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/config.h>
20 #include <linux/string.h>
21 #include <linux/time.h>
22 #include <linux/errno.h>
24 #include <linux/file.h>
25 #include <linux/mount.h>
26 #include <linux/major.h>
27 #include <linux/ext2_fs.h>
28 #include <linux/proc_fs.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/net.h>
32 #include <linux/unistd.h>
33 #include <linux/slab.h>
34 #include <linux/pagemap.h>
36 #include <linux/module.h>
37 #include <linux/namei.h>
38 #include <linux/vfs.h>
39 #include <linux/delay.h>
40 #include <linux/sunrpc/svc.h>
41 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfs3.h>
44 #include <linux/nfsd/xdr3.h>
45 #endif /* CONFIG_NFSD_V3 */
46 #include <linux/nfsd/nfsfh.h>
47 #include <linux/quotaops.h>
48 #include <linux/fsnotify.h>
49 #include <linux/posix_acl.h>
50 #include <linux/posix_acl_xattr.h>
52 #include <linux/xattr.h>
53 #include <linux/nfs4.h>
54 #include <linux/nfs4_acl.h>
55 #include <linux/nfsd_idmap.h>
56 #include <linux/security.h>
57 #endif /* CONFIG_NFSD_V4 */
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 /* We must ignore files (but only files) which might have mandatory
66 * locks on them because there is no way to know if the accesser has
69 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
72 * This is a cache of readahead params that help us choose the proper
73 * readahead strategy. Initially, we set all readahead parameters to 0
74 * and let the VFS handle things.
75 * If you increase the number of cached files very much, you'll need to
76 * add a hash table here.
79 struct raparms *p_next;
84 struct file_ra_state p_ra;
87 static struct raparms * raparml;
88 static struct raparms * raparm_cache;
91 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
93 * Returns -EAGAIN leaving *dpp and *expp unchanged,
94 * or nfs_ok having possibly changed *dpp and *expp
97 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
98 struct svc_export **expp)
100 struct svc_export *exp = *expp, *exp2 = NULL;
101 struct dentry *dentry = *dpp;
102 struct vfsmount *mnt = mntget(exp->ex_mnt);
103 struct dentry *mounts = dget(dentry);
106 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
108 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
115 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
116 /* successfully crossed mount point */
122 if (exp2) exp_put(exp2);
131 * Look up one component of a pathname.
132 * N.B. After this call _both_ fhp and resfh need an fh_put
134 * If the lookup would cross a mountpoint, and the mounted filesystem
135 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
136 * accepted as it stands and the mounted directory is
137 * returned. Otherwise the covered directory is returned.
138 * NOTE: this mountpoint crossing is not supported properly by all
139 * clients and is explicitly disallowed for NFSv3
140 * NeilBrown <neilb@cse.unsw.edu.au>
143 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
144 int len, struct svc_fh *resfh)
146 struct svc_export *exp;
147 struct dentry *dparent;
148 struct dentry *dentry;
151 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
153 /* Obtain dentry and export. */
154 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
158 dparent = fhp->fh_dentry;
159 exp = fhp->fh_export;
164 /* Lookup the name, but don't follow links */
165 if (isdotent(name, len)) {
167 dentry = dget(dparent);
168 else if (dparent != exp->ex_dentry) {
169 dentry = dget_parent(dparent);
170 } else if (!EX_NOHIDE(exp))
171 dentry = dget(dparent); /* .. == . just like at / */
173 /* checking mountpoint crossing is very different when stepping up */
174 struct svc_export *exp2 = NULL;
176 struct vfsmount *mnt = mntget(exp->ex_mnt);
177 dentry = dget(dparent);
178 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
180 dp = dget_parent(dentry);
184 exp2 = exp_parent(exp->ex_client, mnt, dentry,
194 dentry = dget(dparent);
203 dentry = lookup_one_len(name, dparent, len);
204 err = PTR_ERR(dentry);
208 * check if we have crossed a mount point ...
210 if (d_mountpoint(dentry)) {
211 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
218 * Note: we compose the file handle now, but as the
219 * dentry may be negative, it may need to be updated.
221 err = fh_compose(resfh, exp, dentry, fhp);
222 if (!err && !dentry->d_inode)
235 * Set various file attributes.
236 * N.B. After this call fhp needs an fh_put
239 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
240 int check_guard, time_t guardtime)
242 struct dentry *dentry;
244 int accmode = MAY_SATTR;
250 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
251 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
252 if (iap->ia_valid & ATTR_SIZE)
256 err = fh_verify(rqstp, fhp, ftype, accmode);
260 dentry = fhp->fh_dentry;
261 inode = dentry->d_inode;
263 /* Ignore any mode updates on symlinks */
264 if (S_ISLNK(inode->i_mode))
265 iap->ia_valid &= ~ATTR_MODE;
270 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
271 * which only requires access, and "set-[ac]time-to-X" which
272 * requires ownership.
273 * So if it looks like it might be "set both to the same time which
274 * is close to now", and if inode_change_ok fails, then we
275 * convert to "set to now" instead of "set to explicit time"
277 * We only call inode_change_ok as the last test as technically
278 * it is not an interface that we should be using. It is only
279 * valid if the filesystem does not define it's own i_op->setattr.
281 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
282 #define MAX_TOUCH_TIME_ERROR (30*60)
283 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
284 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
286 /* Looks probable. Now just make sure time is in the right ballpark.
287 * Solaris, at least, doesn't seem to care what the time request is.
288 * We require it be within 30 minutes of now.
290 time_t delta = iap->ia_atime.tv_sec - get_seconds();
291 if (delta<0) delta = -delta;
292 if (delta < MAX_TOUCH_TIME_ERROR &&
293 inode_change_ok(inode, iap) != 0) {
294 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
295 * this will cause notify_change to set these times to "now"
297 iap->ia_valid &= ~BOTH_TIME_SET;
301 /* The size case is special. It changes the file as well as the attributes. */
302 if (iap->ia_valid & ATTR_SIZE) {
303 if (iap->ia_size < inode->i_size) {
304 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
310 * If we are changing the size of the file, then
311 * we need to break all leases.
313 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
314 if (err == -EWOULDBLOCK)
316 if (err) /* ENOMEM or EWOULDBLOCK */
319 err = get_write_access(inode);
324 err = locks_verify_truncate(inode, NULL, iap->ia_size);
326 put_write_access(inode);
332 imode = inode->i_mode;
333 if (iap->ia_valid & ATTR_MODE) {
334 iap->ia_mode &= S_IALLUGO;
335 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
338 /* Revoke setuid/setgid bit on chown/chgrp */
339 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
340 iap->ia_valid |= ATTR_KILL_SUID;
341 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
342 iap->ia_valid |= ATTR_KILL_SGID;
344 /* Change the attributes. */
346 iap->ia_valid |= ATTR_CTIME;
348 err = nfserr_notsync;
349 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
351 err = notify_change(dentry, iap);
356 put_write_access(inode);
358 if (EX_ISSYNC(fhp->fh_export))
359 write_inode_now(inode, 1);
368 #if defined(CONFIG_NFSD_V4)
371 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
377 struct inode *inode = dentry->d_inode;
379 buflen = posix_acl_xattr_size(pacl->a_count);
380 buf = kmalloc(buflen, GFP_KERNEL);
385 len = posix_acl_to_xattr(pacl, buf, buflen);
392 if (inode->i_op && inode->i_op->setxattr) {
394 security_inode_setxattr(dentry, key, buf, len, 0);
395 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
397 security_inode_post_setxattr(dentry, key, buf, len, 0);
406 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
407 struct nfs4_acl *acl)
410 struct dentry *dentry;
412 struct posix_acl *pacl = NULL, *dpacl = NULL;
413 unsigned int flags = 0;
416 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
420 dentry = fhp->fh_dentry;
421 inode = dentry->d_inode;
422 if (S_ISDIR(inode->i_mode))
423 flags = NFS4_ACL_DIR;
425 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
426 if (error == -EINVAL) {
427 error = nfserr_attrnotsupp;
429 } else if (error < 0)
433 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
439 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
447 posix_acl_release(pacl);
448 posix_acl_release(dpacl);
451 error = nfserrno(error);
455 static struct posix_acl *
456 _get_posix_acl(struct dentry *dentry, char *key)
458 struct inode *inode = dentry->d_inode;
460 int buflen, error = 0;
461 struct posix_acl *pacl = NULL;
464 if (inode->i_op == NULL)
466 if (inode->i_op->getxattr == NULL)
469 error = security_inode_getxattr(dentry, key);
473 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
475 error = buflen < 0 ? buflen : -ENODATA;
479 buf = kmalloc(buflen, GFP_KERNEL);
485 error = inode->i_op->getxattr(dentry, key, buf, buflen);
489 pacl = posix_acl_from_xattr(buf, buflen);
494 pacl = ERR_PTR(error);
499 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
501 struct inode *inode = dentry->d_inode;
503 struct posix_acl *pacl = NULL, *dpacl = NULL;
504 unsigned int flags = 0;
506 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
507 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
508 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
510 error = PTR_ERR(pacl);
515 if (S_ISDIR(inode->i_mode)) {
516 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
517 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
519 else if (IS_ERR(dpacl)) {
520 error = PTR_ERR(dpacl);
524 flags = NFS4_ACL_DIR;
527 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
529 error = PTR_ERR(*acl);
533 posix_acl_release(pacl);
534 posix_acl_release(dpacl);
538 #endif /* defined(CONFIG_NFS_V4) */
540 #ifdef CONFIG_NFSD_V3
542 * Check server access rights to a file system object
548 static struct accessmap nfs3_regaccess[] = {
549 { NFS3_ACCESS_READ, MAY_READ },
550 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
551 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
552 { NFS3_ACCESS_EXTEND, MAY_WRITE },
557 static struct accessmap nfs3_diraccess[] = {
558 { NFS3_ACCESS_READ, MAY_READ },
559 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
560 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
561 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
562 { NFS3_ACCESS_DELETE, MAY_REMOVE },
567 static struct accessmap nfs3_anyaccess[] = {
568 /* Some clients - Solaris 2.6 at least, make an access call
569 * to the server to check for access for things like /dev/null
570 * (which really, the server doesn't care about). So
571 * We provide simple access checking for them, looking
572 * mainly at mode bits, and we make sure to ignore read-only
575 { NFS3_ACCESS_READ, MAY_READ },
576 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
577 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
578 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
584 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
586 struct accessmap *map;
587 struct svc_export *export;
588 struct dentry *dentry;
589 u32 query, result = 0, sresult = 0;
592 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
596 export = fhp->fh_export;
597 dentry = fhp->fh_dentry;
599 if (S_ISREG(dentry->d_inode->i_mode))
600 map = nfs3_regaccess;
601 else if (S_ISDIR(dentry->d_inode->i_mode))
602 map = nfs3_diraccess;
604 map = nfs3_anyaccess;
608 for (; map->access; map++) {
609 if (map->access & query) {
612 sresult |= map->access;
614 err2 = nfsd_permission(export, dentry, map->how);
617 result |= map->access;
620 /* the following error codes just mean the access was not allowed,
621 * rather than an error occurred */
625 /* simply don't "or" in the access bit. */
635 *supported = sresult;
640 #endif /* CONFIG_NFSD_V3 */
645 * Open an existing file or directory.
646 * The access argument indicates the type of open (read/write/lock)
647 * N.B. After this call fhp needs an fh_put
650 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
651 int access, struct file **filp)
653 struct dentry *dentry;
655 int flags = O_RDONLY|O_LARGEFILE, err;
658 * If we get here, then the client has already done an "open",
659 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
660 * in case a chmod has now revoked permission.
662 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
666 dentry = fhp->fh_dentry;
667 inode = dentry->d_inode;
669 /* Disallow write access to files with the append-only bit set
670 * or any access when mandatory locking enabled
673 if (IS_APPEND(inode) && (access & MAY_WRITE))
675 if (IS_ISMNDLK(inode))
682 * Check to see if there are any leases on this file.
683 * This may block while leases are broken.
685 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
686 if (err == -EWOULDBLOCK)
688 if (err) /* NOMEM or WOULDBLOCK */
691 if (access & MAY_WRITE) {
692 flags = O_WRONLY|O_LARGEFILE;
696 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
698 err = PTR_ERR(*filp);
710 nfsd_close(struct file *filp)
717 * As this calls fsync (not fdatasync) there is no need for a write_inode
720 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
721 struct file_operations *fop)
723 struct inode *inode = dp->d_inode;
724 int (*fsync) (struct file *, struct dentry *, int);
727 filemap_fdatawrite(inode->i_mapping);
728 if (fop && (fsync = fop->fsync))
729 err=fsync(filp, dp, 0);
730 filemap_fdatawait(inode->i_mapping);
732 return nfserrno(err);
737 nfsd_sync(struct file *filp)
740 struct inode *inode = filp->f_dentry->d_inode;
741 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
743 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
750 nfsd_sync_dir(struct dentry *dp)
752 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
756 * Obtain the readahead parameters for the file
757 * specified by (dev, ino).
759 static DEFINE_SPINLOCK(ra_lock);
761 static inline struct raparms *
762 nfsd_get_raparms(dev_t dev, ino_t ino)
764 struct raparms *ra, **rap, **frap = NULL;
768 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
769 if (ra->p_ino == ino && ra->p_dev == dev)
772 if (ra->p_count == 0)
775 depth = nfsdstats.ra_size*11/10;
777 spin_unlock(&ra_lock);
786 if (rap != &raparm_cache) {
788 ra->p_next = raparm_cache;
792 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
793 spin_unlock(&ra_lock);
798 * Grab and keep cached pages assosiated with a file in the svc_rqst
799 * so that they can be passed to the netowork sendmsg/sendpage routines
800 * directrly. They will be released after the sending has completed.
803 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
805 unsigned long count = desc->count;
806 struct svc_rqst *rqstp = desc->arg.data;
811 if (rqstp->rq_res.page_len == 0) {
813 rqstp->rq_respages[rqstp->rq_resused++] = page;
814 rqstp->rq_res.page_base = offset;
815 rqstp->rq_res.page_len = size;
816 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
818 rqstp->rq_respages[rqstp->rq_resused++] = page;
819 rqstp->rq_res.page_len += size;
821 rqstp->rq_res.page_len += size;
824 desc->count = count - size;
825 desc->written += size;
830 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
831 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
839 inode = file->f_dentry->d_inode;
841 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
842 (!lock_may_read(inode, offset, *count)))
846 /* Get readahead parameters */
847 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
850 file->f_ra = ra->p_ra;
852 if (file->f_op->sendfile) {
853 svc_pushback_unused_pages(rqstp);
854 err = file->f_op->sendfile(file, &offset, *count,
855 nfsd_read_actor, rqstp);
859 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
863 /* Write back readahead params */
866 ra->p_ra = file->f_ra;
869 spin_unlock(&ra_lock);
873 nfsdstats.io_read += err;
876 fsnotify_access(file->f_dentry);
883 static void kill_suid(struct dentry *dentry)
886 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
888 down(&dentry->d_inode->i_sem);
889 notify_change(dentry, &ia);
890 up(&dentry->d_inode->i_sem);
894 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
895 loff_t offset, struct kvec *vec, int vlen,
896 unsigned long cnt, int *stablep)
898 struct svc_export *exp;
899 struct dentry *dentry;
903 int stable = *stablep;
908 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
909 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
913 dentry = file->f_dentry;
914 inode = dentry->d_inode;
915 exp = fhp->fh_export;
918 * Request sync writes if
919 * - the sync export option has been set, or
920 * - the client requested O_SYNC behavior (NFSv3 feature).
921 * - The file system doesn't support fsync().
922 * When gathered writes have been configured for this volume,
923 * flushing the data to disk is handled separately below.
926 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
928 *stablep = 2; /* FILE_SYNC */
933 if (stable && !EX_WGATHER(exp))
934 file->f_flags |= O_SYNC;
936 /* Write the data. */
937 oldfs = get_fs(); set_fs(KERNEL_DS);
938 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
941 nfsdstats.io_write += cnt;
942 fsnotify_modify(file->f_dentry);
945 /* clear setuid/setgid flag after write */
946 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
949 if (err >= 0 && stable) {
950 static ino_t last_ino;
951 static dev_t last_dev;
954 * Gathered writes: If another process is currently
955 * writing to the file, there's a high chance
956 * this is another nfsd (triggered by a bulk write
957 * from a client's biod). Rather than syncing the
958 * file with each write request, we sleep for 10 msec.
960 * I don't know if this roughly approximates
961 * C. Juszak's idea of gathered writes, but it's a
962 * nice and simple solution (IMHO), and it seems to
965 if (EX_WGATHER(exp)) {
966 if (atomic_read(&inode->i_writecount) > 1
967 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
968 dprintk("nfsd: write defer %d\n", current->pid);
970 dprintk("nfsd: write resume %d\n", current->pid);
973 if (inode->i_state & I_DIRTY) {
974 dprintk("nfsd: write sync %d\n", current->pid);
978 wake_up(&inode->i_wait);
981 last_ino = inode->i_ino;
982 last_dev = inode->i_sb->s_dev;
985 dprintk("nfsd: write complete err=%d\n", err);
995 * Read data from a file. count must contain the requested read count
996 * on entry. On return, *count contains the number of bytes actually read.
997 * N.B. After this call fhp needs an fh_put
1000 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1001 loff_t offset, struct kvec *vec, int vlen,
1002 unsigned long *count)
1007 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1008 MAY_READ|MAY_OWNER_OVERRIDE);
1011 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1013 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1016 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1024 * Write data to a file.
1025 * The stable flag requests synchronous writes.
1026 * N.B. After this call fhp needs an fh_put
1029 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1030 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1036 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1037 MAY_WRITE|MAY_OWNER_OVERRIDE);
1040 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1043 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1048 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1056 #ifdef CONFIG_NFSD_V3
1058 * Commit all pending writes to stable storage.
1059 * Strictly speaking, we could sync just the indicated file region here,
1060 * but there's currently no way we can ask the VFS to do so.
1062 * Unfortunately we cannot lock the file to make sure we return full WCC
1063 * data to the client, as locking happens lower down in the filesystem.
1066 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1067 loff_t offset, unsigned long count)
1072 if ((u64)count > ~(u64)offset)
1073 return nfserr_inval;
1075 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1077 if (EX_ISSYNC(fhp->fh_export)) {
1078 if (file->f_op && file->f_op->fsync) {
1079 err = nfsd_sync(file);
1081 err = nfserr_notsupp;
1088 #endif /* CONFIG_NFSD_V3 */
1091 * Create a file (regular, directory, device, fifo); UNIX sockets
1092 * not yet implemented.
1093 * If the response fh has been verified, the parent directory should
1094 * already be locked. Note that the parent directory is left locked.
1096 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1099 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1100 char *fname, int flen, struct iattr *iap,
1101 int type, dev_t rdev, struct svc_fh *resfhp)
1103 struct dentry *dentry, *dchild = NULL;
1111 if (isdotent(fname, flen))
1114 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1118 dentry = fhp->fh_dentry;
1119 dirp = dentry->d_inode;
1121 err = nfserr_notdir;
1122 if(!dirp->i_op || !dirp->i_op->lookup)
1125 * Check whether the response file handle has been verified yet.
1126 * If it has, the parent directory should already be locked.
1128 if (!resfhp->fh_dentry) {
1129 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1131 dchild = lookup_one_len(fname, dentry, flen);
1132 err = PTR_ERR(dchild);
1135 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1139 /* called from nfsd_proc_create */
1140 dchild = dget(resfhp->fh_dentry);
1141 if (!fhp->fh_locked) {
1142 /* not actually possible */
1144 "nfsd_create: parent %s/%s not locked!\n",
1145 dentry->d_parent->d_name.name,
1146 dentry->d_name.name);
1152 * Make sure the child dentry is still negative ...
1155 if (dchild->d_inode) {
1156 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1157 dentry->d_name.name, dchild->d_name.name);
1161 if (!(iap->ia_valid & ATTR_MODE))
1163 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1166 * Get the dir op function pointer.
1171 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1174 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1180 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1183 printk("nfsd: bad file type %o in nfsd_create\n", type);
1189 if (EX_ISSYNC(fhp->fh_export)) {
1190 nfsd_sync_dir(dentry);
1191 write_inode_now(dchild->d_inode, 1);
1195 /* Set file attributes. Mode has already been set and
1196 * setting uid/gid works only for root. Irix appears to
1197 * send along the gid when it tries to implement setgid
1198 * directories via NFS.
1201 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1202 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1204 * Update the file handle to get the new inode info.
1207 err = fh_update(resfhp);
1209 if (dchild && !IS_ERR(dchild))
1214 err = nfserrno(err);
1218 #ifdef CONFIG_NFSD_V3
1220 * NFSv3 version of nfsd_create
1223 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1224 char *fname, int flen, struct iattr *iap,
1225 struct svc_fh *resfhp, int createmode, u32 *verifier,
1228 struct dentry *dentry, *dchild = NULL;
1231 __u32 v_mtime=0, v_atime=0;
1238 if (isdotent(fname, flen))
1240 if (!(iap->ia_valid & ATTR_MODE))
1242 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1246 dentry = fhp->fh_dentry;
1247 dirp = dentry->d_inode;
1249 /* Get all the sanity checks out of the way before
1250 * we lock the parent. */
1251 err = nfserr_notdir;
1252 if(!dirp->i_op || !dirp->i_op->lookup)
1257 * Compose the response file handle.
1259 dchild = lookup_one_len(fname, dentry, flen);
1260 err = PTR_ERR(dchild);
1264 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1268 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1269 /* while the verifier would fit in mtime+atime,
1270 * solaris7 gets confused (bugid 4218508) if these have
1271 * the high bit set, so we use the mode as well
1273 v_mtime = verifier[0]&0x7fffffff;
1274 v_atime = verifier[1]&0x7fffffff;
1276 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1277 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1281 if (dchild->d_inode) {
1284 switch (createmode) {
1285 case NFS3_CREATE_UNCHECKED:
1286 if (! S_ISREG(dchild->d_inode->i_mode))
1289 /* in nfsv4, we need to treat this case a little
1290 * differently. we don't want to truncate the
1291 * file now; this would be wrong if the OPEN
1292 * fails for some other reason. furthermore,
1293 * if the size is nonzero, we should ignore it
1294 * according to spec!
1296 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1299 iap->ia_valid &= ATTR_SIZE;
1303 case NFS3_CREATE_EXCLUSIVE:
1304 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1305 && dchild->d_inode->i_atime.tv_sec == v_atime
1306 && dchild->d_inode->i_mode == v_mode
1307 && dchild->d_inode->i_size == 0 )
1310 case NFS3_CREATE_GUARDED:
1316 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1320 if (EX_ISSYNC(fhp->fh_export)) {
1321 nfsd_sync_dir(dentry);
1322 /* setattr will sync the child (or not) */
1326 * Update the filehandle to get the new inode info.
1328 err = fh_update(resfhp);
1332 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1333 /* Cram the verifier into atime/mtime/mode */
1334 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1335 | ATTR_MTIME_SET|ATTR_ATIME_SET
1337 /* XXX someone who knows this better please fix it for nsec */
1338 iap->ia_mtime.tv_sec = v_mtime;
1339 iap->ia_atime.tv_sec = v_atime;
1340 iap->ia_mtime.tv_nsec = 0;
1341 iap->ia_atime.tv_nsec = 0;
1342 iap->ia_mode = v_mode;
1345 /* Set file attributes.
1346 * Mode has already been set but we might need to reset it
1347 * for CREATE_EXCLUSIVE
1348 * Irix appears to send along the gid when it tries to
1349 * implement setgid directories via NFS. Clear out all that cruft.
1352 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1353 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1357 if (dchild && !IS_ERR(dchild))
1362 err = nfserrno(err);
1365 #endif /* CONFIG_NFSD_V3 */
1368 * Read a symlink. On entry, *lenp must contain the maximum path length that
1369 * fits into the buffer. On return, it contains the true length.
1370 * N.B. After this call fhp needs an fh_put
1373 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1375 struct dentry *dentry;
1376 struct inode *inode;
1380 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1384 dentry = fhp->fh_dentry;
1385 inode = dentry->d_inode;
1388 if (!inode->i_op || !inode->i_op->readlink)
1391 touch_atime(fhp->fh_export->ex_mnt, dentry);
1392 /* N.B. Why does this call need a get_fs()??
1393 * Remove the set_fs and watch the fireworks:-) --okir
1396 oldfs = get_fs(); set_fs(KERNEL_DS);
1397 err = inode->i_op->readlink(dentry, buf, *lenp);
1408 err = nfserrno(err);
1413 * Create a symlink and look up its inode
1414 * N.B. After this call _both_ fhp and resfhp need an fh_put
1417 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1418 char *fname, int flen,
1419 char *path, int plen,
1420 struct svc_fh *resfhp,
1423 struct dentry *dentry, *dnew;
1431 if (isdotent(fname, flen))
1434 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1438 dentry = fhp->fh_dentry;
1439 dnew = lookup_one_len(fname, dentry, flen);
1440 err = PTR_ERR(dnew);
1445 /* Only the MODE ATTRibute is even vaguely meaningful */
1446 if (iap && (iap->ia_valid & ATTR_MODE))
1447 mode = iap->ia_mode & S_IALLUGO;
1449 if (unlikely(path[plen] != 0)) {
1450 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1451 if (path_alloced == NULL)
1454 strncpy(path_alloced, path, plen);
1455 path_alloced[plen] = 0;
1456 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1457 kfree(path_alloced);
1460 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1463 if (EX_ISSYNC(fhp->fh_export))
1464 nfsd_sync_dir(dentry);
1466 err = nfserrno(err);
1469 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1471 if (err==0) err = cerr;
1476 err = nfserrno(err);
1482 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1485 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1486 char *name, int len, struct svc_fh *tfhp)
1488 struct dentry *ddir, *dnew, *dold;
1489 struct inode *dirp, *dest;
1492 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1495 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1503 if (isdotent(name, len))
1507 ddir = ffhp->fh_dentry;
1508 dirp = ddir->d_inode;
1510 dnew = lookup_one_len(name, ddir, len);
1511 err = PTR_ERR(dnew);
1515 dold = tfhp->fh_dentry;
1516 dest = dold->d_inode;
1518 err = vfs_link(dold, dirp, dnew);
1520 if (EX_ISSYNC(ffhp->fh_export)) {
1521 nfsd_sync_dir(ddir);
1522 write_inode_now(dest, 1);
1525 if (err == -EXDEV && rqstp->rq_vers == 2)
1528 err = nfserrno(err);
1537 err = nfserrno(err);
1543 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1546 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1547 struct svc_fh *tfhp, char *tname, int tlen)
1549 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1550 struct inode *fdir, *tdir;
1553 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1556 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1560 fdentry = ffhp->fh_dentry;
1561 fdir = fdentry->d_inode;
1563 tdentry = tfhp->fh_dentry;
1564 tdir = tdentry->d_inode;
1566 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1567 if (fdir->i_sb != tdir->i_sb)
1571 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1574 /* cannot use fh_lock as we need deadlock protective ordering
1575 * so do it by hand */
1576 trap = lock_rename(tdentry, fdentry);
1577 ffhp->fh_locked = tfhp->fh_locked = 1;
1581 odentry = lookup_one_len(fname, fdentry, flen);
1582 err = PTR_ERR(odentry);
1583 if (IS_ERR(odentry))
1587 if (!odentry->d_inode)
1590 if (odentry == trap)
1593 ndentry = lookup_one_len(tname, tdentry, tlen);
1594 err = PTR_ERR(ndentry);
1595 if (IS_ERR(ndentry))
1598 if (ndentry == trap)
1602 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1603 ((atomic_read(&odentry->d_count) > 1)
1604 || (atomic_read(&ndentry->d_count) > 1))) {
1608 err = vfs_rename(fdir, odentry, tdir, ndentry);
1609 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1610 nfsd_sync_dir(tdentry);
1611 nfsd_sync_dir(fdentry);
1620 err = nfserrno(err);
1622 /* we cannot reply on fh_unlock on the two filehandles,
1623 * as that would do the wrong thing if the two directories
1624 * were the same, so again we do it by hand
1626 fill_post_wcc(ffhp);
1627 fill_post_wcc(tfhp);
1628 unlock_rename(tdentry, fdentry);
1629 ffhp->fh_locked = tfhp->fh_locked = 0;
1636 * Unlink a file or directory
1637 * N.B. After this call fhp needs an fh_put
1640 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1641 char *fname, int flen)
1643 struct dentry *dentry, *rdentry;
1648 if (!flen || isdotent(fname, flen))
1650 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1655 dentry = fhp->fh_dentry;
1656 dirp = dentry->d_inode;
1658 rdentry = lookup_one_len(fname, dentry, flen);
1659 err = PTR_ERR(rdentry);
1660 if (IS_ERR(rdentry))
1663 if (!rdentry->d_inode) {
1670 type = rdentry->d_inode->i_mode & S_IFMT;
1672 if (type != S_IFDIR) { /* It's UNLINK */
1674 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1675 (atomic_read(&rdentry->d_count) > 1)) {
1679 err = vfs_unlink(dirp, rdentry);
1680 } else { /* It's RMDIR */
1681 err = vfs_rmdir(dirp, rdentry);
1688 if (EX_ISSYNC(fhp->fh_export))
1689 nfsd_sync_dir(dentry);
1695 err = nfserrno(err);
1700 * Read entries from a directory.
1701 * The NFSv3/4 verifier we ignore for now.
1704 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1705 struct readdir_cd *cdp, encode_dent_fn func)
1709 loff_t offset = *offsetp;
1711 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1715 offset = vfs_llseek(file, offset, 0);
1717 err = nfserrno((int)offset);
1722 * Read the directory entries. This silly loop is necessary because
1723 * readdir() is not guaranteed to fill up the entire buffer, but
1724 * may choose to do less.
1728 cdp->err = nfserr_eof; /* will be cleared on successful read */
1729 err = vfs_readdir(file, (filldir_t) func, cdp);
1730 } while (err >=0 && cdp->err == nfs_ok);
1732 err = nfserrno(err);
1735 *offsetp = vfs_llseek(file, 0, 1);
1737 if (err == nfserr_eof || err == nfserr_toosmall)
1738 err = nfs_ok; /* can still be found in ->err */
1746 * Get file system stats
1747 * N.B. After this call fhp needs an fh_put
1750 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1752 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1753 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1759 * Check for a user's access permissions to this inode.
1762 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1764 struct inode *inode = dentry->d_inode;
1770 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1772 (acc & MAY_READ)? " read" : "",
1773 (acc & MAY_WRITE)? " write" : "",
1774 (acc & MAY_EXEC)? " exec" : "",
1775 (acc & MAY_SATTR)? " sattr" : "",
1776 (acc & MAY_TRUNC)? " trunc" : "",
1777 (acc & MAY_LOCK)? " lock" : "",
1778 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1780 IS_IMMUTABLE(inode)? " immut" : "",
1781 IS_APPEND(inode)? " append" : "",
1782 IS_RDONLY(inode)? " ro" : "");
1783 dprintk(" owner %d/%d user %d/%d\n",
1784 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1787 /* Normally we reject any write/sattr etc access on a read-only file
1788 * system. But if it is IRIX doing check on write-access for a
1789 * device special file, we ignore rofs.
1791 if (!(acc & MAY_LOCAL_ACCESS))
1792 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1793 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1795 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1798 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1801 if (acc & MAY_LOCK) {
1802 /* If we cannot rely on authentication in NLM requests,
1803 * just allow locks, otherwise require read permission, or
1806 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1809 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1812 * The file owner always gets access permission for accesses that
1813 * would normally be checked at open time. This is to make
1814 * file access work even when the client has done a fchmod(fd, 0).
1816 * However, `cp foo bar' should fail nevertheless when bar is
1817 * readonly. A sensible way to do this might be to reject all
1818 * attempts to truncate a read-only file, because a creat() call
1819 * always implies file truncation.
1820 * ... but this isn't really fair. A process may reasonably call
1821 * ftruncate on an open file descriptor on a file with perm 000.
1822 * We must trust the client to do permission checking - using "ACCESS"
1825 if ((acc & MAY_OWNER_OVERRIDE) &&
1826 inode->i_uid == current->fsuid)
1829 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1831 /* Allow read access to binaries even when mode 111 */
1832 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1833 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1834 err = permission(inode, MAY_EXEC, NULL);
1836 return err? nfserrno(err) : 0;
1840 nfsd_racache_shutdown(void)
1844 dprintk("nfsd: freeing readahead buffers.\n");
1846 raparm_cache = raparml = NULL;
1849 * Initialize readahead param cache
1852 nfsd_racache_init(int cache_size)
1858 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1860 if (raparml != NULL) {
1861 dprintk("nfsd: allocating %d readahead buffers.\n",
1863 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1864 for (i = 0; i < cache_size - 1; i++) {
1865 raparml[i].p_next = raparml + i + 1;
1867 raparm_cache = raparml;
1870 "nfsd: Could not allocate memory read-ahead cache.\n");
1873 nfsdstats.ra_size = cache_size;
1877 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1879 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1881 struct inode *inode = fhp->fh_dentry->d_inode;
1885 struct posix_acl *acl;
1887 if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr)
1888 return ERR_PTR(-EOPNOTSUPP);
1890 case ACL_TYPE_ACCESS:
1891 name = POSIX_ACL_XATTR_ACCESS;
1893 case ACL_TYPE_DEFAULT:
1894 name = POSIX_ACL_XATTR_DEFAULT;
1897 return ERR_PTR(-EOPNOTSUPP);
1900 size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0);
1903 acl = ERR_PTR(size);
1905 } else if (size > 0) {
1906 value = kmalloc(size, GFP_KERNEL);
1908 acl = ERR_PTR(-ENOMEM);
1911 size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size);
1913 acl = ERR_PTR(size);
1917 acl = posix_acl_from_xattr(value, size);
1925 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1927 struct inode *inode = fhp->fh_dentry->d_inode;
1933 if (!IS_POSIXACL(inode) || !inode->i_op ||
1934 !inode->i_op->setxattr || !inode->i_op->removexattr)
1937 case ACL_TYPE_ACCESS:
1938 name = POSIX_ACL_XATTR_ACCESS;
1940 case ACL_TYPE_DEFAULT:
1941 name = POSIX_ACL_XATTR_DEFAULT;
1947 if (acl && acl->a_count) {
1948 size = posix_acl_xattr_size(acl->a_count);
1949 value = kmalloc(size, GFP_KERNEL);
1952 size = posix_acl_to_xattr(acl, value, size);
1960 if (!fhp->fh_locked)
1961 fh_lock(fhp); /* unlocking is done automatically */
1963 error = inode->i_op->setxattr(fhp->fh_dentry, name,
1966 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1969 error = inode->i_op->removexattr(fhp->fh_dentry, name);
1970 if (error == -ENODATA)
1979 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */