create:
1252 iap->ia_valid &= ~ATTR_MODE;
1254 * Setting uid/gid works only for root. Irix appears to
1255 * send along the gid on create when it tries to implement
1256 * setgid directories via NFS:
1258 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1259 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1261 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1265 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1266 * setting size to 0 may fail for some specific file systems by the permission
1267 * checking which requires WRITE permission but the mode is 000.
1268 * we ignore the resizing(to 0) on the just new created file, since the size is
1269 * 0 after file created.
1271 * call this only after vfs_create() is called.
1274 nfsd_check_ignore_resizing(struct iattr *iap)
1276 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1277 iap->ia_valid &= ~ATTR_SIZE;
1281 * Create a file (regular, directory, device, fifo); UNIX sockets
1282 * not yet implemented.
1283 * If the response fh has been verified, the parent directory should
1284 * already be locked. Note that the parent directory is left locked.
1286 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1289 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1290 char *fname, int flen, struct iattr *iap,
1291 int type, dev_t rdev, struct svc_fh *resfhp)
1293 struct dentry *dentry, *dchild = NULL;
1303 if (isdotent(fname, flen))
1306 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1310 dentry = fhp->fh_dentry;
1311 dirp = dentry->d_inode;
1313 err = nfserr_notdir;
1314 if (!dirp->i_op->lookup)
1317 * Check whether the response file handle has been verified yet.
1318 * If it has, the parent directory should already be locked.
1320 if (!resfhp->fh_dentry) {
1321 host_err = fh_want_write(fhp);
1325 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1326 fh_lock_nested(fhp, I_MUTEX_PARENT);
1327 dchild = lookup_one_len(fname, dentry, flen);
1328 host_err = PTR_ERR(dchild);
1331 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1335 /* called from nfsd_proc_create */
1336 dchild = dget(resfhp->fh_dentry);
1337 if (!fhp->fh_locked) {
1338 /* not actually possible */
1340 "nfsd_create: parent %pd2 not locked!\n",
1347 * Make sure the child dentry is still negative ...
1350 if (dchild->d_inode) {
1351 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1356 if (!(iap->ia_valid & ATTR_MODE))
1358 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1361 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1362 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1368 * Get the dir op function pointer.
1374 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1376 nfsd_check_ignore_resizing(iap);
1379 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1385 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1391 err = nfsd_create_setattr(rqstp, resfhp, iap);
1394 * nfsd_setattr already committed the child. Transactional filesystems
1395 * had a chance to commit changes for both parent and child
1396 * simultaneously making the following commit_metadata a noop.
1398 err2 = nfserrno(commit_metadata(fhp));
1402 * Update the file handle to get the new inode info.
1405 err = fh_update(resfhp);
1407 if (dchild && !IS_ERR(dchild))
1412 err = nfserrno(host_err);
1416 #ifdef CONFIG_NFSD_V3
1418 static inline int nfsd_create_is_exclusive(int createmode)
1420 return createmode == NFS3_CREATE_EXCLUSIVE
1421 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1425 * NFSv3 and NFSv4 version of nfsd_create
1428 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1429 char *fname, int flen, struct iattr *iap,
1430 struct svc_fh *resfhp, int createmode, u32 *verifier,
1431 bool *truncp, bool *created)
1433 struct dentry *dentry, *dchild = NULL;
1437 __u32 v_mtime=0, v_atime=0;
1443 if (isdotent(fname, flen))
1445 if (!(iap->ia_valid & ATTR_MODE))
1447 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1451 dentry = fhp->fh_dentry;
1452 dirp = dentry->d_inode;
1454 /* Get all the sanity checks out of the way before
1455 * we lock the parent. */
1456 err = nfserr_notdir;
1457 if (!dirp->i_op->lookup)
1460 host_err = fh_want_write(fhp);
1464 fh_lock_nested(fhp, I_MUTEX_PARENT);
1467 * Compose the response file handle.
1469 dchild = lookup_one_len(fname, dentry, flen);
1470 host_err = PTR_ERR(dchild);
1474 /* If file doesn't exist, check for permissions to create one */
1475 if (!dchild->d_inode) {
1476 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1481 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1485 if (nfsd_create_is_exclusive(createmode)) {
1486 /* solaris7 gets confused (bugid 4218508) if these have
1487 * the high bit set, so just clear the high bits. If this is
1488 * ever changed to use different attrs for storing the
1489 * verifier, then do_open_lookup() will also need to be fixed
1492 v_mtime = verifier[0]&0x7fffffff;
1493 v_atime = verifier[1]&0x7fffffff;
1496 if (dchild->d_inode) {
1499 switch (createmode) {
1500 case NFS3_CREATE_UNCHECKED:
1501 if (! S_ISREG(dchild->d_inode->i_mode))
1504 /* in nfsv4, we need to treat this case a little
1505 * differently. we don't want to truncate the
1506 * file now; this would be wrong if the OPEN
1507 * fails for some other reason. furthermore,
1508 * if the size is nonzero, we should ignore it
1509 * according to spec!
1511 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1514 iap->ia_valid &= ATTR_SIZE;
1518 case NFS3_CREATE_EXCLUSIVE:
1519 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1520 && dchild->d_inode->i_atime.tv_sec == v_atime
1521 && dchild->d_inode->i_size == 0 ) {
1526 case NFS4_CREATE_EXCLUSIVE4_1:
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 ) {
1535 case NFS3_CREATE_GUARDED:
1542 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1550 nfsd_check_ignore_resizing(iap);
1552 if (nfsd_create_is_exclusive(createmode)) {
1553 /* Cram the verifier into atime/mtime */
1554 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1555 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1556 /* XXX someone who knows this better please fix it for nsec */
1557 iap->ia_mtime.tv_sec = v_mtime;
1558 iap->ia_atime.tv_sec = v_atime;
1559 iap->ia_mtime.tv_nsec = 0;
1560 iap->ia_atime.tv_nsec = 0;
1564 err = nfsd_create_setattr(rqstp, resfhp, iap);
1567 * nfsd_setattr already committed the child (and possibly also the parent).
1570 err = nfserrno(commit_metadata(fhp));
1573 * Update the filehandle to get the new inode info.
1576 err = fh_update(resfhp);
1580 if (dchild && !IS_ERR(dchild))
1586 err = nfserrno(host_err);
1589 #endif /* CONFIG_NFSD_V3 */
1592 * Read a symlink. On entry, *lenp must contain the maximum path length that
1593 * fits into the buffer. On return, it contains the true length.
1594 * N.B. After this call fhp needs an fh_put
1597 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1599 struct inode *inode;
1605 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1609 path.mnt = fhp->fh_export->ex_path.mnt;
1610 path.dentry = fhp->fh_dentry;
1611 inode = path.dentry->d_inode;
1614 if (!inode->i_op->readlink)
1618 /* N.B. Why does this call need a get_fs()??
1619 * Remove the set_fs and watch the fireworks:-) --okir
1622 oldfs = get_fs(); set_fs(KERNEL_DS);
1623 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1634 err = nfserrno(host_err);
1639 * Create a symlink and look up its inode
1640 * N.B. After this call _both_ fhp and resfhp need an fh_put
1643 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1644 char *fname, int flen,
1645 char *path, int plen,
1646 struct svc_fh *resfhp,
1649 struct dentry *dentry, *dnew;
1657 if (isdotent(fname, flen))
1660 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1664 host_err = fh_want_write(fhp);
1669 dentry = fhp->fh_dentry;
1670 dnew = lookup_one_len(fname, dentry, flen);
1671 host_err = PTR_ERR(dnew);
1675 if (unlikely(path[plen] != 0)) {
1676 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1677 if (path_alloced == NULL)
1680 strncpy(path_alloced, path, plen);
1681 path_alloced[plen] = 0;
1682 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1683 kfree(path_alloced);
1686 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1687 err = nfserrno(host_err);
1689 err = nfserrno(commit_metadata(fhp));
1694 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1696 if (err==0) err = cerr;
1701 err = nfserrno(host_err);
1707 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1710 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1711 char *name, int len, struct svc_fh *tfhp)
1713 struct dentry *ddir, *dnew, *dold;
1718 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1721 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1725 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1731 if (isdotent(name, len))
1734 host_err = fh_want_write(tfhp);
1736 err = nfserrno(host_err);
1740 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1741 ddir = ffhp->fh_dentry;
1742 dirp = ddir->d_inode;
1744 dnew = lookup_one_len(name, ddir, len);
1745 host_err = PTR_ERR(dnew);
1749 dold = tfhp->fh_dentry;
1754 host_err = vfs_link(dold, dirp, dnew, NULL);
1756 err = nfserrno(commit_metadata(ffhp));
1758 err = nfserrno(commit_metadata(tfhp));
1760 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1763 err = nfserrno(host_err);
1769 fh_drop_write(tfhp);
1774 err = nfserrno(host_err);
1780 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1783 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1784 struct svc_fh *tfhp, char *tname, int tlen)
1786 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1787 struct inode *fdir, *tdir;
1791 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1794 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1798 fdentry = ffhp->fh_dentry;
1799 fdir = fdentry->d_inode;
1801 tdentry = tfhp->fh_dentry;
1802 tdir = tdentry->d_inode;
1805 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1808 host_err = fh_want_write(ffhp);
1810 err = nfserrno(host_err);
1814 /* cannot use fh_lock as we need deadlock protective ordering
1815 * so do it by hand */
1816 trap = lock_rename(tdentry, fdentry);
1817 ffhp->fh_locked = tfhp->fh_locked = 1;
1821 odentry = lookup_one_len(fname, fdentry, flen);
1822 host_err = PTR_ERR(odentry);
1823 if (IS_ERR(odentry))
1827 if (!odentry->d_inode)
1830 if (odentry == trap)
1833 ndentry = lookup_one_len(tname, tdentry, tlen);
1834 host_err = PTR_ERR(ndentry);
1835 if (IS_ERR(ndentry))
1837 host_err = -ENOTEMPTY;
1838 if (ndentry == trap)
1842 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1844 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1847 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL);
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 if (type != S_IFDIR)
1917 host_err = vfs_unlink(dirp, rdentry, NULL);
1919 host_err = vfs_rmdir(dirp, rdentry);
1921 host_err = commit_metadata(fhp);
1925 err = nfserrno(host_err);
1931 * We do this buffering because we must not call back into the file
1932 * system's ->lookup() method from the filldir callback. That may well
1933 * deadlock a number of file systems.
1935 * This is based heavily on the implementation of same in XFS.
1937 struct buffered_dirent {
1941 unsigned int d_type;
1945 struct readdir_data {
1946 struct dir_context ctx;
1952 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1953 loff_t offset, u64 ino, unsigned int d_type)
1955 struct readdir_data *buf = __buf;
1956 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1957 unsigned int reclen;
1959 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1960 if (buf->used + reclen > PAGE_SIZE) {
1965 de->namlen = namlen;
1966 de->offset = offset;
1968 de->d_type = d_type;
1969 memcpy(de->name, name, namlen);
1970 buf->used += reclen;
1975 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1976 struct readdir_cd *cdp, loff_t *offsetp)
1978 struct buffered_dirent *de;
1982 struct readdir_data buf = {
1983 .ctx.actor = nfsd_buffered_filldir,
1984 .dirent = (void *)__get_free_page(GFP_KERNEL)
1988 return nfserrno(-ENOMEM);
1993 struct inode *dir_inode = file_inode(file);
1994 unsigned int reclen;
1996 cdp->err = nfserr_eof; /* will be cleared on successful read */
2000 host_err = iterate_dir(file, &buf.ctx);
2013 * Various filldir functions may end up calling back into
2014 * lookup_one_len() and the file system's ->lookup() method.
2015 * These expect i_mutex to be held, as it would within readdir.
2017 host_err = mutex_lock_killable(&dir_inode->i_mutex);
2021 de = (struct buffered_dirent *)buf.dirent;
2023 offset = de->offset;
2025 if (func(cdp, de->name, de->namlen, de->offset,
2026 de->ino, de->d_type))
2029 if (cdp->err != nfs_ok)
2032 reclen = ALIGN(sizeof(*de) + de->namlen,
2035 de = (struct buffered_dirent *)((char *)de + reclen);
2037 mutex_unlock(&dir_inode->i_mutex);
2038 if (size > 0) /* We bailed out early */
2041 offset = vfs_llseek(file, 0, SEEK_CUR);
2044 free_page((unsigned long)(buf.dirent));
2047 return nfserrno(host_err);
2054 * Read entries from a directory.
2055 * The NFSv3/4 verifier we ignore for now.
2058 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2059 struct readdir_cd *cdp, filldir_t func)
2063 loff_t offset = *offsetp;
2064 int may_flags = NFSD_MAY_READ;
2066 /* NFSv2 only supports 32 bit cookies */
2067 if (rqstp->rq_vers > 2)
2068 may_flags |= NFSD_MAY_64BIT_COOKIE;
2070 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
2074 offset = vfs_llseek(file, offset, SEEK_SET);
2076 err = nfserrno((int)offset);
2080 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2082 if (err == nfserr_eof || err == nfserr_toosmall)
2083 err = nfs_ok; /* can still be found in ->err */
2091 * Get file system stats
2092 * N.B. After this call fhp needs an fh_put
2095 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2099 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2101 struct path path = {
2102 .mnt = fhp->fh_export->ex_path.mnt,
2103 .dentry = fhp->fh_dentry,
2105 if (vfs_statfs(&path, stat))
2111 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2113 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2117 * Check for a user's access permissions to this inode.
2120 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2121 struct dentry *dentry, int acc)
2123 struct inode *inode = dentry->d_inode;
2126 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
2129 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2131 (acc & NFSD_MAY_READ)? " read" : "",
2132 (acc & NFSD_MAY_WRITE)? " write" : "",
2133 (acc & NFSD_MAY_EXEC)? " exec" : "",
2134 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2135 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2136 (acc & NFSD_MAY_LOCK)? " lock" : "",
2137 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2139 IS_IMMUTABLE(inode)? " immut" : "",
2140 IS_APPEND(inode)? " append" : "",
2141 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2142 dprintk(" owner %d/%d user %d/%d\n",
2143 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2146 /* Normally we reject any write/sattr etc access on a read-only file
2147 * system. But if it is IRIX doing check on write-access for a
2148 * device special file, we ignore rofs.
2150 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2151 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2152 if (exp_rdonly(rqstp, exp) ||
2153 __mnt_is_readonly(exp->ex_path.mnt))
2155 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2158 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2161 if (acc & NFSD_MAY_LOCK) {
2162 /* If we cannot rely on authentication in NLM requests,
2163 * just allow locks, otherwise require read permission, or
2166 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2169 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2172 * The file owner always gets access permission for accesses that
2173 * would normally be checked at open time. This is to make
2174 * file access work even when the client has done a fchmod(fd, 0).
2176 * However, `cp foo bar' should fail nevertheless when bar is
2177 * readonly. A sensible way to do this might be to reject all
2178 * attempts to truncate a read-only file, because a creat() call
2179 * always implies file truncation.
2180 * ... but this isn't really fair. A process may reasonably call
2181 * ftruncate on an open file descriptor on a file with perm 000.
2182 * We must trust the client to do permission checking - using "ACCESS"
2185 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2186 uid_eq(inode->i_uid, current_fsuid()))
2189 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2190 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2192 /* Allow read access to binaries even when mode 111 */
2193 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2194 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2195 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2196 err = inode_permission(inode, MAY_EXEC);
2198 return err? nfserrno(err) : 0;
2202 nfsd_racache_shutdown(void)
2204 struct raparms *raparm, *last_raparm;
2207 dprintk("nfsd: freeing readahead buffers.\n");
2209 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2210 raparm = raparm_hash[i].pb_head;
2212 last_raparm = raparm;
2213 raparm = raparm->p_next;
2216 raparm_hash[i].pb_head = NULL;
2220 * Initialize readahead param cache
2223 nfsd_racache_init(int cache_size)
2228 struct raparms **raparm = NULL;
2231 if (raparm_hash[0].pb_head)
2233 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2236 cache_size = nperbucket * RAPARM_HASH_SIZE;
2238 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2240 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2241 spin_lock_init(&raparm_hash[i].pb_lock);
2243 raparm = &raparm_hash[i].pb_head;
2244 for (j = 0; j < nperbucket; j++) {
2245 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2248 raparm = &(*raparm)->p_next;
2253 nfsdstats.ra_size = cache_size;
2257 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2258 nfsd_racache_shutdown();
2262 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2264 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2266 struct inode *inode = fhp->fh_dentry->d_inode;
2270 struct posix_acl *acl;
2272 if (!IS_POSIXACL(inode))
2273 return ERR_PTR(-EOPNOTSUPP);
2276 case ACL_TYPE_ACCESS:
2277 name = POSIX_ACL_XATTR_ACCESS;
2279 case ACL_TYPE_DEFAULT:
2280 name = POSIX_ACL_XATTR_DEFAULT;
2283 return ERR_PTR(-EOPNOTSUPP);
2286 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2288 return ERR_PTR(size);
2290 acl = posix_acl_from_xattr(&init_user_ns, value, size);
2296 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2298 struct inode *inode = fhp->fh_dentry->d_inode;
2304 if (!IS_POSIXACL(inode) ||
2305 !inode->i_op->setxattr || !inode->i_op->removexattr)
2308 case ACL_TYPE_ACCESS:
2309 name = POSIX_ACL_XATTR_ACCESS;
2311 case ACL_TYPE_DEFAULT:
2312 name = POSIX_ACL_XATTR_DEFAULT;
2318 if (acl && acl->a_count) {
2319 size = posix_acl_xattr_size(acl->a_count);
2320 value = kmalloc(size, GFP_KERNEL);
2323 error = posix_acl_to_xattr(&init_user_ns, acl, value, size);
2330 error = fh_want_write(fhp);
2334 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2336 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2339 error = vfs_removexattr(fhp->fh_dentry, name);
2340 if (error == -ENODATA)
2350 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */