1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
58 #include <linux/ima.h>
60 #include <asm/uaccess.h>
62 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
66 * This is a cache of readahead params that help us choose the proper
67 * readahead strategy. Initially, we set all readahead parameters to 0
68 * and let the VFS handle things.
69 * If you increase the number of cached files very much, you'll need to
70 * add a hash table here.
73 struct raparms *p_next;
78 struct file_ra_state p_ra;
79 unsigned int p_hindex;
82 struct raparm_hbucket {
83 struct raparms *pb_head;
85 } ____cacheline_aligned_in_smp;
87 #define RAPARM_HASH_BITS 4
88 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
93 nfsd_v4client(struct svc_rqst *rq)
95 return rq->rq_prog == NFS_PROGRAM && rq->rq_vers == 4;
99 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
101 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
102 * or nfs_ok having possibly changed *dpp and *expp
105 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
106 struct svc_export **expp)
108 struct svc_export *exp = *expp, *exp2 = NULL;
109 struct dentry *dentry = *dpp;
110 struct path path = {.mnt = mntget(exp->ex_path.mnt),
111 .dentry = dget(dentry)};
114 while (d_mountpoint(path.dentry) && follow_down(&path))
117 exp2 = rqst_exp_get_by_name(rqstp, &path);
119 if (PTR_ERR(exp2) != -ENOENT)
124 if (nfsd_v4client(rqstp) ||
125 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
126 /* successfully crossed mount point */
128 * This is subtle: path.dentry is *not* on path.mnt
129 * at this point. The only reason we are safe is that
130 * original mnt is pinned down by exp, so we should
131 * put path *before* putting exp
134 path.dentry = dentry;
144 static void follow_to_parent(struct path *path)
148 while (path->dentry == path->mnt->mnt_root && follow_up(path))
150 dp = dget_parent(path->dentry);
155 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
157 struct svc_export *exp2;
158 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
159 .dentry = dget(dparent)};
161 follow_to_parent(&path);
163 exp2 = rqst_exp_parent(rqstp, &path);
164 if (PTR_ERR(exp2) == -ENOENT) {
165 *dentryp = dget(dparent);
166 } else if (IS_ERR(exp2)) {
168 return PTR_ERR(exp2);
170 *dentryp = dget(path.dentry);
179 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
180 const char *name, unsigned int len,
181 struct svc_export **exp_ret, struct dentry **dentry_ret)
183 struct svc_export *exp;
184 struct dentry *dparent;
185 struct dentry *dentry;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
191 /* Obtain dentry and export. */
192 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
196 dparent = fhp->fh_dentry;
197 exp = fhp->fh_export;
200 /* Lookup the name, but don't follow links */
201 if (isdotent(name, len)) {
203 dentry = dget(dparent);
204 else if (dparent != exp->ex_path.dentry)
205 dentry = dget_parent(dparent);
206 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
207 dentry = dget(dparent); /* .. == . just like at / */
209 /* checking mountpoint crossing is very different when stepping up */
210 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
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 (d_mountpoint(dentry)) {
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 = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
262 err = check_nfsd_access(exp, rqstp);
266 * Note: we compose the file handle now, but as the
267 * dentry may be negative, it may need to be updated.
269 err = fh_compose(resfh, exp, dentry, fhp);
270 if (!err && !dentry->d_inode)
280 * Set various file attributes.
281 * N.B. After this call fhp needs an fh_put
284 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
285 int check_guard, time_t guardtime)
287 struct dentry *dentry;
289 int accmode = NFSD_MAY_SATTR;
295 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
296 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
297 if (iap->ia_valid & ATTR_SIZE)
301 err = fh_verify(rqstp, fhp, ftype, accmode);
305 dentry = fhp->fh_dentry;
306 inode = dentry->d_inode;
308 /* Ignore any mode updates on symlinks */
309 if (S_ISLNK(inode->i_mode))
310 iap->ia_valid &= ~ATTR_MODE;
316 * NFSv2 does not differentiate between "set-[ac]time-to-now"
317 * which only requires access, and "set-[ac]time-to-X" which
318 * requires ownership.
319 * So if it looks like it might be "set both to the same time which
320 * is close to now", and if inode_change_ok fails, then we
321 * convert to "set to now" instead of "set to explicit time"
323 * We only call inode_change_ok as the last test as technically
324 * it is not an interface that we should be using. It is only
325 * valid if the filesystem does not define it's own i_op->setattr.
327 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
328 #define MAX_TOUCH_TIME_ERROR (30*60)
329 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
330 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
334 * Now just make sure time is in the right ballpark.
335 * Solaris, at least, doesn't seem to care what the time
336 * request is. We require it be within 30 minutes of now.
338 time_t delta = iap->ia_atime.tv_sec - get_seconds();
341 if (delta < MAX_TOUCH_TIME_ERROR &&
342 inode_change_ok(inode, iap) != 0) {
344 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
345 * This will cause notify_change to set these times
348 iap->ia_valid &= ~BOTH_TIME_SET;
353 * The size case is special.
354 * It changes the file as well as the attributes.
356 if (iap->ia_valid & ATTR_SIZE) {
357 if (iap->ia_size < inode->i_size) {
358 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
359 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
365 * If we are changing the size of the file, then
366 * we need to break all leases.
368 host_err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
369 if (host_err == -EWOULDBLOCK)
370 host_err = -ETIMEDOUT;
371 if (host_err) /* ENOMEM or EWOULDBLOCK */
374 host_err = get_write_access(inode);
379 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
381 put_write_access(inode);
387 /* sanitize the mode change */
388 if (iap->ia_valid & ATTR_MODE) {
389 iap->ia_mode &= S_IALLUGO;
390 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
393 /* Revoke setuid/setgid on chown */
394 if (!S_ISDIR(inode->i_mode) &&
395 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
396 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
397 iap->ia_valid |= ATTR_KILL_PRIV;
398 if (iap->ia_valid & ATTR_MODE) {
399 /* we're setting mode too, just clear the s*id bits */
400 iap->ia_mode &= ~S_ISUID;
401 if (iap->ia_mode & S_IXGRP)
402 iap->ia_mode &= ~S_ISGID;
404 /* set ATTR_KILL_* bits and let VFS handle it */
405 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
409 /* Change the attributes. */
411 iap->ia_valid |= ATTR_CTIME;
413 err = nfserr_notsync;
414 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
416 host_err = notify_change(dentry, iap);
417 err = nfserrno(host_err);
421 put_write_access(inode);
423 if (EX_ISSYNC(fhp->fh_export))
424 write_inode_now(inode, 1);
429 err = nfserrno(host_err);
433 #if defined(CONFIG_NFSD_V2_ACL) || \
434 defined(CONFIG_NFSD_V3_ACL) || \
435 defined(CONFIG_NFSD_V4)
436 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
441 buflen = vfs_getxattr(dentry, key, NULL, 0);
445 *buf = kmalloc(buflen, GFP_KERNEL);
449 ret = vfs_getxattr(dentry, key, *buf, buflen);
456 #if defined(CONFIG_NFSD_V4)
458 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
465 buflen = posix_acl_xattr_size(pacl->a_count);
466 buf = kmalloc(buflen, GFP_KERNEL);
471 len = posix_acl_to_xattr(pacl, buf, buflen);
477 error = vfs_setxattr(dentry, key, buf, len, 0);
484 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
485 struct nfs4_acl *acl)
489 struct dentry *dentry;
491 struct posix_acl *pacl = NULL, *dpacl = NULL;
492 unsigned int flags = 0;
495 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
499 dentry = fhp->fh_dentry;
500 inode = dentry->d_inode;
501 if (S_ISDIR(inode->i_mode))
502 flags = NFS4_ACL_DIR;
504 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
505 if (host_error == -EINVAL) {
506 return nfserr_attrnotsupp;
507 } else if (host_error < 0)
510 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
514 if (S_ISDIR(inode->i_mode))
515 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
518 posix_acl_release(pacl);
519 posix_acl_release(dpacl);
521 if (host_error == -EOPNOTSUPP)
522 return nfserr_attrnotsupp;
524 return nfserrno(host_error);
527 static struct posix_acl *
528 _get_posix_acl(struct dentry *dentry, char *key)
531 struct posix_acl *pacl = NULL;
534 buflen = nfsd_getxattr(dentry, key, &buf);
538 return ERR_PTR(buflen);
540 pacl = posix_acl_from_xattr(buf, buflen);
546 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
548 struct inode *inode = dentry->d_inode;
550 struct posix_acl *pacl = NULL, *dpacl = NULL;
551 unsigned int flags = 0;
553 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
554 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
555 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
557 error = PTR_ERR(pacl);
562 if (S_ISDIR(inode->i_mode)) {
563 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
564 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
566 else if (IS_ERR(dpacl)) {
567 error = PTR_ERR(dpacl);
571 flags = NFS4_ACL_DIR;
574 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
576 error = PTR_ERR(*acl);
580 posix_acl_release(pacl);
581 posix_acl_release(dpacl);
585 #endif /* defined(CONFIG_NFS_V4) */
587 #ifdef CONFIG_NFSD_V3
589 * Check server access rights to a file system object
595 static struct accessmap nfs3_regaccess[] = {
596 { NFS3_ACCESS_READ, NFSD_MAY_READ },
597 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
598 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
599 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
604 static struct accessmap nfs3_diraccess[] = {
605 { NFS3_ACCESS_READ, NFSD_MAY_READ },
606 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
607 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
608 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
609 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
614 static struct accessmap nfs3_anyaccess[] = {
615 /* Some clients - Solaris 2.6 at least, make an access call
616 * to the server to check for access for things like /dev/null
617 * (which really, the server doesn't care about). So
618 * We provide simple access checking for them, looking
619 * mainly at mode bits, and we make sure to ignore read-only
622 { NFS3_ACCESS_READ, NFSD_MAY_READ },
623 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
624 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
625 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
631 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
633 struct accessmap *map;
634 struct svc_export *export;
635 struct dentry *dentry;
636 u32 query, result = 0, sresult = 0;
639 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
643 export = fhp->fh_export;
644 dentry = fhp->fh_dentry;
646 if (S_ISREG(dentry->d_inode->i_mode))
647 map = nfs3_regaccess;
648 else if (S_ISDIR(dentry->d_inode->i_mode))
649 map = nfs3_diraccess;
651 map = nfs3_anyaccess;
655 for (; map->access; map++) {
656 if (map->access & query) {
659 sresult |= map->access;
661 err2 = nfsd_permission(rqstp, export, dentry, map->how);
664 result |= map->access;
667 /* the following error codes just mean the access was not allowed,
668 * rather than an error occurred */
672 /* simply don't "or" in the access bit. */
682 *supported = sresult;
687 #endif /* CONFIG_NFSD_V3 */
692 * Open an existing file or directory.
693 * The access argument indicates the type of open (read/write/lock)
694 * N.B. After this call fhp needs an fh_put
697 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
698 int access, struct file **filp)
700 struct dentry *dentry;
702 int flags = O_RDONLY|O_LARGEFILE;
706 validate_process_creds();
709 * If we get here, then the client has already done an "open",
710 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
711 * in case a chmod has now revoked permission.
713 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
717 dentry = fhp->fh_dentry;
718 inode = dentry->d_inode;
720 /* Disallow write access to files with the append-only bit set
721 * or any access when mandatory locking enabled
724 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
727 * We must ignore files (but only files) which might have mandatory
728 * locks on them because there is no way to know if the accesser has
731 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
738 * Check to see if there are any leases on this file.
739 * This may block while leases are broken.
741 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? FMODE_WRITE : 0));
742 if (host_err == -EWOULDBLOCK)
743 host_err = -ETIMEDOUT;
744 if (host_err) /* NOMEM or WOULDBLOCK */
747 if (access & NFSD_MAY_WRITE) {
748 if (access & NFSD_MAY_READ)
749 flags = O_RDWR|O_LARGEFILE;
751 flags = O_WRONLY|O_LARGEFILE;
755 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
756 flags, current_cred());
758 host_err = PTR_ERR(*filp);
760 ima_counts_get(*filp);
762 err = nfserrno(host_err);
764 validate_process_creds();
772 nfsd_close(struct file *filp)
779 * As this calls fsync (not fdatasync) there is no need for a write_inode
782 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
783 const struct file_operations *fop)
785 struct inode *inode = dp->d_inode;
786 int (*fsync) (struct file *, struct dentry *, int);
789 err = filemap_fdatawrite(inode->i_mapping);
790 if (err == 0 && fop && (fsync = fop->fsync))
791 err = fsync(filp, dp, 0);
793 err = filemap_fdatawait(inode->i_mapping);
799 nfsd_sync(struct file *filp)
802 struct inode *inode = filp->f_path.dentry->d_inode;
803 dprintk("nfsd: sync file %s\n", filp->f_path.dentry->d_name.name);
804 mutex_lock(&inode->i_mutex);
805 err=nfsd_dosync(filp, filp->f_path.dentry, filp->f_op);
806 mutex_unlock(&inode->i_mutex);
812 nfsd_sync_dir(struct dentry *dp)
814 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
818 * Obtain the readahead parameters for the file
819 * specified by (dev, ino).
822 static inline struct raparms *
823 nfsd_get_raparms(dev_t dev, ino_t ino)
825 struct raparms *ra, **rap, **frap = NULL;
828 struct raparm_hbucket *rab;
830 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
831 rab = &raparm_hash[hash];
833 spin_lock(&rab->pb_lock);
834 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
835 if (ra->p_ino == ino && ra->p_dev == dev)
838 if (ra->p_count == 0)
841 depth = nfsdstats.ra_size*11/10;
843 spin_unlock(&rab->pb_lock);
853 if (rap != &rab->pb_head) {
855 ra->p_next = rab->pb_head;
859 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
860 spin_unlock(&rab->pb_lock);
865 * Grab and keep cached pages associated with a file in the svc_rqst
866 * so that they can be passed to the network sendmsg/sendpage routines
867 * directly. They will be released after the sending has completed.
870 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
871 struct splice_desc *sd)
873 struct svc_rqst *rqstp = sd->u.data;
874 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
875 struct page *page = buf->page;
879 ret = buf->ops->confirm(pipe, buf);
885 if (rqstp->rq_res.page_len == 0) {
890 rqstp->rq_res.page_base = buf->offset;
891 rqstp->rq_res.page_len = size;
892 } else if (page != pp[-1]) {
898 rqstp->rq_res.page_len += size;
900 rqstp->rq_res.page_len += size;
905 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
906 struct splice_desc *sd)
908 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
911 static inline int svc_msnfs(struct svc_fh *ffhp)
914 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
921 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
922 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
931 inode = file->f_path.dentry->d_inode;
933 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
936 /* Get readahead parameters */
937 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
940 file->f_ra = ra->p_ra;
942 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
943 struct splice_desc sd = {
950 rqstp->rq_resused = 1;
951 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
955 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
959 /* Write back readahead params */
961 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
962 spin_lock(&rab->pb_lock);
963 ra->p_ra = file->f_ra;
966 spin_unlock(&rab->pb_lock);
970 nfsdstats.io_read += host_err;
973 fsnotify_access(file->f_path.dentry);
975 err = nfserrno(host_err);
980 static void kill_suid(struct dentry *dentry)
983 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
985 mutex_lock(&dentry->d_inode->i_mutex);
986 notify_change(dentry, &ia);
987 mutex_unlock(&dentry->d_inode->i_mutex);
991 * Gathered writes: If another process is currently writing to the file,
992 * there's a high chance this is another nfsd (triggered by a bulk write
993 * from a client's biod). Rather than syncing the file with each write
994 * request, we sleep for 10 msec.
996 * I don't know if this roughly approximates C. Juszak's idea of
997 * gathered writes, but it's a nice and simple solution (IMHO), and it
1000 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1001 * better tool (separate unstable writes and commits) for solving this
1004 static int wait_for_concurrent_writes(struct file *file)
1006 struct inode *inode = file->f_path.dentry->d_inode;
1007 static ino_t last_ino;
1008 static dev_t last_dev;
1011 if (atomic_read(&inode->i_writecount) > 1
1012 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
1013 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
1015 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
1018 if (inode->i_state & I_DIRTY) {
1019 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1020 err = nfsd_sync(file);
1022 last_ino = inode->i_ino;
1023 last_dev = inode->i_sb->s_dev;
1028 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1029 loff_t offset, struct kvec *vec, int vlen,
1030 unsigned long *cnt, int *stablep)
1032 struct svc_export *exp;
1033 struct dentry *dentry;
1034 struct inode *inode;
1038 int stable = *stablep;
1044 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1045 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1049 dentry = file->f_path.dentry;
1050 inode = dentry->d_inode;
1051 exp = fhp->fh_export;
1054 * Request sync writes if
1055 * - the sync export option has been set, or
1056 * - the client requested O_SYNC behavior (NFSv3 feature).
1057 * - The file system doesn't support fsync().
1058 * When NFSv2 gathered writes have been configured for this volume,
1059 * flushing the data to disk is handled separately below.
1061 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1063 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1065 *stablep = 2; /* FILE_SYNC */
1068 if (!EX_ISSYNC(exp))
1070 if (stable && !use_wgather) {
1071 spin_lock(&file->f_lock);
1072 file->f_flags |= O_SYNC;
1073 spin_unlock(&file->f_lock);
1076 /* Write the data. */
1077 oldfs = get_fs(); set_fs(KERNEL_DS);
1078 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1083 nfsdstats.io_write += host_err;
1084 fsnotify_modify(file->f_path.dentry);
1086 /* clear setuid/setgid flag after write */
1087 if (inode->i_mode & (S_ISUID | S_ISGID))
1090 if (stable && use_wgather)
1091 host_err = wait_for_concurrent_writes(file);
1094 dprintk("nfsd: write complete host_err=%d\n", host_err);
1098 err = nfserrno(host_err);
1104 * Read data from a file. count must contain the requested read count
1105 * on entry. On return, *count contains the number of bytes actually read.
1106 * N.B. After this call fhp needs an fh_put
1109 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1110 loff_t offset, struct kvec *vec, int vlen,
1111 unsigned long *count)
1116 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1117 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1120 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1122 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1125 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1133 * Write data to a file.
1134 * The stable flag requests synchronous writes.
1135 * N.B. After this call fhp needs an fh_put
1138 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1139 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1145 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1146 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1149 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1152 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1157 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1165 #ifdef CONFIG_NFSD_V3
1167 * Commit all pending writes to stable storage.
1168 * Strictly speaking, we could sync just the indicated file region here,
1169 * but there's currently no way we can ask the VFS to do so.
1171 * Unfortunately we cannot lock the file to make sure we return full WCC
1172 * data to the client, as locking happens lower down in the filesystem.
1175 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1176 loff_t offset, unsigned long count)
1181 if ((u64)count > ~(u64)offset)
1182 return nfserr_inval;
1184 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1187 if (EX_ISSYNC(fhp->fh_export)) {
1188 if (file->f_op && file->f_op->fsync) {
1189 err = nfserrno(nfsd_sync(file));
1191 err = nfserr_notsupp;
1198 #endif /* CONFIG_NFSD_V3 */
1201 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1205 * Mode has already been set earlier in create:
1207 iap->ia_valid &= ~ATTR_MODE;
1209 * Setting uid/gid works only for root. Irix appears to
1210 * send along the gid on create when it tries to implement
1211 * setgid directories via NFS:
1213 if (current_fsuid() != 0)
1214 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1216 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1220 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1221 * setting size to 0 may fail for some specific file systems by the permission
1222 * checking which requires WRITE permission but the mode is 000.
1223 * we ignore the resizing(to 0) on the just new created file, since the size is
1224 * 0 after file created.
1226 * call this only after vfs_create() is called.
1229 nfsd_check_ignore_resizing(struct iattr *iap)
1231 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1232 iap->ia_valid &= ~ATTR_SIZE;
1236 * Create a file (regular, directory, device, fifo); UNIX sockets
1237 * not yet implemented.
1238 * If the response fh has been verified, the parent directory should
1239 * already be locked. Note that the parent directory is left locked.
1241 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1244 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1245 char *fname, int flen, struct iattr *iap,
1246 int type, dev_t rdev, struct svc_fh *resfhp)
1248 struct dentry *dentry, *dchild = NULL;
1258 if (isdotent(fname, flen))
1261 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1265 dentry = fhp->fh_dentry;
1266 dirp = dentry->d_inode;
1268 err = nfserr_notdir;
1269 if (!dirp->i_op->lookup)
1272 * Check whether the response file handle has been verified yet.
1273 * If it has, the parent directory should already be locked.
1275 if (!resfhp->fh_dentry) {
1276 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1277 fh_lock_nested(fhp, I_MUTEX_PARENT);
1278 dchild = lookup_one_len(fname, dentry, flen);
1279 host_err = PTR_ERR(dchild);
1282 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1286 /* called from nfsd_proc_create */
1287 dchild = dget(resfhp->fh_dentry);
1288 if (!fhp->fh_locked) {
1289 /* not actually possible */
1291 "nfsd_create: parent %s/%s not locked!\n",
1292 dentry->d_parent->d_name.name,
1293 dentry->d_name.name);
1299 * Make sure the child dentry is still negative ...
1302 if (dchild->d_inode) {
1303 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1304 dentry->d_name.name, dchild->d_name.name);
1308 if (!(iap->ia_valid & ATTR_MODE))
1310 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1313 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1314 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1319 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1324 * Get the dir op function pointer.
1329 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1331 nfsd_check_ignore_resizing(iap);
1334 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1340 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1344 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1348 if (EX_ISSYNC(fhp->fh_export)) {
1349 err = nfserrno(nfsd_sync_dir(dentry));
1350 write_inode_now(dchild->d_inode, 1);
1353 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1356 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1358 * Update the file handle to get the new inode info.
1361 err = fh_update(resfhp);
1363 if (dchild && !IS_ERR(dchild))
1368 err = nfserrno(host_err);
1372 #ifdef CONFIG_NFSD_V3
1374 * NFSv3 version of nfsd_create
1377 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1378 char *fname, int flen, struct iattr *iap,
1379 struct svc_fh *resfhp, int createmode, u32 *verifier,
1380 int *truncp, int *created)
1382 struct dentry *dentry, *dchild = NULL;
1387 __u32 v_mtime=0, v_atime=0;
1393 if (isdotent(fname, flen))
1395 if (!(iap->ia_valid & ATTR_MODE))
1397 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1401 dentry = fhp->fh_dentry;
1402 dirp = dentry->d_inode;
1404 /* Get all the sanity checks out of the way before
1405 * we lock the parent. */
1406 err = nfserr_notdir;
1407 if (!dirp->i_op->lookup)
1409 fh_lock_nested(fhp, I_MUTEX_PARENT);
1412 * Compose the response file handle.
1414 dchild = lookup_one_len(fname, dentry, flen);
1415 host_err = PTR_ERR(dchild);
1419 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1423 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1424 /* solaris7 gets confused (bugid 4218508) if these have
1425 * the high bit set, so just clear the high bits. If this is
1426 * ever changed to use different attrs for storing the
1427 * verifier, then do_open_lookup() will also need to be fixed
1430 v_mtime = verifier[0]&0x7fffffff;
1431 v_atime = verifier[1]&0x7fffffff;
1434 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1437 if (dchild->d_inode) {
1440 switch (createmode) {
1441 case NFS3_CREATE_UNCHECKED:
1442 if (! S_ISREG(dchild->d_inode->i_mode))
1445 /* in nfsv4, we need to treat this case a little
1446 * differently. we don't want to truncate the
1447 * file now; this would be wrong if the OPEN
1448 * fails for some other reason. furthermore,
1449 * if the size is nonzero, we should ignore it
1450 * according to spec!
1452 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1455 iap->ia_valid &= ATTR_SIZE;
1459 case NFS3_CREATE_EXCLUSIVE:
1460 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1461 && dchild->d_inode->i_atime.tv_sec == v_atime
1462 && dchild->d_inode->i_size == 0 )
1465 case NFS3_CREATE_GUARDED:
1468 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1472 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1474 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1480 if (EX_ISSYNC(fhp->fh_export)) {
1481 err = nfserrno(nfsd_sync_dir(dentry));
1482 /* setattr will sync the child (or not) */
1485 nfsd_check_ignore_resizing(iap);
1487 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1488 /* Cram the verifier into atime/mtime */
1489 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1490 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1491 /* XXX someone who knows this better please fix it for nsec */
1492 iap->ia_mtime.tv_sec = v_mtime;
1493 iap->ia_atime.tv_sec = v_atime;
1494 iap->ia_mtime.tv_nsec = 0;
1495 iap->ia_atime.tv_nsec = 0;
1499 err2 = nfsd_create_setattr(rqstp, resfhp, iap);
1503 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1505 * Update the filehandle to get the new inode info.
1508 err = fh_update(resfhp);
1512 if (dchild && !IS_ERR(dchild))
1517 err = nfserrno(host_err);
1520 #endif /* CONFIG_NFSD_V3 */
1523 * Read a symlink. On entry, *lenp must contain the maximum path length that
1524 * fits into the buffer. On return, it contains the true length.
1525 * N.B. After this call fhp needs an fh_put
1528 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1530 struct dentry *dentry;
1531 struct inode *inode;
1536 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1540 dentry = fhp->fh_dentry;
1541 inode = dentry->d_inode;
1544 if (!inode->i_op->readlink)
1547 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1548 /* N.B. Why does this call need a get_fs()??
1549 * Remove the set_fs and watch the fireworks:-) --okir
1552 oldfs = get_fs(); set_fs(KERNEL_DS);
1553 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1564 err = nfserrno(host_err);
1569 * Create a symlink and look up its inode
1570 * N.B. After this call _both_ fhp and resfhp need an fh_put
1573 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1574 char *fname, int flen,
1575 char *path, int plen,
1576 struct svc_fh *resfhp,
1579 struct dentry *dentry, *dnew;
1587 if (isdotent(fname, flen))
1590 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1594 dentry = fhp->fh_dentry;
1595 dnew = lookup_one_len(fname, dentry, flen);
1596 host_err = PTR_ERR(dnew);
1600 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1604 if (unlikely(path[plen] != 0)) {
1605 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1606 if (path_alloced == NULL)
1609 strncpy(path_alloced, path, plen);
1610 path_alloced[plen] = 0;
1611 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1612 kfree(path_alloced);
1615 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1618 if (EX_ISSYNC(fhp->fh_export))
1619 host_err = nfsd_sync_dir(dentry);
1621 err = nfserrno(host_err);
1624 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1626 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1628 if (err==0) err = cerr;
1633 err = nfserrno(host_err);
1639 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1642 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1643 char *name, int len, struct svc_fh *tfhp)
1645 struct dentry *ddir, *dnew, *dold;
1646 struct inode *dirp, *dest;
1650 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1653 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1661 if (isdotent(name, len))
1664 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1665 ddir = ffhp->fh_dentry;
1666 dirp = ddir->d_inode;
1668 dnew = lookup_one_len(name, ddir, len);
1669 host_err = PTR_ERR(dnew);
1673 dold = tfhp->fh_dentry;
1674 dest = dold->d_inode;
1676 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1678 err = nfserrno(host_err);
1681 host_err = vfs_link(dold, dirp, dnew);
1683 if (EX_ISSYNC(ffhp->fh_export)) {
1684 err = nfserrno(nfsd_sync_dir(ddir));
1685 write_inode_now(dest, 1);
1689 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1692 err = nfserrno(host_err);
1694 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1703 err = nfserrno(host_err);
1709 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1712 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1713 struct svc_fh *tfhp, char *tname, int tlen)
1715 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1716 struct inode *fdir, *tdir;
1720 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1723 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1727 fdentry = ffhp->fh_dentry;
1728 fdir = fdentry->d_inode;
1730 tdentry = tfhp->fh_dentry;
1731 tdir = tdentry->d_inode;
1733 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1734 if (ffhp->fh_export != tfhp->fh_export)
1738 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1741 /* cannot use fh_lock as we need deadlock protective ordering
1742 * so do it by hand */
1743 trap = lock_rename(tdentry, fdentry);
1744 ffhp->fh_locked = tfhp->fh_locked = 1;
1748 odentry = lookup_one_len(fname, fdentry, flen);
1749 host_err = PTR_ERR(odentry);
1750 if (IS_ERR(odentry))
1754 if (!odentry->d_inode)
1757 if (odentry == trap)
1760 ndentry = lookup_one_len(tname, tdentry, tlen);
1761 host_err = PTR_ERR(ndentry);
1762 if (IS_ERR(ndentry))
1764 host_err = -ENOTEMPTY;
1765 if (ndentry == trap)
1768 if (svc_msnfs(ffhp) &&
1769 ((atomic_read(&odentry->d_count) > 1)
1770 || (atomic_read(&ndentry->d_count) > 1))) {
1776 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1778 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1782 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1783 if (!host_err && EX_ISSYNC(tfhp->fh_export)) {
1784 host_err = nfsd_sync_dir(tdentry);
1786 host_err = nfsd_sync_dir(fdentry);
1789 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1796 err = nfserrno(host_err);
1798 /* we cannot reply on fh_unlock on the two filehandles,
1799 * as that would do the wrong thing if the two directories
1800 * were the same, so again we do it by hand
1802 fill_post_wcc(ffhp);
1803 fill_post_wcc(tfhp);
1804 unlock_rename(tdentry, fdentry);
1805 ffhp->fh_locked = tfhp->fh_locked = 0;
1812 * Unlink a file or directory
1813 * N.B. After this call fhp needs an fh_put
1816 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1817 char *fname, int flen)
1819 struct dentry *dentry, *rdentry;
1825 if (!flen || isdotent(fname, flen))
1827 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1831 fh_lock_nested(fhp, I_MUTEX_PARENT);
1832 dentry = fhp->fh_dentry;
1833 dirp = dentry->d_inode;
1835 rdentry = lookup_one_len(fname, dentry, flen);
1836 host_err = PTR_ERR(rdentry);
1837 if (IS_ERR(rdentry))
1840 if (!rdentry->d_inode) {
1847 type = rdentry->d_inode->i_mode & S_IFMT;
1849 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1853 if (type != S_IFDIR) { /* It's UNLINK */
1855 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1856 (atomic_read(&rdentry->d_count) > 1)) {
1860 host_err = vfs_unlink(dirp, rdentry);
1861 } else { /* It's RMDIR */
1862 host_err = vfs_rmdir(dirp, rdentry);
1869 if (EX_ISSYNC(fhp->fh_export))
1870 host_err = nfsd_sync_dir(dentry);
1873 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1875 err = nfserrno(host_err);
1881 * We do this buffering because we must not call back into the file
1882 * system's ->lookup() method from the filldir callback. That may well
1883 * deadlock a number of file systems.
1885 * This is based heavily on the implementation of same in XFS.
1887 struct buffered_dirent {
1891 unsigned int d_type;
1895 struct readdir_data {
1901 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1902 loff_t offset, u64 ino, unsigned int d_type)
1904 struct readdir_data *buf = __buf;
1905 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1906 unsigned int reclen;
1908 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1909 if (buf->used + reclen > PAGE_SIZE) {
1914 de->namlen = namlen;
1915 de->offset = offset;
1917 de->d_type = d_type;
1918 memcpy(de->name, name, namlen);
1919 buf->used += reclen;
1924 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1925 struct readdir_cd *cdp, loff_t *offsetp)
1927 struct readdir_data buf;
1928 struct buffered_dirent *de;
1933 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1935 return nfserrno(-ENOMEM);
1940 struct inode *dir_inode = file->f_path.dentry->d_inode;
1941 unsigned int reclen;
1943 cdp->err = nfserr_eof; /* will be cleared on successful read */
1947 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1960 * Various filldir functions may end up calling back into
1961 * lookup_one_len() and the file system's ->lookup() method.
1962 * These expect i_mutex to be held, as it would within readdir.
1964 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1968 de = (struct buffered_dirent *)buf.dirent;
1970 offset = de->offset;
1972 if (func(cdp, de->name, de->namlen, de->offset,
1973 de->ino, de->d_type))
1976 if (cdp->err != nfs_ok)
1979 reclen = ALIGN(sizeof(*de) + de->namlen,
1982 de = (struct buffered_dirent *)((char *)de + reclen);
1984 mutex_unlock(&dir_inode->i_mutex);
1985 if (size > 0) /* We bailed out early */
1988 offset = vfs_llseek(file, 0, SEEK_CUR);
1991 free_page((unsigned long)(buf.dirent));
1994 return nfserrno(host_err);
2001 * Read entries from a directory.
2002 * The NFSv3/4 verifier we ignore for now.
2005 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
2006 struct readdir_cd *cdp, filldir_t func)
2010 loff_t offset = *offsetp;
2012 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2016 offset = vfs_llseek(file, offset, 0);
2018 err = nfserrno((int)offset);
2022 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2024 if (err == nfserr_eof || err == nfserr_toosmall)
2025 err = nfs_ok; /* can still be found in ->err */
2033 * Get file system stats
2034 * N.B. After this call fhp needs an fh_put
2037 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2039 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2040 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2045 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2047 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2051 * Check for a user's access permissions to this inode.
2054 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2055 struct dentry *dentry, int acc)
2057 struct inode *inode = dentry->d_inode;
2061 if (acc == NFSD_MAY_NOP)
2064 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2066 (acc & NFSD_MAY_READ)? " read" : "",
2067 (acc & NFSD_MAY_WRITE)? " write" : "",
2068 (acc & NFSD_MAY_EXEC)? " exec" : "",
2069 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2070 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2071 (acc & NFSD_MAY_LOCK)? " lock" : "",
2072 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2074 IS_IMMUTABLE(inode)? " immut" : "",
2075 IS_APPEND(inode)? " append" : "",
2076 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2077 dprintk(" owner %d/%d user %d/%d\n",
2078 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2081 /* Normally we reject any write/sattr etc access on a read-only file
2082 * system. But if it is IRIX doing check on write-access for a
2083 * device special file, we ignore rofs.
2085 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2086 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2087 if (exp_rdonly(rqstp, exp) ||
2088 __mnt_is_readonly(exp->ex_path.mnt))
2090 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2093 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2096 if (acc & NFSD_MAY_LOCK) {
2097 /* If we cannot rely on authentication in NLM requests,
2098 * just allow locks, otherwise require read permission, or
2101 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2104 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2107 * The file owner always gets access permission for accesses that
2108 * would normally be checked at open time. This is to make
2109 * file access work even when the client has done a fchmod(fd, 0).
2111 * However, `cp foo bar' should fail nevertheless when bar is
2112 * readonly. A sensible way to do this might be to reject all
2113 * attempts to truncate a read-only file, because a creat() call
2114 * always implies file truncation.
2115 * ... but this isn't really fair. A process may reasonably call
2116 * ftruncate on an open file descriptor on a file with perm 000.
2117 * We must trust the client to do permission checking - using "ACCESS"
2120 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2121 inode->i_uid == current_fsuid())
2124 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2125 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2127 /* Allow read access to binaries even when mode 111 */
2128 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2129 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2130 err = inode_permission(inode, MAY_EXEC);
2134 /* Do integrity (permission) checking now, but defer incrementing
2135 * IMA counts to the actual file open.
2137 path.mnt = exp->ex_path.mnt;
2138 path.dentry = dentry;
2139 err = ima_path_check(&path, acc & (MAY_READ | MAY_WRITE | MAY_EXEC),
2142 return err? nfserrno(err) : 0;
2146 nfsd_racache_shutdown(void)
2148 struct raparms *raparm, *last_raparm;
2151 dprintk("nfsd: freeing readahead buffers.\n");
2153 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2154 raparm = raparm_hash[i].pb_head;
2156 last_raparm = raparm;
2157 raparm = raparm->p_next;
2160 raparm_hash[i].pb_head = NULL;
2164 * Initialize readahead param cache
2167 nfsd_racache_init(int cache_size)
2172 struct raparms **raparm = NULL;
2175 if (raparm_hash[0].pb_head)
2177 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2180 cache_size = nperbucket * RAPARM_HASH_SIZE;
2182 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2184 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2185 spin_lock_init(&raparm_hash[i].pb_lock);
2187 raparm = &raparm_hash[i].pb_head;
2188 for (j = 0; j < nperbucket; j++) {
2189 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2192 raparm = &(*raparm)->p_next;
2197 nfsdstats.ra_size = cache_size;
2201 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2202 nfsd_racache_shutdown();
2206 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2208 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2210 struct inode *inode = fhp->fh_dentry->d_inode;
2214 struct posix_acl *acl;
2216 if (!IS_POSIXACL(inode))
2217 return ERR_PTR(-EOPNOTSUPP);
2220 case ACL_TYPE_ACCESS:
2221 name = POSIX_ACL_XATTR_ACCESS;
2223 case ACL_TYPE_DEFAULT:
2224 name = POSIX_ACL_XATTR_DEFAULT;
2227 return ERR_PTR(-EOPNOTSUPP);
2230 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2232 return ERR_PTR(size);
2234 acl = posix_acl_from_xattr(value, size);
2240 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2242 struct inode *inode = fhp->fh_dentry->d_inode;
2248 if (!IS_POSIXACL(inode) ||
2249 !inode->i_op->setxattr || !inode->i_op->removexattr)
2252 case ACL_TYPE_ACCESS:
2253 name = POSIX_ACL_XATTR_ACCESS;
2255 case ACL_TYPE_DEFAULT:
2256 name = POSIX_ACL_XATTR_DEFAULT;
2262 if (acl && acl->a_count) {
2263 size = posix_acl_xattr_size(acl->a_count);
2264 value = kmalloc(size, GFP_KERNEL);
2267 error = posix_acl_to_xattr(acl, value, size);
2274 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2278 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2280 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2283 error = vfs_removexattr(fhp->fh_dentry, name);
2284 if (error == -ENODATA)
2288 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2294 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */