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/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
37 #include <linux/nfs4_acl.h>
38 #include <linux/nfsd_idmap.h>
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms *p_next;
60 struct file_ra_state p_ra;
61 unsigned int p_hindex;
64 struct raparm_hbucket {
65 struct raparms *pb_head;
67 } ____cacheline_aligned_in_smp;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
82 struct svc_export **expp)
84 struct svc_export *exp = *expp, *exp2 = NULL;
85 struct dentry *dentry = *dpp;
86 struct path path = {.mnt = mntget(exp->ex_path.mnt),
87 .dentry = dget(dentry)};
90 while (d_mountpoint(path.dentry) && follow_down(&path))
93 exp2 = rqst_exp_get_by_name(rqstp, &path);
97 * We normally allow NFS clients to continue
98 * "underneath" a mountpoint that is not exported.
99 * The exception is V4ROOT, where no traversal is ever
100 * allowed without an explicit export of the new
103 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
108 if (nfsd_v4client(rqstp) ||
109 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
110 /* successfully crossed mount point */
112 * This is subtle: path.dentry is *not* on path.mnt
113 * at this point. The only reason we are safe is that
114 * original mnt is pinned down by exp, so we should
115 * put path *before* putting exp
118 path.dentry = dentry;
128 static void follow_to_parent(struct path *path)
132 while (path->dentry == path->mnt->mnt_root && follow_up(path))
134 dp = dget_parent(path->dentry);
139 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
141 struct svc_export *exp2;
142 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
143 .dentry = dget(dparent)};
145 follow_to_parent(&path);
147 exp2 = rqst_exp_parent(rqstp, &path);
148 if (PTR_ERR(exp2) == -ENOENT) {
149 *dentryp = dget(dparent);
150 } else if (IS_ERR(exp2)) {
152 return PTR_ERR(exp2);
154 *dentryp = dget(path.dentry);
163 * For nfsd purposes, we treat V4ROOT exports as though there was an
164 * export at *every* directory.
166 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
168 if (d_mountpoint(dentry))
170 if (!(exp->ex_flags & NFSEXP_V4ROOT))
172 return dentry->d_inode != NULL;
176 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
177 const char *name, unsigned int len,
178 struct svc_export **exp_ret, struct dentry **dentry_ret)
180 struct svc_export *exp;
181 struct dentry *dparent;
182 struct dentry *dentry;
186 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
188 /* Obtain dentry and export. */
189 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
193 dparent = fhp->fh_dentry;
194 exp = fhp->fh_export;
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name, len)) {
200 dentry = dget(dparent);
201 else if (dparent != exp->ex_path.dentry)
202 dentry = dget_parent(dparent);
203 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
204 dentry = dget(dparent); /* .. == . just like at / */
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
213 dentry = lookup_one_len(name, dparent, len);
214 host_err = PTR_ERR(dentry);
218 * check if we have crossed a mount point ...
220 if (nfsd_mountpoint(dentry, exp)) {
221 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
227 *dentry_ret = dentry;
233 return nfserrno(host_err);
237 * Look up one component of a pathname.
238 * N.B. After this call _both_ fhp and resfh need an fh_put
240 * If the lookup would cross a mountpoint, and the mounted filesystem
241 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
242 * accepted as it stands and the mounted directory is
243 * returned. Otherwise the covered directory is returned.
244 * NOTE: this mountpoint crossing is not supported properly by all
245 * clients and is explicitly disallowed for NFSv3
246 * NeilBrown <neilb@cse.unsw.edu.au>
249 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
250 unsigned int len, struct svc_fh *resfh)
252 struct svc_export *exp;
253 struct dentry *dentry;
256 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
259 err = check_nfsd_access(exp, rqstp);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err = fh_compose(resfh, exp, dentry, fhp);
267 if (!err && !dentry->d_inode)
276 * Commit metadata changes to stable storage.
279 commit_metadata(struct svc_fh *fhp)
281 struct inode *inode = fhp->fh_dentry->d_inode;
282 const struct export_operations *export_ops = inode->i_sb->s_export_op;
285 if (!EX_ISSYNC(fhp->fh_export))
288 if (export_ops->commit_metadata) {
289 error = export_ops->commit_metadata(inode);
291 struct writeback_control wbc = {
292 .sync_mode = WB_SYNC_ALL,
293 .nr_to_write = 0, /* metadata only */
296 error = sync_inode(inode, &wbc);
303 * Set various file attributes.
304 * N.B. After this call fhp needs an fh_put
307 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
308 int check_guard, time_t guardtime)
310 struct dentry *dentry;
312 int accmode = NFSD_MAY_SATTR;
318 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
319 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
320 if (iap->ia_valid & ATTR_SIZE)
324 err = fh_verify(rqstp, fhp, ftype, accmode);
328 dentry = fhp->fh_dentry;
329 inode = dentry->d_inode;
331 /* Ignore any mode updates on symlinks */
332 if (S_ISLNK(inode->i_mode))
333 iap->ia_valid &= ~ATTR_MODE;
339 * NFSv2 does not differentiate between "set-[ac]time-to-now"
340 * which only requires access, and "set-[ac]time-to-X" which
341 * requires ownership.
342 * So if it looks like it might be "set both to the same time which
343 * is close to now", and if inode_change_ok fails, then we
344 * convert to "set to now" instead of "set to explicit time"
346 * We only call inode_change_ok as the last test as technically
347 * it is not an interface that we should be using. It is only
348 * valid if the filesystem does not define it's own i_op->setattr.
350 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
351 #define MAX_TOUCH_TIME_ERROR (30*60)
352 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
353 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
357 * Now just make sure time is in the right ballpark.
358 * Solaris, at least, doesn't seem to care what the time
359 * request is. We require it be within 30 minutes of now.
361 time_t delta = iap->ia_atime.tv_sec - get_seconds();
364 if (delta < MAX_TOUCH_TIME_ERROR &&
365 inode_change_ok(inode, iap) != 0) {
367 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
368 * This will cause notify_change to set these times
371 iap->ia_valid &= ~BOTH_TIME_SET;
376 * The size case is special.
377 * It changes the file as well as the attributes.
379 if (iap->ia_valid & ATTR_SIZE) {
380 if (iap->ia_size < inode->i_size) {
381 err = nfsd_permission(rqstp, fhp->fh_export, dentry,
382 NFSD_MAY_TRUNC|NFSD_MAY_OWNER_OVERRIDE);
388 * If we are changing the size of the file, then
389 * we need to break all leases.
391 host_err = break_lease(inode, O_WRONLY | O_NONBLOCK);
392 if (host_err == -EWOULDBLOCK)
393 host_err = -ETIMEDOUT;
394 if (host_err) /* ENOMEM or EWOULDBLOCK */
397 host_err = get_write_access(inode);
402 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
404 put_write_access(inode);
409 /* sanitize the mode change */
410 if (iap->ia_valid & ATTR_MODE) {
411 iap->ia_mode &= S_IALLUGO;
412 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
415 /* Revoke setuid/setgid on chown */
416 if (!S_ISDIR(inode->i_mode) &&
417 (((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid) ||
418 ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid))) {
419 iap->ia_valid |= ATTR_KILL_PRIV;
420 if (iap->ia_valid & ATTR_MODE) {
421 /* we're setting mode too, just clear the s*id bits */
422 iap->ia_mode &= ~S_ISUID;
423 if (iap->ia_mode & S_IXGRP)
424 iap->ia_mode &= ~S_ISGID;
426 /* set ATTR_KILL_* bits and let VFS handle it */
427 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
431 /* Change the attributes. */
433 iap->ia_valid |= ATTR_CTIME;
435 err = nfserr_notsync;
436 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
438 host_err = notify_change(dentry, iap);
439 err = nfserrno(host_err);
443 put_write_access(inode);
445 if (EX_ISSYNC(fhp->fh_export))
446 write_inode_now(inode, 1);
451 err = nfserrno(host_err);
455 #if defined(CONFIG_NFSD_V2_ACL) || \
456 defined(CONFIG_NFSD_V3_ACL) || \
457 defined(CONFIG_NFSD_V4)
458 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
463 buflen = vfs_getxattr(dentry, key, NULL, 0);
467 *buf = kmalloc(buflen, GFP_KERNEL);
471 ret = vfs_getxattr(dentry, key, *buf, buflen);
478 #if defined(CONFIG_NFSD_V4)
480 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
487 buflen = posix_acl_xattr_size(pacl->a_count);
488 buf = kmalloc(buflen, GFP_KERNEL);
493 len = posix_acl_to_xattr(pacl, buf, buflen);
499 error = vfs_setxattr(dentry, key, buf, len, 0);
506 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
507 struct nfs4_acl *acl)
511 struct dentry *dentry;
513 struct posix_acl *pacl = NULL, *dpacl = NULL;
514 unsigned int flags = 0;
517 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
521 dentry = fhp->fh_dentry;
522 inode = dentry->d_inode;
523 if (S_ISDIR(inode->i_mode))
524 flags = NFS4_ACL_DIR;
526 host_error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
527 if (host_error == -EINVAL) {
528 return nfserr_attrnotsupp;
529 } else if (host_error < 0)
532 host_error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
536 if (S_ISDIR(inode->i_mode))
537 host_error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
540 posix_acl_release(pacl);
541 posix_acl_release(dpacl);
543 if (host_error == -EOPNOTSUPP)
544 return nfserr_attrnotsupp;
546 return nfserrno(host_error);
549 static struct posix_acl *
550 _get_posix_acl(struct dentry *dentry, char *key)
553 struct posix_acl *pacl = NULL;
556 buflen = nfsd_getxattr(dentry, key, &buf);
560 return ERR_PTR(buflen);
562 pacl = posix_acl_from_xattr(buf, buflen);
568 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
570 struct inode *inode = dentry->d_inode;
572 struct posix_acl *pacl = NULL, *dpacl = NULL;
573 unsigned int flags = 0;
575 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
576 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
577 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
579 error = PTR_ERR(pacl);
584 if (S_ISDIR(inode->i_mode)) {
585 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
586 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
588 else if (IS_ERR(dpacl)) {
589 error = PTR_ERR(dpacl);
593 flags = NFS4_ACL_DIR;
596 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
598 error = PTR_ERR(*acl);
602 posix_acl_release(pacl);
603 posix_acl_release(dpacl);
607 #endif /* defined(CONFIG_NFS_V4) */
609 #ifdef CONFIG_NFSD_V3
611 * Check server access rights to a file system object
617 static struct accessmap nfs3_regaccess[] = {
618 { NFS3_ACCESS_READ, NFSD_MAY_READ },
619 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
620 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
621 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
626 static struct accessmap nfs3_diraccess[] = {
627 { NFS3_ACCESS_READ, NFSD_MAY_READ },
628 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
629 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
630 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
631 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
636 static struct accessmap nfs3_anyaccess[] = {
637 /* Some clients - Solaris 2.6 at least, make an access call
638 * to the server to check for access for things like /dev/null
639 * (which really, the server doesn't care about). So
640 * We provide simple access checking for them, looking
641 * mainly at mode bits, and we make sure to ignore read-only
644 { NFS3_ACCESS_READ, NFSD_MAY_READ },
645 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
646 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
647 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
653 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
655 struct accessmap *map;
656 struct svc_export *export;
657 struct dentry *dentry;
658 u32 query, result = 0, sresult = 0;
661 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
665 export = fhp->fh_export;
666 dentry = fhp->fh_dentry;
668 if (S_ISREG(dentry->d_inode->i_mode))
669 map = nfs3_regaccess;
670 else if (S_ISDIR(dentry->d_inode->i_mode))
671 map = nfs3_diraccess;
673 map = nfs3_anyaccess;
677 for (; map->access; map++) {
678 if (map->access & query) {
681 sresult |= map->access;
683 err2 = nfsd_permission(rqstp, export, dentry, map->how);
686 result |= map->access;
689 /* the following error codes just mean the access was not allowed,
690 * rather than an error occurred */
694 /* simply don't "or" in the access bit. */
704 *supported = sresult;
709 #endif /* CONFIG_NFSD_V3 */
714 * Open an existing file or directory.
715 * The access argument indicates the type of open (read/write/lock)
716 * N.B. After this call fhp needs an fh_put
719 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
720 int access, struct file **filp)
722 struct dentry *dentry;
724 int flags = O_RDONLY|O_LARGEFILE;
728 validate_process_creds();
731 * If we get here, then the client has already done an "open",
732 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
733 * in case a chmod has now revoked permission.
735 err = fh_verify(rqstp, fhp, type, access | NFSD_MAY_OWNER_OVERRIDE);
739 dentry = fhp->fh_dentry;
740 inode = dentry->d_inode;
742 /* Disallow write access to files with the append-only bit set
743 * or any access when mandatory locking enabled
746 if (IS_APPEND(inode) && (access & NFSD_MAY_WRITE))
749 * We must ignore files (but only files) which might have mandatory
750 * locks on them because there is no way to know if the accesser has
753 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
760 * Check to see if there are any leases on this file.
761 * This may block while leases are broken.
763 if (!(access & NFSD_MAY_NOT_BREAK_LEASE))
764 host_err = break_lease(inode, O_NONBLOCK | ((access & NFSD_MAY_WRITE) ? O_WRONLY : 0));
765 if (host_err == -EWOULDBLOCK)
766 host_err = -ETIMEDOUT;
767 if (host_err) /* NOMEM or WOULDBLOCK */
770 if (access & NFSD_MAY_WRITE) {
771 if (access & NFSD_MAY_READ)
772 flags = O_RDWR|O_LARGEFILE;
774 flags = O_WRONLY|O_LARGEFILE;
776 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_path.mnt),
777 flags, current_cred());
779 host_err = PTR_ERR(*filp);
781 host_err = ima_file_check(*filp, access);
783 err = nfserrno(host_err);
785 validate_process_creds();
793 nfsd_close(struct file *filp)
799 * Obtain the readahead parameters for the file
800 * specified by (dev, ino).
803 static inline struct raparms *
804 nfsd_get_raparms(dev_t dev, ino_t ino)
806 struct raparms *ra, **rap, **frap = NULL;
809 struct raparm_hbucket *rab;
811 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
812 rab = &raparm_hash[hash];
814 spin_lock(&rab->pb_lock);
815 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
816 if (ra->p_ino == ino && ra->p_dev == dev)
819 if (ra->p_count == 0)
822 depth = nfsdstats.ra_size*11/10;
824 spin_unlock(&rab->pb_lock);
834 if (rap != &rab->pb_head) {
836 ra->p_next = rab->pb_head;
840 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
841 spin_unlock(&rab->pb_lock);
846 * Grab and keep cached pages associated with a file in the svc_rqst
847 * so that they can be passed to the network sendmsg/sendpage routines
848 * directly. They will be released after the sending has completed.
851 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
852 struct splice_desc *sd)
854 struct svc_rqst *rqstp = sd->u.data;
855 struct page **pp = rqstp->rq_respages + rqstp->rq_resused;
856 struct page *page = buf->page;
860 ret = buf->ops->confirm(pipe, buf);
866 if (rqstp->rq_res.page_len == 0) {
871 rqstp->rq_res.page_base = buf->offset;
872 rqstp->rq_res.page_len = size;
873 } else if (page != pp[-1]) {
879 rqstp->rq_res.page_len += size;
881 rqstp->rq_res.page_len += size;
886 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
887 struct splice_desc *sd)
889 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
892 static inline int svc_msnfs(struct svc_fh *ffhp)
895 return (ffhp->fh_export->ex_flags & NFSEXP_MSNFS);
902 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
903 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
912 inode = file->f_path.dentry->d_inode;
914 if (svc_msnfs(fhp) && !lock_may_read(inode, offset, *count))
917 /* Get readahead parameters */
918 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
921 file->f_ra = ra->p_ra;
923 if (file->f_op->splice_read && rqstp->rq_splice_ok) {
924 struct splice_desc sd = {
931 rqstp->rq_resused = 1;
932 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
936 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
940 /* Write back readahead params */
942 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
943 spin_lock(&rab->pb_lock);
944 ra->p_ra = file->f_ra;
947 spin_unlock(&rab->pb_lock);
951 nfsdstats.io_read += host_err;
954 fsnotify_access(file->f_path.dentry);
956 err = nfserrno(host_err);
961 static void kill_suid(struct dentry *dentry)
964 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_KILL_PRIV;
966 mutex_lock(&dentry->d_inode->i_mutex);
967 notify_change(dentry, &ia);
968 mutex_unlock(&dentry->d_inode->i_mutex);
972 * Gathered writes: If another process is currently writing to the file,
973 * there's a high chance this is another nfsd (triggered by a bulk write
974 * from a client's biod). Rather than syncing the file with each write
975 * request, we sleep for 10 msec.
977 * I don't know if this roughly approximates C. Juszak's idea of
978 * gathered writes, but it's a nice and simple solution (IMHO), and it
981 * Note: we do this only in the NFSv2 case, since v3 and higher have a
982 * better tool (separate unstable writes and commits) for solving this
985 static int wait_for_concurrent_writes(struct file *file)
987 struct inode *inode = file->f_path.dentry->d_inode;
988 static ino_t last_ino;
989 static dev_t last_dev;
992 if (atomic_read(&inode->i_writecount) > 1
993 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
994 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
996 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
999 if (inode->i_state & I_DIRTY) {
1000 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
1001 err = vfs_fsync(file, file->f_path.dentry, 0);
1003 last_ino = inode->i_ino;
1004 last_dev = inode->i_sb->s_dev;
1009 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1010 loff_t offset, struct kvec *vec, int vlen,
1011 unsigned long *cnt, int *stablep)
1013 struct svc_export *exp;
1014 struct dentry *dentry;
1015 struct inode *inode;
1019 int stable = *stablep;
1025 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1026 (!lock_may_write(file->f_path.dentry->d_inode, offset, *cnt)))
1030 dentry = file->f_path.dentry;
1031 inode = dentry->d_inode;
1032 exp = fhp->fh_export;
1035 * Request sync writes if
1036 * - the sync export option has been set, or
1037 * - the client requested O_SYNC behavior (NFSv3 feature).
1038 * - The file system doesn't support fsync().
1039 * When NFSv2 gathered writes have been configured for this volume,
1040 * flushing the data to disk is handled separately below.
1042 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
1044 if (!file->f_op->fsync) {/* COMMIT3 cannot work */
1046 *stablep = 2; /* FILE_SYNC */
1049 if (!EX_ISSYNC(exp))
1051 if (stable && !use_wgather) {
1052 spin_lock(&file->f_lock);
1053 file->f_flags |= O_SYNC;
1054 spin_unlock(&file->f_lock);
1057 /* Write the data. */
1058 oldfs = get_fs(); set_fs(KERNEL_DS);
1059 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
1064 nfsdstats.io_write += host_err;
1065 fsnotify_modify(file->f_path.dentry);
1067 /* clear setuid/setgid flag after write */
1068 if (inode->i_mode & (S_ISUID | S_ISGID))
1071 if (stable && use_wgather)
1072 host_err = wait_for_concurrent_writes(file);
1075 dprintk("nfsd: write complete host_err=%d\n", host_err);
1079 err = nfserrno(host_err);
1085 * Read data from a file. count must contain the requested read count
1086 * on entry. On return, *count contains the number of bytes actually read.
1087 * N.B. After this call fhp needs an fh_put
1090 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1091 loff_t offset, struct kvec *vec, int vlen,
1092 unsigned long *count)
1097 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1098 NFSD_MAY_READ|NFSD_MAY_OWNER_OVERRIDE);
1101 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1103 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1106 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1114 * Write data to a file.
1115 * The stable flag requests synchronous writes.
1116 * N.B. After this call fhp needs an fh_put
1119 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1120 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1126 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1127 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1130 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1133 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1138 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1146 #ifdef CONFIG_NFSD_V3
1148 * Commit all pending writes to stable storage.
1150 * Note: we only guarantee that data that lies within the range specified
1151 * by the 'offset' and 'count' parameters will be synced.
1153 * Unfortunately we cannot lock the file to make sure we return full WCC
1154 * data to the client, as locking happens lower down in the filesystem.
1157 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1158 loff_t offset, unsigned long count)
1161 loff_t end = LLONG_MAX;
1162 __be32 err = nfserr_inval;
1167 end = offset + (loff_t)count - 1;
1172 err = nfsd_open(rqstp, fhp, S_IFREG,
1173 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1176 if (EX_ISSYNC(fhp->fh_export)) {
1177 int err2 = vfs_fsync_range(file, file->f_path.dentry,
1180 if (err2 != -EINVAL)
1181 err = nfserrno(err2);
1183 err = nfserr_notsupp;
1190 #endif /* CONFIG_NFSD_V3 */
1193 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1197 * Mode has already been set earlier in create:
1199 iap->ia_valid &= ~ATTR_MODE;
1201 * Setting uid/gid works only for root. Irix appears to
1202 * send along the gid on create when it tries to implement
1203 * setgid directories via NFS:
1205 if (current_fsuid() != 0)
1206 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1208 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1212 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1213 * setting size to 0 may fail for some specific file systems by the permission
1214 * checking which requires WRITE permission but the mode is 000.
1215 * we ignore the resizing(to 0) on the just new created file, since the size is
1216 * 0 after file created.
1218 * call this only after vfs_create() is called.
1221 nfsd_check_ignore_resizing(struct iattr *iap)
1223 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1224 iap->ia_valid &= ~ATTR_SIZE;
1228 * Create a file (regular, directory, device, fifo); UNIX sockets
1229 * not yet implemented.
1230 * If the response fh has been verified, the parent directory should
1231 * already be locked. Note that the parent directory is left locked.
1233 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1236 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1237 char *fname, int flen, struct iattr *iap,
1238 int type, dev_t rdev, struct svc_fh *resfhp)
1240 struct dentry *dentry, *dchild = NULL;
1250 if (isdotent(fname, flen))
1253 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1257 dentry = fhp->fh_dentry;
1258 dirp = dentry->d_inode;
1260 err = nfserr_notdir;
1261 if (!dirp->i_op->lookup)
1264 * Check whether the response file handle has been verified yet.
1265 * If it has, the parent directory should already be locked.
1267 if (!resfhp->fh_dentry) {
1268 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1269 fh_lock_nested(fhp, I_MUTEX_PARENT);
1270 dchild = lookup_one_len(fname, dentry, flen);
1271 host_err = PTR_ERR(dchild);
1274 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1278 /* called from nfsd_proc_create */
1279 dchild = dget(resfhp->fh_dentry);
1280 if (!fhp->fh_locked) {
1281 /* not actually possible */
1283 "nfsd_create: parent %s/%s not locked!\n",
1284 dentry->d_parent->d_name.name,
1285 dentry->d_name.name);
1291 * Make sure the child dentry is still negative ...
1294 if (dchild->d_inode) {
1295 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1296 dentry->d_name.name, dchild->d_name.name);
1300 if (!(iap->ia_valid & ATTR_MODE))
1302 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1305 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1306 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1311 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1316 * Get the dir op function pointer.
1321 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1323 nfsd_check_ignore_resizing(iap);
1326 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1332 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1336 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1340 err = nfsd_create_setattr(rqstp, resfhp, iap);
1343 * nfsd_setattr already committed the child. Transactional filesystems
1344 * had a chance to commit changes for both parent and child
1345 * simultaneously making the following commit_metadata a noop.
1347 err2 = nfserrno(commit_metadata(fhp));
1350 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1352 * Update the file handle to get the new inode info.
1355 err = fh_update(resfhp);
1357 if (dchild && !IS_ERR(dchild))
1362 err = nfserrno(host_err);
1366 #ifdef CONFIG_NFSD_V3
1368 * NFSv3 version of nfsd_create
1371 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1372 char *fname, int flen, struct iattr *iap,
1373 struct svc_fh *resfhp, int createmode, u32 *verifier,
1374 int *truncp, int *created)
1376 struct dentry *dentry, *dchild = NULL;
1380 __u32 v_mtime=0, v_atime=0;
1386 if (isdotent(fname, flen))
1388 if (!(iap->ia_valid & ATTR_MODE))
1390 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1394 dentry = fhp->fh_dentry;
1395 dirp = dentry->d_inode;
1397 /* Get all the sanity checks out of the way before
1398 * we lock the parent. */
1399 err = nfserr_notdir;
1400 if (!dirp->i_op->lookup)
1402 fh_lock_nested(fhp, I_MUTEX_PARENT);
1405 * Compose the response file handle.
1407 dchild = lookup_one_len(fname, dentry, flen);
1408 host_err = PTR_ERR(dchild);
1412 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1416 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1417 /* solaris7 gets confused (bugid 4218508) if these have
1418 * the high bit set, so just clear the high bits. If this is
1419 * ever changed to use different attrs for storing the
1420 * verifier, then do_open_lookup() will also need to be fixed
1423 v_mtime = verifier[0]&0x7fffffff;
1424 v_atime = verifier[1]&0x7fffffff;
1427 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1430 if (dchild->d_inode) {
1433 switch (createmode) {
1434 case NFS3_CREATE_UNCHECKED:
1435 if (! S_ISREG(dchild->d_inode->i_mode))
1438 /* in nfsv4, we need to treat this case a little
1439 * differently. we don't want to truncate the
1440 * file now; this would be wrong if the OPEN
1441 * fails for some other reason. furthermore,
1442 * if the size is nonzero, we should ignore it
1443 * according to spec!
1445 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1448 iap->ia_valid &= ATTR_SIZE;
1452 case NFS3_CREATE_EXCLUSIVE:
1453 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1454 && dchild->d_inode->i_atime.tv_sec == v_atime
1455 && dchild->d_inode->i_size == 0 )
1458 case NFS3_CREATE_GUARDED:
1461 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1465 host_err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1467 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1473 nfsd_check_ignore_resizing(iap);
1475 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1476 /* Cram the verifier into atime/mtime */
1477 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1478 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1479 /* XXX someone who knows this better please fix it for nsec */
1480 iap->ia_mtime.tv_sec = v_mtime;
1481 iap->ia_atime.tv_sec = v_atime;
1482 iap->ia_mtime.tv_nsec = 0;
1483 iap->ia_atime.tv_nsec = 0;
1487 err = nfsd_create_setattr(rqstp, resfhp, iap);
1490 * nfsd_setattr already committed the child (and possibly also the parent).
1493 err = nfserrno(commit_metadata(fhp));
1495 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1497 * Update the filehandle to get the new inode info.
1500 err = fh_update(resfhp);
1504 if (dchild && !IS_ERR(dchild))
1509 err = nfserrno(host_err);
1512 #endif /* CONFIG_NFSD_V3 */
1515 * Read a symlink. On entry, *lenp must contain the maximum path length that
1516 * fits into the buffer. On return, it contains the true length.
1517 * N.B. After this call fhp needs an fh_put
1520 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1522 struct dentry *dentry;
1523 struct inode *inode;
1528 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1532 dentry = fhp->fh_dentry;
1533 inode = dentry->d_inode;
1536 if (!inode->i_op->readlink)
1539 touch_atime(fhp->fh_export->ex_path.mnt, dentry);
1540 /* N.B. Why does this call need a get_fs()??
1541 * Remove the set_fs and watch the fireworks:-) --okir
1544 oldfs = get_fs(); set_fs(KERNEL_DS);
1545 host_err = inode->i_op->readlink(dentry, buf, *lenp);
1556 err = nfserrno(host_err);
1561 * Create a symlink and look up its inode
1562 * N.B. After this call _both_ fhp and resfhp need an fh_put
1565 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1566 char *fname, int flen,
1567 char *path, int plen,
1568 struct svc_fh *resfhp,
1571 struct dentry *dentry, *dnew;
1579 if (isdotent(fname, flen))
1582 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1586 dentry = fhp->fh_dentry;
1587 dnew = lookup_one_len(fname, dentry, flen);
1588 host_err = PTR_ERR(dnew);
1592 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1596 if (unlikely(path[plen] != 0)) {
1597 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1598 if (path_alloced == NULL)
1601 strncpy(path_alloced, path, plen);
1602 path_alloced[plen] = 0;
1603 host_err = vfs_symlink(dentry->d_inode, dnew, path_alloced);
1604 kfree(path_alloced);
1607 host_err = vfs_symlink(dentry->d_inode, dnew, path);
1608 err = nfserrno(host_err);
1610 err = nfserrno(commit_metadata(fhp));
1613 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1615 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1617 if (err==0) err = cerr;
1622 err = nfserrno(host_err);
1628 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1631 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1632 char *name, int len, struct svc_fh *tfhp)
1634 struct dentry *ddir, *dnew, *dold;
1635 struct inode *dirp, *dest;
1639 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1642 err = fh_verify(rqstp, tfhp, -S_IFDIR, NFSD_MAY_NOP);
1650 if (isdotent(name, len))
1653 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1654 ddir = ffhp->fh_dentry;
1655 dirp = ddir->d_inode;
1657 dnew = lookup_one_len(name, ddir, len);
1658 host_err = PTR_ERR(dnew);
1662 dold = tfhp->fh_dentry;
1663 dest = dold->d_inode;
1665 host_err = mnt_want_write(tfhp->fh_export->ex_path.mnt);
1667 err = nfserrno(host_err);
1670 host_err = vfs_link(dold, dirp, dnew);
1672 err = nfserrno(commit_metadata(ffhp));
1674 err = nfserrno(commit_metadata(tfhp));
1676 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1679 err = nfserrno(host_err);
1681 mnt_drop_write(tfhp->fh_export->ex_path.mnt);
1690 err = nfserrno(host_err);
1696 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1699 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1700 struct svc_fh *tfhp, char *tname, int tlen)
1702 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1703 struct inode *fdir, *tdir;
1707 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1710 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1714 fdentry = ffhp->fh_dentry;
1715 fdir = fdentry->d_inode;
1717 tdentry = tfhp->fh_dentry;
1718 tdir = tdentry->d_inode;
1720 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1721 if (ffhp->fh_export != tfhp->fh_export)
1725 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1728 /* cannot use fh_lock as we need deadlock protective ordering
1729 * so do it by hand */
1730 trap = lock_rename(tdentry, fdentry);
1731 ffhp->fh_locked = tfhp->fh_locked = 1;
1735 odentry = lookup_one_len(fname, fdentry, flen);
1736 host_err = PTR_ERR(odentry);
1737 if (IS_ERR(odentry))
1741 if (!odentry->d_inode)
1744 if (odentry == trap)
1747 ndentry = lookup_one_len(tname, tdentry, tlen);
1748 host_err = PTR_ERR(ndentry);
1749 if (IS_ERR(ndentry))
1751 host_err = -ENOTEMPTY;
1752 if (ndentry == trap)
1755 if (svc_msnfs(ffhp) &&
1756 ((atomic_read(&odentry->d_count) > 1)
1757 || (atomic_read(&ndentry->d_count) > 1))) {
1763 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1765 host_err = mnt_want_write(ffhp->fh_export->ex_path.mnt);
1769 host_err = vfs_rename(fdir, odentry, tdir, ndentry);
1771 host_err = commit_metadata(tfhp);
1773 host_err = commit_metadata(ffhp);
1776 mnt_drop_write(ffhp->fh_export->ex_path.mnt);
1783 err = nfserrno(host_err);
1785 /* we cannot reply on fh_unlock on the two filehandles,
1786 * as that would do the wrong thing if the two directories
1787 * were the same, so again we do it by hand
1789 fill_post_wcc(ffhp);
1790 fill_post_wcc(tfhp);
1791 unlock_rename(tdentry, fdentry);
1792 ffhp->fh_locked = tfhp->fh_locked = 0;
1799 * Unlink a file or directory
1800 * N.B. After this call fhp needs an fh_put
1803 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1804 char *fname, int flen)
1806 struct dentry *dentry, *rdentry;
1812 if (!flen || isdotent(fname, flen))
1814 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1818 fh_lock_nested(fhp, I_MUTEX_PARENT);
1819 dentry = fhp->fh_dentry;
1820 dirp = dentry->d_inode;
1822 rdentry = lookup_one_len(fname, dentry, flen);
1823 host_err = PTR_ERR(rdentry);
1824 if (IS_ERR(rdentry))
1827 if (!rdentry->d_inode) {
1834 type = rdentry->d_inode->i_mode & S_IFMT;
1836 host_err = mnt_want_write(fhp->fh_export->ex_path.mnt);
1840 if (type != S_IFDIR) { /* It's UNLINK */
1842 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1843 (atomic_read(&rdentry->d_count) > 1)) {
1847 host_err = vfs_unlink(dirp, rdentry);
1848 } else { /* It's RMDIR */
1849 host_err = vfs_rmdir(dirp, rdentry);
1855 host_err = commit_metadata(fhp);
1857 mnt_drop_write(fhp->fh_export->ex_path.mnt);
1859 err = nfserrno(host_err);
1865 * We do this buffering because we must not call back into the file
1866 * system's ->lookup() method from the filldir callback. That may well
1867 * deadlock a number of file systems.
1869 * This is based heavily on the implementation of same in XFS.
1871 struct buffered_dirent {
1875 unsigned int d_type;
1879 struct readdir_data {
1885 static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1886 loff_t offset, u64 ino, unsigned int d_type)
1888 struct readdir_data *buf = __buf;
1889 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1890 unsigned int reclen;
1892 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1893 if (buf->used + reclen > PAGE_SIZE) {
1898 de->namlen = namlen;
1899 de->offset = offset;
1901 de->d_type = d_type;
1902 memcpy(de->name, name, namlen);
1903 buf->used += reclen;
1908 static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1909 struct readdir_cd *cdp, loff_t *offsetp)
1911 struct readdir_data buf;
1912 struct buffered_dirent *de;
1917 buf.dirent = (void *)__get_free_page(GFP_KERNEL);
1919 return nfserrno(-ENOMEM);
1924 struct inode *dir_inode = file->f_path.dentry->d_inode;
1925 unsigned int reclen;
1927 cdp->err = nfserr_eof; /* will be cleared on successful read */
1931 host_err = vfs_readdir(file, nfsd_buffered_filldir, &buf);
1944 * Various filldir functions may end up calling back into
1945 * lookup_one_len() and the file system's ->lookup() method.
1946 * These expect i_mutex to be held, as it would within readdir.
1948 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1952 de = (struct buffered_dirent *)buf.dirent;
1954 offset = de->offset;
1956 if (func(cdp, de->name, de->namlen, de->offset,
1957 de->ino, de->d_type))
1960 if (cdp->err != nfs_ok)
1963 reclen = ALIGN(sizeof(*de) + de->namlen,
1966 de = (struct buffered_dirent *)((char *)de + reclen);
1968 mutex_unlock(&dir_inode->i_mutex);
1969 if (size > 0) /* We bailed out early */
1972 offset = vfs_llseek(file, 0, SEEK_CUR);
1975 free_page((unsigned long)(buf.dirent));
1978 return nfserrno(host_err);
1985 * Read entries from a directory.
1986 * The NFSv3/4 verifier we ignore for now.
1989 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1990 struct readdir_cd *cdp, filldir_t func)
1994 loff_t offset = *offsetp;
1996 err = nfsd_open(rqstp, fhp, S_IFDIR, NFSD_MAY_READ, &file);
2000 offset = vfs_llseek(file, offset, 0);
2002 err = nfserrno((int)offset);
2006 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
2008 if (err == nfserr_eof || err == nfserr_toosmall)
2009 err = nfs_ok; /* can still be found in ->err */
2017 * Get file system stats
2018 * N.B. After this call fhp needs an fh_put
2021 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
2023 __be32 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
2024 if (!err && vfs_statfs(fhp->fh_dentry,stat))
2029 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
2031 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
2035 * Check for a user's access permissions to this inode.
2038 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
2039 struct dentry *dentry, int acc)
2041 struct inode *inode = dentry->d_inode;
2045 if (acc == NFSD_MAY_NOP)
2048 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2050 (acc & NFSD_MAY_READ)? " read" : "",
2051 (acc & NFSD_MAY_WRITE)? " write" : "",
2052 (acc & NFSD_MAY_EXEC)? " exec" : "",
2053 (acc & NFSD_MAY_SATTR)? " sattr" : "",
2054 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
2055 (acc & NFSD_MAY_LOCK)? " lock" : "",
2056 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
2058 IS_IMMUTABLE(inode)? " immut" : "",
2059 IS_APPEND(inode)? " append" : "",
2060 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
2061 dprintk(" owner %d/%d user %d/%d\n",
2062 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
2065 /* Normally we reject any write/sattr etc access on a read-only file
2066 * system. But if it is IRIX doing check on write-access for a
2067 * device special file, we ignore rofs.
2069 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2070 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
2071 if (exp_rdonly(rqstp, exp) ||
2072 __mnt_is_readonly(exp->ex_path.mnt))
2074 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
2077 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
2080 if (acc & NFSD_MAY_LOCK) {
2081 /* If we cannot rely on authentication in NLM requests,
2082 * just allow locks, otherwise require read permission, or
2085 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2088 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2091 * The file owner always gets access permission for accesses that
2092 * would normally be checked at open time. This is to make
2093 * file access work even when the client has done a fchmod(fd, 0).
2095 * However, `cp foo bar' should fail nevertheless when bar is
2096 * readonly. A sensible way to do this might be to reject all
2097 * attempts to truncate a read-only file, because a creat() call
2098 * always implies file truncation.
2099 * ... but this isn't really fair. A process may reasonably call
2100 * ftruncate on an open file descriptor on a file with perm 000.
2101 * We must trust the client to do permission checking - using "ACCESS"
2104 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2105 inode->i_uid == current_fsuid())
2108 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2109 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2111 /* Allow read access to binaries even when mode 111 */
2112 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2113 acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE))
2114 err = inode_permission(inode, MAY_EXEC);
2118 /* Do integrity (permission) checking now, but defer incrementing
2119 * IMA counts to the actual file open.
2121 path.mnt = exp->ex_path.mnt;
2122 path.dentry = dentry;
2124 return err? nfserrno(err) : 0;
2128 nfsd_racache_shutdown(void)
2130 struct raparms *raparm, *last_raparm;
2133 dprintk("nfsd: freeing readahead buffers.\n");
2135 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2136 raparm = raparm_hash[i].pb_head;
2138 last_raparm = raparm;
2139 raparm = raparm->p_next;
2142 raparm_hash[i].pb_head = NULL;
2146 * Initialize readahead param cache
2149 nfsd_racache_init(int cache_size)
2154 struct raparms **raparm = NULL;
2157 if (raparm_hash[0].pb_head)
2159 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2162 cache_size = nperbucket * RAPARM_HASH_SIZE;
2164 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2166 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2167 spin_lock_init(&raparm_hash[i].pb_lock);
2169 raparm = &raparm_hash[i].pb_head;
2170 for (j = 0; j < nperbucket; j++) {
2171 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2174 raparm = &(*raparm)->p_next;
2179 nfsdstats.ra_size = cache_size;
2183 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2184 nfsd_racache_shutdown();
2188 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2190 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
2192 struct inode *inode = fhp->fh_dentry->d_inode;
2196 struct posix_acl *acl;
2198 if (!IS_POSIXACL(inode))
2199 return ERR_PTR(-EOPNOTSUPP);
2202 case ACL_TYPE_ACCESS:
2203 name = POSIX_ACL_XATTR_ACCESS;
2205 case ACL_TYPE_DEFAULT:
2206 name = POSIX_ACL_XATTR_DEFAULT;
2209 return ERR_PTR(-EOPNOTSUPP);
2212 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
2214 return ERR_PTR(size);
2216 acl = posix_acl_from_xattr(value, size);
2222 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
2224 struct inode *inode = fhp->fh_dentry->d_inode;
2230 if (!IS_POSIXACL(inode) ||
2231 !inode->i_op->setxattr || !inode->i_op->removexattr)
2234 case ACL_TYPE_ACCESS:
2235 name = POSIX_ACL_XATTR_ACCESS;
2237 case ACL_TYPE_DEFAULT:
2238 name = POSIX_ACL_XATTR_DEFAULT;
2244 if (acl && acl->a_count) {
2245 size = posix_acl_xattr_size(acl->a_count);
2246 value = kmalloc(size, GFP_KERNEL);
2249 error = posix_acl_to_xattr(acl, value, size);
2256 error = mnt_want_write(fhp->fh_export->ex_path.mnt);
2260 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
2262 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
2265 error = vfs_removexattr(fhp->fh_dentry, name);
2266 if (error == -ENODATA)
2270 mnt_drop_write(fhp->fh_export->ex_path.mnt);
2276 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */