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/dnotify.h>
49 #include <linux/xattr_acl.h>
50 #include <linux/posix_acl.h>
52 #include <linux/posix_acl_xattr.h>
53 #include <linux/xattr_acl.h>
54 #include <linux/xattr.h>
55 #include <linux/nfs4.h>
56 #include <linux/nfs4_acl.h>
57 #include <linux/nfsd_idmap.h>
58 #include <linux/security.h>
59 #endif /* CONFIG_NFSD_V4 */
61 #include <asm/uaccess.h>
63 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
67 /* We must ignore files (but only files) which might have mandatory
68 * locks on them because there is no way to know if the accesser has
71 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
74 * This is a cache of readahead params that help us choose the proper
75 * readahead strategy. Initially, we set all readahead parameters to 0
76 * and let the VFS handle things.
77 * If you increase the number of cached files very much, you'll need to
78 * add a hash table here.
81 struct raparms *p_next;
86 struct file_ra_state p_ra;
89 static struct raparms * raparml;
90 static struct raparms * raparm_cache;
93 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
95 * Returns -EAGAIN leaving *dpp and *expp unchanged,
96 * or nfs_ok having possibly changed *dpp and *expp
99 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
100 struct svc_export **expp)
102 struct svc_export *exp = *expp, *exp2 = NULL;
103 struct dentry *dentry = *dpp;
104 struct vfsmount *mnt = mntget(exp->ex_mnt);
105 struct dentry *mounts = dget(dentry);
108 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
110 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
117 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
118 /* successfully crossed mount point */
124 if (exp2) exp_put(exp2);
133 * Look up one component of a pathname.
134 * N.B. After this call _both_ fhp and resfh need an fh_put
136 * If the lookup would cross a mountpoint, and the mounted filesystem
137 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
138 * accepted as it stands and the mounted directory is
139 * returned. Otherwise the covered directory is returned.
140 * NOTE: this mountpoint crossing is not supported properly by all
141 * clients and is explicitly disallowed for NFSv3
142 * NeilBrown <neilb@cse.unsw.edu.au>
145 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
146 int len, struct svc_fh *resfh)
148 struct svc_export *exp;
149 struct dentry *dparent;
150 struct dentry *dentry;
153 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
155 /* Obtain dentry and export. */
156 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
160 dparent = fhp->fh_dentry;
161 exp = fhp->fh_export;
166 /* Lookup the name, but don't follow links */
167 if (isdotent(name, len)) {
169 dentry = dget(dparent);
170 else if (dparent != exp->ex_dentry) {
171 dentry = dget_parent(dparent);
172 } else if (!EX_NOHIDE(exp))
173 dentry = dget(dparent); /* .. == . just like at / */
175 /* checking mountpoint crossing is very different when stepping up */
176 struct svc_export *exp2 = NULL;
178 struct vfsmount *mnt = mntget(exp->ex_mnt);
179 dentry = dget(dparent);
180 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
182 dp = dget_parent(dentry);
186 exp2 = exp_parent(exp->ex_client, mnt, dentry,
196 dentry = dget(dparent);
205 dentry = lookup_one_len(name, dparent, len);
206 err = PTR_ERR(dentry);
210 * check if we have crossed a mount point ...
212 if (d_mountpoint(dentry)) {
213 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
220 * Note: we compose the file handle now, but as the
221 * dentry may be negative, it may need to be updated.
223 err = fh_compose(resfh, exp, dentry, fhp);
224 if (!err && !dentry->d_inode)
237 * Set various file attributes.
238 * N.B. After this call fhp needs an fh_put
241 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
242 int check_guard, time_t guardtime)
244 struct dentry *dentry;
246 int accmode = MAY_SATTR;
252 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
253 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
254 if (iap->ia_valid & ATTR_SIZE)
258 err = fh_verify(rqstp, fhp, ftype, accmode);
259 if (err || !iap->ia_valid)
262 dentry = fhp->fh_dentry;
263 inode = dentry->d_inode;
265 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
266 * which only requires access, and "set-[ac]time-to-X" which
267 * requires ownership.
268 * So if it looks like it might be "set both to the same time which
269 * is close to now", and if inode_change_ok fails, then we
270 * convert to "set to now" instead of "set to explicit time"
272 * We only call inode_change_ok as the last test as technically
273 * it is not an interface that we should be using. It is only
274 * valid if the filesystem does not define it's own i_op->setattr.
276 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
277 #define MAX_TOUCH_TIME_ERROR (30*60)
278 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
279 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
281 /* Looks probable. Now just make sure time is in the right ballpark.
282 * Solaris, at least, doesn't seem to care what the time request is.
283 * We require it be within 30 minutes of now.
285 time_t delta = iap->ia_atime.tv_sec - get_seconds();
286 if (delta<0) delta = -delta;
287 if (delta < MAX_TOUCH_TIME_ERROR &&
288 inode_change_ok(inode, iap) != 0) {
289 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
290 * this will cause notify_change to set these times to "now"
292 iap->ia_valid &= ~BOTH_TIME_SET;
296 /* The size case is special. It changes the file as well as the attributes. */
297 if (iap->ia_valid & ATTR_SIZE) {
298 if (iap->ia_size < inode->i_size) {
299 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
305 * If we are changing the size of the file, then
306 * we need to break all leases.
308 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
309 if (err == -EWOULDBLOCK)
311 if (err) /* ENOMEM or EWOULDBLOCK */
314 err = get_write_access(inode);
319 err = locks_verify_truncate(inode, NULL, iap->ia_size);
321 put_write_access(inode);
327 imode = inode->i_mode;
328 if (iap->ia_valid & ATTR_MODE) {
329 iap->ia_mode &= S_IALLUGO;
330 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
333 /* Revoke setuid/setgid bit on chown/chgrp */
334 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
335 iap->ia_valid |= ATTR_KILL_SUID;
336 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
337 iap->ia_valid |= ATTR_KILL_SGID;
339 /* Change the attributes. */
341 iap->ia_valid |= ATTR_CTIME;
343 err = nfserr_notsync;
344 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
346 err = notify_change(dentry, iap);
351 put_write_access(inode);
353 if (EX_ISSYNC(fhp->fh_export))
354 write_inode_now(inode, 1);
363 #if defined(CONFIG_NFSD_V4)
366 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
372 struct inode *inode = dentry->d_inode;
374 buflen = posix_acl_xattr_size(pacl->a_count);
375 buf = kmalloc(buflen, GFP_KERNEL);
380 len = posix_acl_to_xattr(pacl, buf, buflen);
387 if (inode->i_op && inode->i_op->setxattr) {
389 security_inode_setxattr(dentry, key, buf, len, 0);
390 error = inode->i_op->setxattr(dentry, key, buf, len, 0);
392 security_inode_post_setxattr(dentry, key, buf, len, 0);
401 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
402 struct nfs4_acl *acl)
405 struct dentry *dentry;
407 struct posix_acl *pacl = NULL, *dpacl = NULL;
408 unsigned int flags = 0;
411 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
415 dentry = fhp->fh_dentry;
416 inode = dentry->d_inode;
417 if (S_ISDIR(inode->i_mode))
418 flags = NFS4_ACL_DIR;
420 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
421 if (error == -EINVAL) {
422 error = nfserr_attrnotsupp;
424 } else if (error < 0)
428 error = set_nfsv4_acl_one(dentry, pacl, XATTR_NAME_ACL_ACCESS);
434 error = set_nfsv4_acl_one(dentry, dpacl, XATTR_NAME_ACL_DEFAULT);
442 posix_acl_release(pacl);
443 posix_acl_release(dpacl);
446 error = nfserrno(error);
450 static struct posix_acl *
451 _get_posix_acl(struct dentry *dentry, char *key)
453 struct inode *inode = dentry->d_inode;
455 int buflen, error = 0;
456 struct posix_acl *pacl = NULL;
459 if (inode->i_op == NULL)
461 if (inode->i_op->getxattr == NULL)
464 error = security_inode_getxattr(dentry, key);
468 buflen = inode->i_op->getxattr(dentry, key, NULL, 0);
470 error = buflen < 0 ? buflen : -ENODATA;
474 buf = kmalloc(buflen, GFP_KERNEL);
480 error = inode->i_op->getxattr(dentry, key, buf, buflen);
484 pacl = posix_acl_from_xattr(buf, buflen);
489 pacl = ERR_PTR(error);
494 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
496 struct inode *inode = dentry->d_inode;
498 struct posix_acl *pacl = NULL, *dpacl = NULL;
499 unsigned int flags = 0;
501 pacl = _get_posix_acl(dentry, XATTR_NAME_ACL_ACCESS);
502 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
503 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
505 error = PTR_ERR(pacl);
510 if (S_ISDIR(inode->i_mode)) {
511 dpacl = _get_posix_acl(dentry, XATTR_NAME_ACL_DEFAULT);
512 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
514 else if (IS_ERR(dpacl)) {
515 error = PTR_ERR(dpacl);
519 flags = NFS4_ACL_DIR;
522 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
524 error = PTR_ERR(*acl);
528 posix_acl_release(pacl);
529 posix_acl_release(dpacl);
533 #endif /* defined(CONFIG_NFS_V4) */
535 #ifdef CONFIG_NFSD_V3
537 * Check server access rights to a file system object
543 static struct accessmap nfs3_regaccess[] = {
544 { NFS3_ACCESS_READ, MAY_READ },
545 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
546 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
547 { NFS3_ACCESS_EXTEND, MAY_WRITE },
552 static struct accessmap nfs3_diraccess[] = {
553 { NFS3_ACCESS_READ, MAY_READ },
554 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
555 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
556 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
557 { NFS3_ACCESS_DELETE, MAY_REMOVE },
562 static struct accessmap nfs3_anyaccess[] = {
563 /* Some clients - Solaris 2.6 at least, make an access call
564 * to the server to check for access for things like /dev/null
565 * (which really, the server doesn't care about). So
566 * We provide simple access checking for them, looking
567 * mainly at mode bits, and we make sure to ignore read-only
570 { NFS3_ACCESS_READ, MAY_READ },
571 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
572 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
573 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
579 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
581 struct accessmap *map;
582 struct svc_export *export;
583 struct dentry *dentry;
584 u32 query, result = 0, sresult = 0;
587 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
591 export = fhp->fh_export;
592 dentry = fhp->fh_dentry;
594 if (S_ISREG(dentry->d_inode->i_mode))
595 map = nfs3_regaccess;
596 else if (S_ISDIR(dentry->d_inode->i_mode))
597 map = nfs3_diraccess;
599 map = nfs3_anyaccess;
603 for (; map->access; map++) {
604 if (map->access & query) {
607 sresult |= map->access;
609 err2 = nfsd_permission(export, dentry, map->how);
612 result |= map->access;
615 /* the following error codes just mean the access was not allowed,
616 * rather than an error occurred */
620 /* simply don't "or" in the access bit. */
630 *supported = sresult;
635 #endif /* CONFIG_NFSD_V3 */
640 * Open an existing file or directory.
641 * The access argument indicates the type of open (read/write/lock)
642 * N.B. After this call fhp needs an fh_put
645 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
646 int access, struct file **filp)
648 struct dentry *dentry;
650 int flags = O_RDONLY|O_LARGEFILE, err;
653 * If we get here, then the client has already done an "open",
654 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
655 * in case a chmod has now revoked permission.
657 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
661 dentry = fhp->fh_dentry;
662 inode = dentry->d_inode;
664 /* Disallow write access to files with the append-only bit set
665 * or any access when mandatory locking enabled
668 if (IS_APPEND(inode) && (access & MAY_WRITE))
670 if (IS_ISMNDLK(inode))
677 * Check to see if there are any leases on this file.
678 * This may block while leases are broken.
680 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
681 if (err == -EWOULDBLOCK)
683 if (err) /* NOMEM or WOULDBLOCK */
686 if (access & MAY_WRITE) {
687 flags = O_WRONLY|O_LARGEFILE;
691 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
693 err = PTR_ERR(*filp);
705 nfsd_close(struct file *filp)
712 * As this calls fsync (not fdatasync) there is no need for a write_inode
715 static inline void nfsd_dosync(struct file *filp, struct dentry *dp,
716 struct file_operations *fop)
718 struct inode *inode = dp->d_inode;
719 int (*fsync) (struct file *, struct dentry *, int);
721 filemap_fdatawrite(inode->i_mapping);
722 if (fop && (fsync = fop->fsync))
724 filemap_fdatawait(inode->i_mapping);
729 nfsd_sync(struct file *filp)
731 struct inode *inode = filp->f_dentry->d_inode;
732 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
734 nfsd_dosync(filp, filp->f_dentry, filp->f_op);
739 nfsd_sync_dir(struct dentry *dp)
741 nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
745 * Obtain the readahead parameters for the file
746 * specified by (dev, ino).
748 static DEFINE_SPINLOCK(ra_lock);
750 static inline struct raparms *
751 nfsd_get_raparms(dev_t dev, ino_t ino)
753 struct raparms *ra, **rap, **frap = NULL;
757 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
758 if (ra->p_ino == ino && ra->p_dev == dev)
761 if (ra->p_count == 0)
764 depth = nfsdstats.ra_size*11/10;
766 spin_unlock(&ra_lock);
775 if (rap != &raparm_cache) {
777 ra->p_next = raparm_cache;
781 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
782 spin_unlock(&ra_lock);
787 * Grab and keep cached pages assosiated with a file in the svc_rqst
788 * so that they can be passed to the netowork sendmsg/sendpage routines
789 * directrly. They will be released after the sending has completed.
792 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
794 unsigned long count = desc->count;
795 struct svc_rqst *rqstp = desc->arg.data;
800 if (rqstp->rq_res.page_len == 0) {
802 rqstp->rq_respages[rqstp->rq_resused++] = page;
803 rqstp->rq_res.page_base = offset;
804 rqstp->rq_res.page_len = size;
805 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
807 rqstp->rq_respages[rqstp->rq_resused++] = page;
808 rqstp->rq_res.page_len += size;
810 rqstp->rq_res.page_len += size;
813 desc->count = count - size;
814 desc->written += size;
819 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
820 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
828 inode = file->f_dentry->d_inode;
830 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
831 (!lock_may_read(inode, offset, *count)))
835 /* Get readahead parameters */
836 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
839 file->f_ra = ra->p_ra;
841 if (file->f_op->sendfile) {
842 svc_pushback_unused_pages(rqstp);
843 err = file->f_op->sendfile(file, &offset, *count,
844 nfsd_read_actor, rqstp);
848 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
852 /* Write back readahead params */
855 ra->p_ra = file->f_ra;
858 spin_unlock(&ra_lock);
862 nfsdstats.io_read += err;
865 dnotify_parent(file->f_dentry, DN_ACCESS);
873 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
874 loff_t offset, struct kvec *vec, int vlen,
875 unsigned long cnt, int *stablep)
877 struct svc_export *exp;
878 struct dentry *dentry;
882 int stable = *stablep;
887 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
888 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
892 dentry = file->f_dentry;
893 inode = dentry->d_inode;
894 exp = fhp->fh_export;
897 * Request sync writes if
898 * - the sync export option has been set, or
899 * - the client requested O_SYNC behavior (NFSv3 feature).
900 * - The file system doesn't support fsync().
901 * When gathered writes have been configured for this volume,
902 * flushing the data to disk is handled separately below.
905 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
907 *stablep = 2; /* FILE_SYNC */
912 if (stable && !EX_WGATHER(exp))
913 file->f_flags |= O_SYNC;
915 /* Write the data. */
916 oldfs = get_fs(); set_fs(KERNEL_DS);
917 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
920 nfsdstats.io_write += cnt;
921 dnotify_parent(file->f_dentry, DN_MODIFY);
924 /* clear setuid/setgid flag after write */
925 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID))) {
927 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
930 notify_change(dentry, &ia);
934 if (err >= 0 && stable) {
935 static ino_t last_ino;
936 static dev_t last_dev;
939 * Gathered writes: If another process is currently
940 * writing to the file, there's a high chance
941 * this is another nfsd (triggered by a bulk write
942 * from a client's biod). Rather than syncing the
943 * file with each write request, we sleep for 10 msec.
945 * I don't know if this roughly approximates
946 * C. Juszak's idea of gathered writes, but it's a
947 * nice and simple solution (IMHO), and it seems to
950 if (EX_WGATHER(exp)) {
951 if (atomic_read(&inode->i_writecount) > 1
952 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
953 dprintk("nfsd: write defer %d\n", current->pid);
955 dprintk("nfsd: write resume %d\n", current->pid);
958 if (inode->i_state & I_DIRTY) {
959 dprintk("nfsd: write sync %d\n", current->pid);
963 wake_up(&inode->i_wait);
966 last_ino = inode->i_ino;
967 last_dev = inode->i_sb->s_dev;
970 dprintk("nfsd: write complete err=%d\n", err);
980 * Read data from a file. count must contain the requested read count
981 * on entry. On return, *count contains the number of bytes actually read.
982 * N.B. After this call fhp needs an fh_put
985 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
986 loff_t offset, struct kvec *vec, int vlen,
987 unsigned long *count)
992 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
993 MAY_READ|MAY_OWNER_OVERRIDE);
996 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
998 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1001 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1009 * Write data to a file.
1010 * The stable flag requests synchronous writes.
1011 * N.B. After this call fhp needs an fh_put
1014 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1015 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1021 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1022 MAY_WRITE|MAY_OWNER_OVERRIDE);
1025 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1028 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1033 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1041 #ifdef CONFIG_NFSD_V3
1043 * Commit all pending writes to stable storage.
1044 * Strictly speaking, we could sync just the indicated file region here,
1045 * but there's currently no way we can ask the VFS to do so.
1047 * Unfortunately we cannot lock the file to make sure we return full WCC
1048 * data to the client, as locking happens lower down in the filesystem.
1051 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1052 loff_t offset, unsigned long count)
1057 if ((u64)count > ~(u64)offset)
1058 return nfserr_inval;
1060 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1062 if (EX_ISSYNC(fhp->fh_export)) {
1063 if (file->f_op && file->f_op->fsync) {
1066 err = nfserr_notsupp;
1073 #endif /* CONFIG_NFSD_V3 */
1076 * Create a file (regular, directory, device, fifo); UNIX sockets
1077 * not yet implemented.
1078 * If the response fh has been verified, the parent directory should
1079 * already be locked. Note that the parent directory is left locked.
1081 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1084 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1085 char *fname, int flen, struct iattr *iap,
1086 int type, dev_t rdev, struct svc_fh *resfhp)
1088 struct dentry *dentry, *dchild = NULL;
1096 if (isdotent(fname, flen))
1099 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1103 dentry = fhp->fh_dentry;
1104 dirp = dentry->d_inode;
1106 err = nfserr_notdir;
1107 if(!dirp->i_op || !dirp->i_op->lookup)
1110 * Check whether the response file handle has been verified yet.
1111 * If it has, the parent directory should already be locked.
1113 if (!resfhp->fh_dentry) {
1114 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1116 dchild = lookup_one_len(fname, dentry, flen);
1117 err = PTR_ERR(dchild);
1120 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1124 /* called from nfsd_proc_create */
1125 dchild = dget(resfhp->fh_dentry);
1126 if (!fhp->fh_locked) {
1127 /* not actually possible */
1129 "nfsd_create: parent %s/%s not locked!\n",
1130 dentry->d_parent->d_name.name,
1131 dentry->d_name.name);
1137 * Make sure the child dentry is still negative ...
1140 if (dchild->d_inode) {
1141 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1142 dentry->d_name.name, dchild->d_name.name);
1146 if (!(iap->ia_valid & ATTR_MODE))
1148 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1151 * Get the dir op function pointer.
1156 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1159 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1165 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1168 printk("nfsd: bad file type %o in nfsd_create\n", type);
1174 if (EX_ISSYNC(fhp->fh_export)) {
1175 nfsd_sync_dir(dentry);
1176 write_inode_now(dchild->d_inode, 1);
1180 /* Set file attributes. Mode has already been set and
1181 * setting uid/gid works only for root. Irix appears to
1182 * send along the gid when it tries to implement setgid
1183 * directories via NFS.
1186 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0)
1187 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1189 * Update the file handle to get the new inode info.
1192 err = fh_update(resfhp);
1194 if (dchild && !IS_ERR(dchild))
1199 err = nfserrno(err);
1203 #ifdef CONFIG_NFSD_V3
1205 * NFSv3 version of nfsd_create
1208 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1209 char *fname, int flen, struct iattr *iap,
1210 struct svc_fh *resfhp, int createmode, u32 *verifier,
1213 struct dentry *dentry, *dchild = NULL;
1216 __u32 v_mtime=0, v_atime=0;
1223 if (isdotent(fname, flen))
1225 if (!(iap->ia_valid & ATTR_MODE))
1227 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1231 dentry = fhp->fh_dentry;
1232 dirp = dentry->d_inode;
1234 /* Get all the sanity checks out of the way before
1235 * we lock the parent. */
1236 err = nfserr_notdir;
1237 if(!dirp->i_op || !dirp->i_op->lookup)
1242 * Compose the response file handle.
1244 dchild = lookup_one_len(fname, dentry, flen);
1245 err = PTR_ERR(dchild);
1249 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1253 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1254 /* while the verifier would fit in mtime+atime,
1255 * solaris7 gets confused (bugid 4218508) if these have
1256 * the high bit set, so we use the mode as well
1258 v_mtime = verifier[0]&0x7fffffff;
1259 v_atime = verifier[1]&0x7fffffff;
1261 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1262 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1266 if (dchild->d_inode) {
1269 switch (createmode) {
1270 case NFS3_CREATE_UNCHECKED:
1271 if (! S_ISREG(dchild->d_inode->i_mode))
1274 /* in nfsv4, we need to treat this case a little
1275 * differently. we don't want to truncate the
1276 * file now; this would be wrong if the OPEN
1277 * fails for some other reason. furthermore,
1278 * if the size is nonzero, we should ignore it
1279 * according to spec!
1281 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1284 iap->ia_valid &= ATTR_SIZE;
1288 case NFS3_CREATE_EXCLUSIVE:
1289 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1290 && dchild->d_inode->i_atime.tv_sec == v_atime
1291 && dchild->d_inode->i_mode == v_mode
1292 && dchild->d_inode->i_size == 0 )
1295 case NFS3_CREATE_GUARDED:
1301 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1305 if (EX_ISSYNC(fhp->fh_export)) {
1306 nfsd_sync_dir(dentry);
1307 /* setattr will sync the child (or not) */
1311 * Update the filehandle to get the new inode info.
1313 err = fh_update(resfhp);
1317 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1318 /* Cram the verifier into atime/mtime/mode */
1319 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1320 | ATTR_MTIME_SET|ATTR_ATIME_SET
1322 /* XXX someone who knows this better please fix it for nsec */
1323 iap->ia_mtime.tv_sec = v_mtime;
1324 iap->ia_atime.tv_sec = v_atime;
1325 iap->ia_mtime.tv_nsec = 0;
1326 iap->ia_atime.tv_nsec = 0;
1327 iap->ia_mode = v_mode;
1330 /* Set file attributes.
1331 * Mode has already been set but we might need to reset it
1332 * for CREATE_EXCLUSIVE
1333 * Irix appears to send along the gid when it tries to
1334 * implement setgid directories via NFS. Clear out all that cruft.
1337 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0)
1338 err = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1342 if (dchild && !IS_ERR(dchild))
1347 err = nfserrno(err);
1350 #endif /* CONFIG_NFSD_V3 */
1353 * Read a symlink. On entry, *lenp must contain the maximum path length that
1354 * fits into the buffer. On return, it contains the true length.
1355 * N.B. After this call fhp needs an fh_put
1358 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1360 struct dentry *dentry;
1361 struct inode *inode;
1365 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1369 dentry = fhp->fh_dentry;
1370 inode = dentry->d_inode;
1373 if (!inode->i_op || !inode->i_op->readlink)
1376 touch_atime(fhp->fh_export->ex_mnt, dentry);
1377 /* N.B. Why does this call need a get_fs()??
1378 * Remove the set_fs and watch the fireworks:-) --okir
1381 oldfs = get_fs(); set_fs(KERNEL_DS);
1382 err = inode->i_op->readlink(dentry, buf, *lenp);
1393 err = nfserrno(err);
1398 * Create a symlink and look up its inode
1399 * N.B. After this call _both_ fhp and resfhp need an fh_put
1402 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1403 char *fname, int flen,
1404 char *path, int plen,
1405 struct svc_fh *resfhp,
1408 struct dentry *dentry, *dnew;
1416 if (isdotent(fname, flen))
1419 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1423 dentry = fhp->fh_dentry;
1424 dnew = lookup_one_len(fname, dentry, flen);
1425 err = PTR_ERR(dnew);
1430 /* Only the MODE ATTRibute is even vaguely meaningful */
1431 if (iap && (iap->ia_valid & ATTR_MODE))
1432 mode = iap->ia_mode & S_IALLUGO;
1434 if (unlikely(path[plen] != 0)) {
1435 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1436 if (path_alloced == NULL)
1439 strncpy(path_alloced, path, plen);
1440 path_alloced[plen] = 0;
1441 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1442 kfree(path_alloced);
1445 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1448 if (EX_ISSYNC(fhp->fh_export))
1449 nfsd_sync_dir(dentry);
1451 err = nfserrno(err);
1454 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1456 if (err==0) err = cerr;
1461 err = nfserrno(err);
1467 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1470 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1471 char *name, int len, struct svc_fh *tfhp)
1473 struct dentry *ddir, *dnew, *dold;
1474 struct inode *dirp, *dest;
1477 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1480 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1488 if (isdotent(name, len))
1492 ddir = ffhp->fh_dentry;
1493 dirp = ddir->d_inode;
1495 dnew = lookup_one_len(name, ddir, len);
1496 err = PTR_ERR(dnew);
1500 dold = tfhp->fh_dentry;
1501 dest = dold->d_inode;
1503 err = vfs_link(dold, dirp, dnew);
1505 if (EX_ISSYNC(ffhp->fh_export)) {
1506 nfsd_sync_dir(ddir);
1507 write_inode_now(dest, 1);
1510 if (err == -EXDEV && rqstp->rq_vers == 2)
1513 err = nfserrno(err);
1522 err = nfserrno(err);
1528 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1531 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1532 struct svc_fh *tfhp, char *tname, int tlen)
1534 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1535 struct inode *fdir, *tdir;
1538 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1541 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1545 fdentry = ffhp->fh_dentry;
1546 fdir = fdentry->d_inode;
1548 tdentry = tfhp->fh_dentry;
1549 tdir = tdentry->d_inode;
1551 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1552 if (fdir->i_sb != tdir->i_sb)
1556 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1559 /* cannot use fh_lock as we need deadlock protective ordering
1560 * so do it by hand */
1561 trap = lock_rename(tdentry, fdentry);
1562 ffhp->fh_locked = tfhp->fh_locked = 1;
1566 odentry = lookup_one_len(fname, fdentry, flen);
1567 err = PTR_ERR(odentry);
1568 if (IS_ERR(odentry))
1572 if (!odentry->d_inode)
1575 if (odentry == trap)
1578 ndentry = lookup_one_len(tname, tdentry, tlen);
1579 err = PTR_ERR(ndentry);
1580 if (IS_ERR(ndentry))
1583 if (ndentry == trap)
1587 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1588 ((atomic_read(&odentry->d_count) > 1)
1589 || (atomic_read(&ndentry->d_count) > 1))) {
1593 err = vfs_rename(fdir, odentry, tdir, ndentry);
1594 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1595 nfsd_sync_dir(tdentry);
1596 nfsd_sync_dir(fdentry);
1605 err = nfserrno(err);
1607 /* we cannot reply on fh_unlock on the two filehandles,
1608 * as that would do the wrong thing if the two directories
1609 * were the same, so again we do it by hand
1611 fill_post_wcc(ffhp);
1612 fill_post_wcc(tfhp);
1613 unlock_rename(tdentry, fdentry);
1614 ffhp->fh_locked = tfhp->fh_locked = 0;
1621 * Unlink a file or directory
1622 * N.B. After this call fhp needs an fh_put
1625 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1626 char *fname, int flen)
1628 struct dentry *dentry, *rdentry;
1633 if (!flen || isdotent(fname, flen))
1635 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1640 dentry = fhp->fh_dentry;
1641 dirp = dentry->d_inode;
1643 rdentry = lookup_one_len(fname, dentry, flen);
1644 err = PTR_ERR(rdentry);
1645 if (IS_ERR(rdentry))
1648 if (!rdentry->d_inode) {
1655 type = rdentry->d_inode->i_mode & S_IFMT;
1657 if (type != S_IFDIR) { /* It's UNLINK */
1659 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1660 (atomic_read(&rdentry->d_count) > 1)) {
1664 err = vfs_unlink(dirp, rdentry);
1665 } else { /* It's RMDIR */
1666 err = vfs_rmdir(dirp, rdentry);
1673 if (EX_ISSYNC(fhp->fh_export))
1674 nfsd_sync_dir(dentry);
1680 err = nfserrno(err);
1685 * Read entries from a directory.
1686 * The NFSv3/4 verifier we ignore for now.
1689 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1690 struct readdir_cd *cdp, encode_dent_fn func)
1694 loff_t offset = *offsetp;
1696 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1700 offset = vfs_llseek(file, offset, 0);
1702 err = nfserrno((int)offset);
1707 * Read the directory entries. This silly loop is necessary because
1708 * readdir() is not guaranteed to fill up the entire buffer, but
1709 * may choose to do less.
1713 cdp->err = nfserr_eof; /* will be cleared on successful read */
1714 err = vfs_readdir(file, (filldir_t) func, cdp);
1715 } while (err >=0 && cdp->err == nfs_ok);
1717 err = nfserrno(err);
1720 *offsetp = vfs_llseek(file, 0, 1);
1722 if (err == nfserr_eof || err == nfserr_toosmall)
1723 err = nfs_ok; /* can still be found in ->err */
1731 * Get file system stats
1732 * N.B. After this call fhp needs an fh_put
1735 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1737 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1738 if (!err && vfs_statfs(fhp->fh_dentry->d_inode->i_sb,stat))
1744 * Check for a user's access permissions to this inode.
1747 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1749 struct inode *inode = dentry->d_inode;
1755 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1757 (acc & MAY_READ)? " read" : "",
1758 (acc & MAY_WRITE)? " write" : "",
1759 (acc & MAY_EXEC)? " exec" : "",
1760 (acc & MAY_SATTR)? " sattr" : "",
1761 (acc & MAY_TRUNC)? " trunc" : "",
1762 (acc & MAY_LOCK)? " lock" : "",
1763 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1765 IS_IMMUTABLE(inode)? " immut" : "",
1766 IS_APPEND(inode)? " append" : "",
1767 IS_RDONLY(inode)? " ro" : "");
1768 dprintk(" owner %d/%d user %d/%d\n",
1769 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1772 /* Normally we reject any write/sattr etc access on a read-only file
1773 * system. But if it is IRIX doing check on write-access for a
1774 * device special file, we ignore rofs.
1776 if (!(acc & MAY_LOCAL_ACCESS))
1777 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1778 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1780 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1783 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1786 if (acc & MAY_LOCK) {
1787 /* If we cannot rely on authentication in NLM requests,
1788 * just allow locks, otherwise require read permission, or
1791 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1794 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1797 * The file owner always gets access permission for accesses that
1798 * would normally be checked at open time. This is to make
1799 * file access work even when the client has done a fchmod(fd, 0).
1801 * However, `cp foo bar' should fail nevertheless when bar is
1802 * readonly. A sensible way to do this might be to reject all
1803 * attempts to truncate a read-only file, because a creat() call
1804 * always implies file truncation.
1805 * ... but this isn't really fair. A process may reasonably call
1806 * ftruncate on an open file descriptor on a file with perm 000.
1807 * We must trust the client to do permission checking - using "ACCESS"
1810 if ((acc & MAY_OWNER_OVERRIDE) &&
1811 inode->i_uid == current->fsuid)
1814 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1816 /* Allow read access to binaries even when mode 111 */
1817 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1818 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1819 err = permission(inode, MAY_EXEC, NULL);
1821 return err? nfserrno(err) : 0;
1825 nfsd_racache_shutdown(void)
1829 dprintk("nfsd: freeing readahead buffers.\n");
1831 raparm_cache = raparml = NULL;
1834 * Initialize readahead param cache
1837 nfsd_racache_init(int cache_size)
1843 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1845 if (raparml != NULL) {
1846 dprintk("nfsd: allocating %d readahead buffers.\n",
1848 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1849 for (i = 0; i < cache_size - 1; i++) {
1850 raparml[i].p_next = raparml + i + 1;
1852 raparm_cache = raparml;
1855 "nfsd: Could not allocate memory read-ahead cache.\n");
1858 nfsdstats.ra_size = cache_size;
1862 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1864 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1866 struct inode *inode = fhp->fh_dentry->d_inode;
1870 struct posix_acl *acl;
1872 if (!IS_POSIXACL(inode) || !inode->i_op || !inode->i_op->getxattr)
1873 return ERR_PTR(-EOPNOTSUPP);
1875 case ACL_TYPE_ACCESS:
1876 name = XATTR_NAME_ACL_ACCESS;
1878 case ACL_TYPE_DEFAULT:
1879 name = XATTR_NAME_ACL_DEFAULT;
1882 return ERR_PTR(-EOPNOTSUPP);
1885 size = inode->i_op->getxattr(fhp->fh_dentry, name, NULL, 0);
1888 acl = ERR_PTR(size);
1890 } else if (size > 0) {
1891 value = kmalloc(size, GFP_KERNEL);
1893 acl = ERR_PTR(-ENOMEM);
1896 size = inode->i_op->getxattr(fhp->fh_dentry, name, value, size);
1898 acl = ERR_PTR(size);
1902 acl = posix_acl_from_xattr(value, size);
1910 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1912 struct inode *inode = fhp->fh_dentry->d_inode;
1918 if (!IS_POSIXACL(inode) || !inode->i_op ||
1919 !inode->i_op->setxattr || !inode->i_op->removexattr)
1922 case ACL_TYPE_ACCESS:
1923 name = XATTR_NAME_ACL_ACCESS;
1925 case ACL_TYPE_DEFAULT:
1926 name = XATTR_NAME_ACL_DEFAULT;
1932 if (acl && acl->a_count) {
1933 size = xattr_acl_size(acl->a_count);
1934 value = kmalloc(size, GFP_KERNEL);
1937 size = posix_acl_to_xattr(acl, value, size);
1945 if (!fhp->fh_locked)
1946 fh_lock(fhp); /* unlocking is done automatically */
1948 error = inode->i_op->setxattr(fhp->fh_dentry, name,
1951 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1954 error = inode->i_op->removexattr(fhp->fh_dentry, name);
1955 if (error == -ENODATA)
1964 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */