1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/quotaops.h>
24 #include <linux/fsnotify.h>
25 #include <linux/posix_acl_xattr.h>
26 #include <linux/xattr.h>
27 #include <linux/jhash.h>
28 #include <linux/ima.h>
29 #include <asm/uaccess.h>
33 #endif /* CONFIG_NFSD_V3 */
36 #include <linux/nfs4_acl.h>
37 #include <linux/nfsd_idmap.h>
38 #endif /* CONFIG_NFSD_V4 */
43 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
47 * This is a cache of readahead params that help us choose the proper
48 * readahead strategy. Initially, we set all readahead parameters to 0
49 * and let the VFS handle things.
50 * If you increase the number of cached files very much, you'll need to
51 * add a hash table here.
54 struct raparms *p_next;
59 struct file_ra_state p_ra;
60 unsigned int p_hindex;
63 struct raparm_hbucket {
64 struct raparms *pb_head;
66 } ____cacheline_aligned_in_smp;
68 #define RAPARM_HASH_BITS 4
69 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
70 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
71 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
74 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
76 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
77 * or nfs_ok having possibly changed *dpp and *expp
80 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
81 struct svc_export **expp)
83 struct svc_export *exp = *expp, *exp2 = NULL;
84 struct dentry *dentry = *dpp;
85 struct path path = {.mnt = mntget(exp->ex_path.mnt),
86 .dentry = dget(dentry)};
89 while (d_mountpoint(path.dentry) && follow_down(&path))
92 exp2 = rqst_exp_get_by_name(rqstp, &path);
96 * We normally allow NFS clients to continue
97 * "underneath" a mountpoint that is not exported.
98 * The exception is V4ROOT, where no traversal is ever
99 * allowed without an explicit export of the new
102 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
107 if (nfsd_v4client(rqstp) ||
108 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
109 /* successfully crossed mount point */
111 * This is subtle: path.dentry is *not* on path.mnt
112 * at this point. The only reason we are safe is that
113 * original mnt is pinned down by exp, so we should
114 * put path *before* putting exp
117 path.dentry = dentry;
127 static void follow_to_parent(struct path *path)
131 while (path->dentry == path->mnt->mnt_root && follow_up(path))
133 dp = dget_parent(path->dentry);
138 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
140 struct svc_export *exp2;
141 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
142 .dentry = dget(dparent)};
144 follow_to_parent(&path);
146 exp2 = rqst_exp_parent(rqstp, &path);
147 if (PTR_ERR(exp2) == -ENOENT) {
148 *dentryp = dget(dparent);
149 } else if (IS_ERR(exp2)) {
151 return PTR_ERR(exp2);
153 *dentryp = dget(path.dentry);
162 * For nfsd purposes, we treat V4ROOT exports as though there was an
163 * export at *every* directory.
165 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
167 if (d_mountpoint(dentry))
169 if (!(exp->ex_flags & NFSEXP_V4ROOT))
171 return dentry->d_inode != NULL;
175 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
176 const char *name, unsigned int len,
177 struct svc_export **exp_ret, struct dentry **dentry_ret)
179 struct svc_export *exp;
180 struct dentry *dparent;
181 struct dentry *dentry;
185 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
187 /* Obtain dentry and export. */
188 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
192 dparent = fhp->fh_dentry;
193 exp = fhp->fh_export;
196 /* Lookup the name, but don't follow links */
197 if (isdotent(name, len)) {
199 dentry = dget(dparent);
200 else if (dparent != exp->ex_path.dentry)
201 dentry = dget_parent(dparent);
202 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
203 dentry = dget(dparent); /* .. == . just like at / */
205 /* checking mountpoint crossing is very different when stepping up */
206 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
212 dentry = lookup_one_len(name, dparent, len);
213 host_err = PTR_ERR(dentry);
217 * check if we have crossed a mount point ...
219 if (nfsd_mountpoint(dentry, exp)) {
220 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
226 *dentry_ret = dentry;
232 return nfserrno(host_err);
236 * Look up one component of a pathname.
237 * N.B. After this call _both_ fhp and resfh need an fh_put
239 * If the lookup would cross a mountpoint, and the mounted filesystem
240 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
241 * accepted as it stands and the mounted directory is
242 * returned. Otherwise the covered directory is returned.
243 * NOTE: this mountpoint crossing is not supported properly by all
244 * clients and is explicitly disallowed for NFSv3
245 * NeilBrown <neilb@cse.unsw.edu.au>
248 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
249 unsigned int len, struct svc_fh *resfh)
251 struct svc_export *exp;
252 struct dentry *dentry;
255 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
258 err = check_nfsd_access(exp, rqstp);
262 * Note: we compose the file handle now, but as the
263 * dentry may be negative, it may need to be updated.
265 err = fh_compose(resfh, exp, dentry, fhp);
266 if (!err && !dentry->d_inode)
276 * Set various file attributes.
277 * N.B. After this call fhp needs an fh_put
280 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
281 int check_guard, time_t guardtime)
283 struct dentry *dentry;
285 int accmode = NFSD_MAY_SATTR;
291 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
292 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
293 if (iap->ia_valid & ATTR_SIZE)
297 err = fh_verify(rqstp, fhp, ftype, accmode);
301 dentry = fhp->fh_dentry;
302 inode = dentry->d_inode;
304 /* Ignore any mode updates on symlinks */
305 if (S_ISLNK(inode->i_mode))
306 iap->ia_valid &= ~ATTR_MODE;
312 * NFSv2 does not differentiate between "set-[ac]time-to-now"
313 * which only requires access, and "set-[ac]time-to-X" which
314 * requires ownership.
315 * So if it looks like it might be "set both to the same time which
316 * is close to now", and if inode_change_ok fails, then we
317 * convert to "set to now" instead of "set to explicit time"
319 * We only call inode_change_ok as the last test as technically
320 * it is not an interface that we should be using. It is only
321 * valid if the filesystem does not define it's own i_op->setattr.
323 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
324 #define MAX_TOUCH_TIME_ERROR (30*60)
325 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
326 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
330 * Now just make sure time is in the right ballpark.
331 * Solaris, at least, doesn't seem to care what the time
332 * request is. We require it be within 30 minutes of now.
334 time_t delta = iap->ia_atime.tv_sec - get_seconds();
337 if (delta < MAX_TOUCH_TIME_ERROR &&
338 inode_change_ok(inode, iap) != 0) {
340 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
341 * This will cause notify_change to set these times
344 iap->ia_valid &= ~BOTH_TIME_SET;
349 * The size case is special.
350 * It changes the file as well as the attributes.
352 if (iap->ia_valid & ATTR_SIZE) {
353 if (iap->ia_size < inode->i_size) {
354 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
355 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
361 * If we are changing the size of the file, then
362 * we need to break all leases.
364 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
365 if (host_err == -EWOULDBLOCK)
366 host_err = -ETIMEDOUT;
367 if (host_err) /* ENOMEM or EWOULDBLOCK */
370 host_err = get_write_access(inode);
375 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
377 put_write_access(inode);
383 /* sanitize the mode change */
384 if (iap->ia_valid & ATTR_MODE) {
385 iap->ia_mode &= S_IALLUGO;
386 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
389 /* Revoke setuid/setgid on chown */
390 if (!S_ISDIR(inode->i_mode) &&
391 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
392 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
393 iap->ia_valid |= ATTR_KILL_PRIV;
394 if (iap->ia_valid & ATTR_MODE) {
395 /* we're setting mode too, just clear the s*id bits */
396 iap->ia_mode &= ~S_ISUID;
397 if (iap->ia_mode & S_IXGRP)
398 iap->ia_mode &= ~S_ISGID;
400 /* set ATTR_KILL_* bits and let VFS handle it */
401 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
405 /* Change the attributes. */
407 iap->ia_valid |= ATTR_CTIME;
409 err = nfserr_notsync;
410 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
412 host_err = notify_change(dentry, iap);
413 err = nfserrno(host_err);
417 put_write_access(inode);
419 if (EX_ISSYNC(fhp->fh_export))
420 write_inode_now(inode, 1);
425 err = nfserrno(host_err);
429 #if defined(CONFIG_NFSD_V2_ACL) || \
430 defined(CONFIG_NFSD_V3_ACL) || \
431 defined(CONFIG_NFSD_V4)
432 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
437 buflen = vfs_getxattr(dentry, key, NULL, 0);
441 *buf = kmalloc(buflen, GFP_KERNEL);
445 ret = vfs_getxattr(dentry, key, *buf, buflen);
452 #if defined(CONFIG_NFSD_V4)
454 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
461 buflen = posix_acl_xattr_size(pacl->a_count);
462 buf = kmalloc(buflen, GFP_KERNEL);
467 len = posix_acl_to_xattr(pacl, buf, buflen);
473 error = vfs_setxattr(dentry, key, buf, len, 0);
480 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
481 struct nfs4_acl *acl)
485 struct dentry *dentry;
487 struct posix_acl *pacl = NULL, *dpacl = NULL;
488 unsigned int flags = 0;
491 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
495 dentry = fhp->fh_dentry;
496 inode = dentry->d_inode;
497 if (S_ISDIR(inode->i_mode))
498 flags = NFS4_ACL_DIR;
500 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
501 if (host_error == -EINVAL) {
502 return nfserr_attrnotsupp;
503 } else if (host_error < 0)
506 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
510 if (S_ISDIR(inode->i_mode))
511 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
514 posix_acl_release(pacl);
515 posix_acl_release(dpacl);
517 if (host_error == -EOPNOTSUPP)
518 return nfserr_attrnotsupp;
520 return nfserrno(host_error);
523 static struct posix_acl *
524 _get_posix_acl(struct dentry *dentry, char *key)
527 struct posix_acl *pacl = NULL;
530 buflen = nfsd_getxattr(dentry, key, &buf);
534 return ERR_PTR(buflen);
536 pacl = posix_acl_from_xattr(buf, buflen);
542 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
544 struct inode *inode = dentry->d_inode;
546 struct posix_acl *pacl = NULL, *dpacl = NULL;
547 unsigned int flags = 0;
549 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
550 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
551 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
553 error = PTR_ERR(pacl);
558 if (S_ISDIR(inode->i_mode)) {
559 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
560 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
562 else if (IS_ERR(dpacl)) {
563 error = PTR_ERR(dpacl);
567 flags = NFS4_ACL_DIR;
570 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
572 error = PTR_ERR(*acl);
576 posix_acl_release(pacl);
577 posix_acl_release(dpacl);
581 #endif /* defined(CONFIG_NFS_V4) */
583 #ifdef CONFIG_NFSD_V3
585 * Check server access rights to a file system object
591 static struct accessmap nfs3_regaccess[] = {
592 { NFS3_ACCESS_READ, NFSD_MAY_READ },
593 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
594 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
595 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
600 static struct accessmap nfs3_diraccess[] = {
601 { NFS3_ACCESS_READ, NFSD_MAY_READ },
602 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
603 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
604 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
605 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
610 static struct accessmap nfs3_anyaccess[] = {
611 /* Some clients - Solaris 2.6 at least, make an access call
612 * to the server to check for access for things like /dev/null
613 * (which really, the server doesn't care about). So
614 * We provide simple access checking for them, looking
615 * mainly at mode bits, and we make sure to ignore read-only
618 { NFS3_ACCESS_READ, NFSD_MAY_READ },
619 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
620 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
621 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
627 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
629 struct accessmap *map;
630 struct svc_export *export;
631 struct dentry *dentry;
632 u32 query, result = 0, sresult = 0;
635 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
639 export = fhp->fh_export;
640 dentry = fhp->fh_dentry;
642 if (S_ISREG(dentry->d_inode->i_mode))
643 map = nfs3_regaccess;
644 else if (S_ISDIR(dentry->d_inode->i_mode))
645 map = nfs3_diraccess;
647 map = nfs3_anyaccess;
651 for (; map->access; map++) {
652 if (map->access & query) {
655 sresult |= map->access;
657 err2 = nfsd_permission(rqstp, export, dentry, map->how);
660 result |= map->access;
663 /* the following error codes just mean the access was not allowed,
664 * rather than an error occurred */
668 /* simply don't "or" in the access bit. */
678 *supported = sresult;
683 #endif /* CONFIG_NFSD_V3 */
688 * Open an existing file or directory.
689 * The access argument indicates the type of open (read/write/lock)
690 * N.B. After this call fhp needs an fh_put
693 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
694 int access, struct file **filp)
696 struct dentry *dentry;
698 int flags = O_RDONLY|O_LARGEFILE;
702 validate_process_creds();
705 * If we get here, then the client has already done an "open",
706 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
707 * in case a chmod has now revoked permission.
709 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
713 dentry = fhp->fh_dentry;
714 inode = dentry->d_inode;
716 /* Disallow write access to files with the append-only bit set
717 * or any access when mandatory locking enabled
720 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
723 * We must ignore files (but only files) which might have mandatory
724 * locks on them because there is no way to know if the accesser has
727 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
734 * Check to see if there are any leases on this file.
735 * This may block while leases are broken.
737 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
738 if (host_err == -EWOULDBLOCK)
739 host_err = -ETIMEDOUT;
740 if (host_err) /* NOMEM or WOULDBLOCK */
743 if (access & NFSD_MAY_WRITE) {
744 if (access & NFSD_MAY_READ)
745 flags = O_RDWR|O_LARGEFILE;
747 flags = O_WRONLY|O_LARGEFILE;
751 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
752 flags, current_cred());
754 host_err = PTR_ERR(*filp);
756 ima_counts_get(*filp);
758 err = nfserrno(host_err);
760 validate_process_creds();
768 nfsd_close(struct file *filp)
775 * As this calls fsync (not fdatasync) there is no need for a write_inode
778 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
779 const struct file_operations *fop)
781 struct inode *inode = dp->d_inode;
782 int (*fsync) (struct file *, struct dentry *, int);
785 err = filemap_fdatawrite(inode->i_mapping);
786 if (err == 0 && fop && (fsync = fop->fsync))
787 err = fsync(filp, dp, 0);
789 err = filemap_fdatawait(inode->i_mapping);
795 nfsd_sync(struct file *filp)
798 struct inode *inode = filp->f_path.dentry->d_inode;
799 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
800 mutex_lock(&inode->i_mutex);
801 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
802 mutex_unlock(&inode->i_mutex);
808 nfsd_sync_dir(struct dentry *dp)
810 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
814 * Obtain the readahead parameters for the file
815 * specified by (dev, ino).
818 static inline struct raparms *
819 nfsd_get_raparms(dev_t dev, ino_t ino)
821 struct raparms *ra, **rap, **frap = NULL;
824 struct raparm_hbucket *rab;
826 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
827 rab = &raparm_hash[hash];
829 spin_lock(&rab->pb_lock);
830 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
831 if (ra->p_ino == ino && ra->p_dev == dev)
834 if (ra->p_count == 0)
837 depth = nfsdstats.ra_size*11/10;
839 spin_unlock(&rab->pb_lock);
849 if (rap != &rab->pb_head) {
851 ra->p_next = rab->pb_head;
855 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
856 spin_unlock(&rab->pb_lock);
861 * Grab and keep cached pages associated with a file in the svc_rqst
862 * so that they can be passed to the network sendmsg/sendpage routines
863 * directly. They will be released after the sending has completed.
866 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
867 struct splice_desc *sd)
869 struct svc_rqst *rqstp = sd->u.data;
870 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
871 struct page *page = buf->page;
875 ret = buf->ops->confirm(pipe, buf);
881 if (rqstp->rq_res.page_len == 0) {
886 rqstp->rq_res.page_base = buf->offset;
887 rqstp->rq_res.page_len = size;
888 } else if (page != pp[-1]) {
894 rqstp->rq_res.page_len += size;
896 rqstp->rq_res.page_len += size;
901 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
902 struct splice_desc *sd)
904 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
907 static inline int svc_msnfs(struct svc_fh *ffhp)
910 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
917 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
918 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
927 inode = file->f_path.dentry->d_inode;
929 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
932 /* Get readahead parameters */
933 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
936 file->f_ra = ra->p_ra;
938 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
939 struct splice_desc sd = {
946 rqstp->rq_resused = 1;
947 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
951 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
955 /* Write back readahead params */
957 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
958 spin_lock(&rab->pb_lock);
959 ra->p_ra = file->f_ra;
962 spin_unlock(&rab->pb_lock);
966 nfsdstats.io_read += host_err;
969 fsnotify_access(file->f_path.dentry);
971 err = nfserrno(host_err);
976 static void kill_suid(struct dentry *dentry)
979 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
981 mutex_lock(&dentry->d_inode->i_mutex);
982 notify_change(dentry, &ia);
983 mutex_unlock(&dentry->d_inode->i_mutex);
987 * Gathered writes: If another process is currently writing to the file,
988 * there's a high chance this is another nfsd (triggered by a bulk write
989 * from a client's biod). Rather than syncing the file with each write
990 * request, we sleep for 10 msec.
992 * I don't know if this roughly approximates C. Juszak's idea of
993 * gathered writes, but it's a nice and simple solution (IMHO), and it
996 * Note: we do this only in the NFSv2 case, since v3 and higher have a
997 * better tool (separate unstable writes and commits) for solving this
1000 static int wait_for_concurrent_writes(struct file *file)
1002 struct inode *inode = file->f_path.dentry->d_inode;
1003 static ino_t last_ino;
1004 static dev_t last_dev;
1007 if (atomic_read(&inode->i_writecount) > 1
1008 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1009 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1011 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1014 if (inode->i_state & I_DIRTY) {
1015 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1016 err = nfsd_sync(file);
1018 last_ino = inode->i_ino;
1019 last_dev = inode->i_sb->s_dev;
1024 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1025 loff_t offset, struct kvec *vec, int vlen,
1026 unsigned long *cnt, int *stablep)
1028 struct svc_export *exp;
1029 struct dentry *dentry;
1030 struct inode *inode;
1034 int stable = *stablep;
1040 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1041 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1045 dentry = file->f_path.dentry;
1046 inode = dentry->d_inode;
1047 exp = fhp->fh_export;
1050 * Request sync writes if
1051 * - the sync export option has been set, or
1052 * - the client requested O_SYNC behavior (NFSv3 feature).
1053 * - The file system doesn't support fsync().
1054 * When NFSv2 gathered writes have been configured for this volume,
1055 * flushing the data to disk is handled separately below.
1057 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1059 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1061 *stablep = 2; /* FILE_SYNC */
1064 if (!EX_ISSYNC(exp))
1066 if (stable && !use_wgather) {
1067 spin_lock(&file->f_lock);
1068 file->f_flags |= O_SYNC;
1069 spin_unlock(&file->f_lock);
1072 /* Write the data. */
1073 oldfs = get_fs(); set_fs(KERNEL_DS);
1074 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1079 nfsdstats.io_write += host_err;
1080 fsnotify_modify(file->f_path.dentry);
1082 /* clear setuid/setgid flag after write */
1083 if (inode->i_mode & (S_ISUID | S_ISGID))
1086 if (stable && use_wgather)
1087 host_err = wait_for_concurrent_writes(file);
1090 dprintk("nfsd: write complete host_err=%d\n", host_err);
1094 err = nfserrno(host_err);
1100 * Read data from a file. count must contain the requested read count
1101 * on entry. On return, *count contains the number of bytes actually read.
1102 * N.B. After this call fhp needs an fh_put
1105 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1106 loff_t offset, struct kvec *vec, int vlen,
1107 unsigned long *count)
1112 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1113 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1116 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1118 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1121 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1129 * Write data to a file.
1130 * The stable flag requests synchronous writes.
1131 * N.B. After this call fhp needs an fh_put
1134 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1135 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1141 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1142 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1145 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1148 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1153 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1161 #ifdef CONFIG_NFSD_V3
1163 * Commit all pending writes to stable storage.
1164 * Strictly speaking, we could sync just the indicated file region here,
1165 * but there's currently no way we can ask the VFS to do so.
1167 * Unfortunately we cannot lock the file to make sure we return full WCC
1168 * data to the client, as locking happens lower down in the filesystem.
1171 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1172 loff_t offset, unsigned long count)
1177 if ((u64)count > ~(u64)offset)
1178 return nfserr_inval;
1180 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1183 if (EX_ISSYNC(fhp->fh_export)) {
1184 if (file->f_op && file->f_op->fsync) {
1185 err = nfserrno(nfsd_sync(file));
1187 err = nfserr_notsupp;
1194 #endif /* CONFIG_NFSD_V3 */
1197 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1201 * Mode has already been set earlier in create:
1203 iap->ia_valid &= ~ATTR_MODE;
1205 * Setting uid/gid works only for root. Irix appears to
1206 * send along the gid on create when it tries to implement
1207 * setgid directories via NFS:
1209 if (current_fsuid() != 0)
1210 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1212 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1216 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1217 * setting size to 0 may fail for some specific file systems by the permission
1218 * checking which requires WRITE permission but the mode is 000.
1219 * we ignore the resizing(to 0) on the just new created file, since the size is
1220 * 0 after file created.
1222 * call this only after vfs_create() is called.
1225 nfsd_check_ignore_resizing(struct iattr *iap)
1227 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1228 iap->ia_valid &= ~ATTR_SIZE;
1232 * Create a file (regular, directory, device, fifo); UNIX sockets
1233 * not yet implemented.
1234 * If the response fh has been verified, the parent directory should
1235 * already be locked. Note that the parent directory is left locked.
1237 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1240 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1241 char *fname, int flen, struct iattr *iap,
1242 int type, dev_t rdev, struct svc_fh *resfhp)
1244 struct dentry *dentry, *dchild = NULL;
1254 if (isdotent(fname, flen))
1257 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1261 dentry = fhp->fh_dentry;
1262 dirp = dentry->d_inode;
1264 err = nfserr_notdir;
1265 if (!dirp->i_op->lookup)
1268 * Check whether the response file handle has been verified yet.
1269 * If it has, the parent directory should already be locked.
1271 if (!resfhp->fh_dentry) {
1272 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1273 fh_lock_nested(fhp, I_MUTEX_PARENT);
1274 dchild = lookup_one_len(fname, dentry, flen);
1275 host_err = PTR_ERR(dchild);
1278 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1282 /* called from nfsd_proc_create */
1283 dchild = dget(resfhp->fh_dentry);
1284 if (!fhp->fh_locked) {
1285 /* not actually possible */
1287 "nfsd_create: parent %s/%s not locked!\n",
1288 dentry->d_parent->d_name.name,
1289 dentry->d_name.name);
1295 * Make sure the child dentry is still negative ...
1298 if (dchild->d_inode) {
1299 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1300 dentry->d_name.name, dchild->d_name.name);
1304 if (!(iap->ia_valid & ATTR_MODE))
1306 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1309 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1310 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1315 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1320 * Get the dir op function pointer.
1325 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1327 nfsd_check_ignore_resizing(iap);
1330 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1336 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1340 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1344 if (EX_ISSYNC(fhp->fh_export)) {
1345 err = nfserrno(nfsd_sync_dir(dentry));
1346 write_inode_now(dchild->d_inode, 1);
1349 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1352 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1354 * Update the file handle to get the new inode info.
1357 err = fh_update(resfhp);
1359 if (dchild && !IS_ERR(dchild))
1364 err = nfserrno(host_err);
1368 #ifdef CONFIG_NFSD_V3
1370 * NFSv3 version of nfsd_create
1373 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1374 char *fname, int flen, struct iattr *iap,
1375 struct svc_fh *resfhp, int createmode, u32 *verifier,
1376 int *truncp, int *created)
1378 struct dentry *dentry, *dchild = NULL;
1383 __u32 v_mtime=0, v_atime=0;
1389 if (isdotent(fname, flen))
1391 if (!(iap->ia_valid & ATTR_MODE))
1393 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1397 dentry = fhp->fh_dentry;
1398 dirp = dentry->d_inode;
1400 /* Get all the sanity checks out of the way before
1401 * we lock the parent. */
1402 err = nfserr_notdir;
1403 if (!dirp->i_op->lookup)
1405 fh_lock_nested(fhp, I_MUTEX_PARENT);
1408 * Compose the response file handle.
1410 dchild = lookup_one_len(fname, dentry, flen);
1411 host_err = PTR_ERR(dchild);
1415 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1419 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1420 /* solaris7 gets confused (bugid 4218508) if these have
1421 * the high bit set, so just clear the high bits. If this is
1422 * ever changed to use different attrs for storing the
1423 * verifier, then do_open_lookup() will also need to be fixed
1426 v_mtime = verifier[0]&0x7fffffff;
1427 v_atime = verifier[1]&0x7fffffff;
1430 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1433 if (dchild->d_inode) {
1436 switch (createmode) {
1437 case NFS3_CREATE_UNCHECKED:
1438 if (! S_ISREG(dchild->d_inode->i_mode))
1441 /* in nfsv4, we need to treat this case a little
1442 * differently. we don't want to truncate the
1443 * file now; this would be wrong if the OPEN
1444 * fails for some other reason. furthermore,
1445 * if the size is nonzero, we should ignore it
1446 * according to spec!
1448 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1451 iap->ia_valid &= ATTR_SIZE;
1455 case NFS3_CREATE_EXCLUSIVE:
1456 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1457 && dchild->d_inode->i_atime.tv_sec == v_atime
1458 && dchild->d_inode->i_size == 0 )
1461 case NFS3_CREATE_GUARDED:
1464 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1468 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1470 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1476 if (EX_ISSYNC(fhp->fh_export)) {
1477 err = nfserrno(nfsd_sync_dir(dentry));
1478 /* setattr will sync the child (or not) */
1481 nfsd_check_ignore_resizing(iap);
1483 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1484 /* Cram the verifier into atime/mtime */
1485 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1486 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1487 /* XXX someone who knows this better please fix it for nsec */
1488 iap->ia_mtime.tv_sec = v_mtime;
1489 iap->ia_atime.tv_sec = v_atime;
1490 iap->ia_mtime.tv_nsec = 0;
1491 iap->ia_atime.tv_nsec = 0;
1495 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1499 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1501 * Update the filehandle to get the new inode info.
1504 err = fh_update(resfhp);
1508 if (dchild && !IS_ERR(dchild))
1513 err = nfserrno(host_err);
1516 #endif /* CONFIG_NFSD_V3 */
1519 * Read a symlink. On entry, *lenp must contain the maximum path length that
1520 * fits into the buffer. On return, it contains the true length.
1521 * N.B. After this call fhp needs an fh_put
1524 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1526 struct dentry *dentry;
1527 struct inode *inode;
1532 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1536 dentry = fhp->fh_dentry;
1537 inode = dentry->d_inode;
1540 if (!inode->i_op->readlink)
1543 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1544 /* N.B. Why does this call need a get_fs()??
1545 * Remove the set_fs and watch the fireworks:-) --okir
1548 oldfs = get_fs(); set_fs(KERNEL_DS);
1549 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1560 err = nfserrno(host_err);
1565 * Create a symlink and look up its inode
1566 * N.B. After this call _both_ fhp and resfhp need an fh_put
1569 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1570 char *fname, int flen,
1571 char *path, int plen,
1572 struct svc_fh *resfhp,
1575 struct dentry *dentry, *dnew;
1583 if (isdotent(fname, flen))
1586 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1590 dentry = fhp->fh_dentry;
1591 dnew = lookup_one_len(fname, dentry, flen);
1592 host_err = PTR_ERR(dnew);
1596 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1600 if (unlikely(path[plen] != 0)) {
1601 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1602 if (path_alloced == NULL)
1605 strncpy(path_alloced, path, plen);
1606 path_alloced[plen] = 0;
1607 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1608 kfree(path_alloced);
1611 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1614 if (EX_ISSYNC(fhp->fh_export))
1615 host_err = nfsd_sync_dir(dentry);
1617 err = nfserrno(host_err);
1620 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1622 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1624 if (err==0) err = cerr;
1629 err = nfserrno(host_err);
1635 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1638 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1639 char *name, int len, struct svc_fh *tfhp)
1641 struct dentry *ddir, *dnew, *dold;
1642 struct inode *dirp, *dest;
1646 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1649 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1657 if (isdotent(name, len))
1660 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1661 ddir = ffhp->fh_dentry;
1662 dirp = ddir->d_inode;
1664 dnew = lookup_one_len(name, ddir, len);
1665 host_err = PTR_ERR(dnew);
1669 dold = tfhp->fh_dentry;
1670 dest = dold->d_inode;
1672 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1674 err = nfserrno(host_err);
1677 host_err = vfs_link(dold, dirp, dnew);
1679 if (EX_ISSYNC(ffhp->fh_export)) {
1680 err = nfserrno(nfsd_sync_dir(ddir));
1681 write_inode_now(dest, 1);
1685 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1688 err = nfserrno(host_err);
1690 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1699 err = nfserrno(host_err);
1705 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1708 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1709 struct svc_fh *tfhp, char *tname, int tlen)
1711 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1712 struct inode *fdir, *tdir;
1716 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1719 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1723 fdentry = ffhp->fh_dentry;
1724 fdir = fdentry->d_inode;
1726 tdentry = tfhp->fh_dentry;
1727 tdir = tdentry->d_inode;
1729 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1730 if (ffhp->fh_export != tfhp->fh_export)
1734 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1737 /* cannot use fh_lock as we need deadlock protective ordering
1738 * so do it by hand */
1739 trap = lock_rename(tdentry, fdentry);
1740 ffhp->fh_locked = tfhp->fh_locked = 1;
1744 odentry = lookup_one_len(fname, fdentry, flen);
1745 host_err = PTR_ERR(odentry);
1746 if (IS_ERR(odentry))
1750 if (!odentry->d_inode)
1753 if (odentry == trap)
1756 ndentry = lookup_one_len(tname, tdentry, tlen);
1757 host_err = PTR_ERR(ndentry);
1758 if (IS_ERR(ndentry))
1760 host_err = -ENOTEMPTY;
1761 if (ndentry == trap)
1764 if (svc_msnfs(ffhp) &&
1765 ((atomic_read(&odentry->d_count) > 1)
1766 || (atomic_read(&ndentry->d_count) > 1))) {
1772 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1774 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1778 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1779 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1780 host_err = nfsd_sync_dir(tdentry);
1782 host_err = nfsd_sync_dir(fdentry);
1785 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1792 err = nfserrno(host_err);
1794 /* we cannot reply on fh_unlock on the two filehandles,
1795 * as that would do the wrong thing if the two directories
1796 * were the same, so again we do it by hand
1798 fill_post_wcc(ffhp);
1799 fill_post_wcc(tfhp);
1800 unlock_rename(tdentry, fdentry);
1801 ffhp->fh_locked = tfhp->fh_locked = 0;
1808 * Unlink a file or directory
1809 * N.B. After this call fhp needs an fh_put
1812 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1813 char *fname, int flen)
1815 struct dentry *dentry, *rdentry;
1821 if (!flen || isdotent(fname, flen))
1823 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1827 fh_lock_nested(fhp, I_MUTEX_PARENT);
1828 dentry = fhp->fh_dentry;
1829 dirp = dentry->d_inode;
1831 rdentry = lookup_one_len(fname, dentry, flen);
1832 host_err = PTR_ERR(rdentry);
1833 if (IS_ERR(rdentry))
1836 if (!rdentry->d_inode) {
1843 type = rdentry->d_inode->i_mode & S_IFMT;
1845 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1849 if (type != S_IFDIR) { /* It's UNLINK */
1851 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1852 (atomic_read(&rdentry->d_count) > 1)) {
1856 host_err = vfs_unlink(dirp, rdentry);
1857 } else { /* It's RMDIR */
1858 host_err = vfs_rmdir(dirp, rdentry);
1865 if (EX_ISSYNC(fhp->fh_export))
1866 host_err = nfsd_sync_dir(dentry);
1869 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1871 err = nfserrno(host_err);
1877 * We do this buffering because we must not call back into the file
1878 * system's ->lookup() method from the filldir callback. That may well
1879 * deadlock a number of file systems.
1881 * This is based heavily on the implementation of same in XFS.
1883 struct buffered_dirent {
1887 unsigned int d_type;
1891 struct readdir_data {
1897 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1898 loff_t offset, u64 ino, unsigned int d_type)
1900 struct readdir_data *buf = __buf;
1901 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1902 unsigned int reclen;
1904 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1905 if (buf->used + reclen > PAGE_SIZE) {
1910 de->namlen = namlen;
1911 de->offset = offset;
1913 de->d_type = d_type;
1914 memcpy(de->name, name, namlen);
1915 buf->used += reclen;
1920 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1921 struct readdir_cd *cdp, loff_t *offsetp)
1923 struct readdir_data buf;
1924 struct buffered_dirent *de;
1929 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1931 return nfserrno(-ENOMEM);
1936 struct inode *dir_inode = file->f_path.dentry->d_inode;
1937 unsigned int reclen;
1939 cdp->err = nfserr_eof; /* will be cleared on successful read */
1943 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1956 * Various filldir functions may end up calling back into
1957 * lookup_one_len() and the file system's ->lookup() method.
1958 * These expect i_mutex to be held, as it would within readdir.
1960 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1964 de = (struct buffered_dirent *)buf.dirent;
1966 offset = de->offset;
1968 if (func(cdp, de->name, de->namlen, de->offset,
1969 de->ino, de->d_type))
1972 if (cdp->err != nfs_ok)
1975 reclen = ALIGN(sizeof(*de) + de->namlen,
1978 de = (struct buffered_dirent *)((char *)de + reclen);
1980 mutex_unlock(&dir_inode->i_mutex);
1981 if (size > 0) /* We bailed out early */
1984 offset = vfs_llseek(file, 0, SEEK_CUR);
1987 free_page((unsigned long)(buf.dirent));
1990 return nfserrno(host_err);
1997 * Read entries from a directory.
1998 * The NFSv3/4 verifier we ignore for now.
2001 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2002 struct readdir_cd *cdp, filldir_t func)
2006 loff_t offset = *offsetp;
2008 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2012 offset = vfs_llseek(file, offset, 0);
2014 err = nfserrno((int)offset);
2018 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2020 if (err == nfserr_eof || err == nfserr_toosmall)
2021 err = nfs_ok; /* can still be found in ->err */
2029 * Get file system stats
2030 * N.B. After this call fhp needs an fh_put
2033 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2035 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2036 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2041 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2043 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2047 * Check for a user's access permissions to this inode.
2050 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2051 struct dentry *dentry, int acc)
2053 struct inode *inode = dentry->d_inode;
2057 if (acc == NFSD_MAY_NOP)
2060 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2062 (acc & NFSD_MAY_READ)? " read" : "",
2063 (acc & NFSD_MAY_WRITE)? " write" : "",
2064 (acc & NFSD_MAY_EXEC)? " exec" : "",
2065 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2066 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2067 (acc & NFSD_MAY_LOCK)? " lock" : "",
2068 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2070 IS_IMMUTABLE(inode)? " immut" : "",
2071 IS_APPEND(inode)? " append" : "",
2072 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2073 dprintk(" owner %d/%d user %d/%d\n",
2074 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2077 /* Normally we reject any write/sattr etc access on a read-only file
2078 * system. But if it is IRIX doing check on write-access for a
2079 * device special file, we ignore rofs.
2081 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2082 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2083 if (exp_rdonly(rqstp, exp) ||
2084 __mnt_is_readonly(exp->ex_path.mnt))
2086 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2089 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2092 if (acc & NFSD_MAY_LOCK) {
2093 /* If we cannot rely on authentication in NLM requests,
2094 * just allow locks, otherwise require read permission, or
2097 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2100 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2103 * The file owner always gets access permission for accesses that
2104 * would normally be checked at open time. This is to make
2105 * file access work even when the client has done a fchmod(fd, 0).
2107 * However, `cp foo bar' should fail nevertheless when bar is
2108 * readonly. A sensible way to do this might be to reject all
2109 * attempts to truncate a read-only file, because a creat() call
2110 * always implies file truncation.
2111 * ... but this isn't really fair. A process may reasonably call
2112 * ftruncate on an open file descriptor on a file with perm 000.
2113 * We must trust the client to do permission checking - using "ACCESS"
2116 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2117 inode->i_uid == current_fsuid())
2120 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2121 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2123 /* Allow read access to binaries even when mode 111 */
2124 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2125 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2126 err = inode_permission(inode, MAY_EXEC);
2130 /* Do integrity (permission) checking now, but defer incrementing
2131 * IMA counts to the actual file open.
2133 path.mnt = exp->ex_path.mnt;
2134 path.dentry = dentry;
2135 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2138 return err? nfserrno(err) : 0;
2142 nfsd_racache_shutdown(void)
2144 struct raparms *raparm, *last_raparm;
2147 dprintk("nfsd: freeing readahead buffers.\n");
2149 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2150 raparm = raparm_hash[i].pb_head;
2152 last_raparm = raparm;
2153 raparm = raparm->p_next;
2156 raparm_hash[i].pb_head = NULL;
2160 * Initialize readahead param cache
2163 nfsd_racache_init(int cache_size)
2168 struct raparms **raparm = NULL;
2171 if (raparm_hash[0].pb_head)
2173 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2176 cache_size = nperbucket * RAPARM_HASH_SIZE;
2178 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2180 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2181 spin_lock_init(&raparm_hash[i].pb_lock);
2183 raparm = &raparm_hash[i].pb_head;
2184 for (j = 0; j < nperbucket; j++) {
2185 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2188 raparm = &(*raparm)->p_next;
2193 nfsdstats.ra_size = cache_size;
2197 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2198 nfsd_racache_shutdown();
2202 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2204 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2206 struct inode *inode = fhp->fh_dentry->d_inode;
2210 struct posix_acl *acl;
2212 if (!IS_POSIXACL(inode))
2213 return ERR_PTR(-EOPNOTSUPP);
2216 case ACL_TYPE_ACCESS:
2217 name = POSIX_ACL_XATTR_ACCESS;
2219 case ACL_TYPE_DEFAULT:
2220 name = POSIX_ACL_XATTR_DEFAULT;
2223 return ERR_PTR(-EOPNOTSUPP);
2226 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2228 return ERR_PTR(size);
2230 acl = posix_acl_from_xattr(value, size);
2236 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2238 struct inode *inode = fhp->fh_dentry->d_inode;
2244 if (!IS_POSIXACL(inode) ||
2245 !inode->i_op->setxattr || !inode->i_op->removexattr)
2248 case ACL_TYPE_ACCESS:
2249 name = POSIX_ACL_XATTR_ACCESS;
2251 case ACL_TYPE_DEFAULT:
2252 name = POSIX_ACL_XATTR_DEFAULT;
2258 if (acl && acl->a_count) {
2259 size = posix_acl_xattr_size(acl->a_count);
2260 value = kmalloc(size, GFP_KERNEL);
2263 error = posix_acl_to_xattr(acl, value, size);
2270 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2274 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2276 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2279 error = vfs_removexattr(fhp->fh_dentry, name);
2280 if (error == -ENODATA)
2284 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2290 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */