1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/ext2_fs.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
58 #include <asm/uaccess.h>
60 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
64 /* We must ignore files (but only files) which might have mandatory
65 * locks on them because there is no way to know if the accesser has
68 #define IS_ISMNDLK(i) (S_ISREG((i)->i_mode) && MANDATORY_LOCK(i))
71 * This is a cache of readahead params that help us choose the proper
72 * readahead strategy. Initially, we set all readahead parameters to 0
73 * and let the VFS handle things.
74 * If you increase the number of cached files very much, you'll need to
75 * add a hash table here.
78 struct raparms *p_next;
83 struct file_ra_state p_ra;
86 static struct raparms * raparml;
87 static struct raparms * raparm_cache;
90 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
92 * Returns -EAGAIN leaving *dpp and *expp unchanged,
93 * or nfs_ok having possibly changed *dpp and *expp
96 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
97 struct svc_export **expp)
99 struct svc_export *exp = *expp, *exp2 = NULL;
100 struct dentry *dentry = *dpp;
101 struct vfsmount *mnt = mntget(exp->ex_mnt);
102 struct dentry *mounts = dget(dentry);
105 while (follow_down(&mnt,&mounts)&&d_mountpoint(mounts));
107 exp2 = exp_get_by_name(exp->ex_client, mnt, mounts, &rqstp->rq_chandle);
114 if (exp2 && ((exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2))) {
115 /* successfully crossed mount point */
121 if (exp2) exp_put(exp2);
130 * Look up one component of a pathname.
131 * N.B. After this call _both_ fhp and resfh need an fh_put
133 * If the lookup would cross a mountpoint, and the mounted filesystem
134 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
135 * accepted as it stands and the mounted directory is
136 * returned. Otherwise the covered directory is returned.
137 * NOTE: this mountpoint crossing is not supported properly by all
138 * clients and is explicitly disallowed for NFSv3
139 * NeilBrown <neilb@cse.unsw.edu.au>
142 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
143 int len, struct svc_fh *resfh)
145 struct svc_export *exp;
146 struct dentry *dparent;
147 struct dentry *dentry;
150 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
152 /* Obtain dentry and export. */
153 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_EXEC);
157 dparent = fhp->fh_dentry;
158 exp = fhp->fh_export;
163 /* Lookup the name, but don't follow links */
164 if (isdotent(name, len)) {
166 dentry = dget(dparent);
167 else if (dparent != exp->ex_dentry) {
168 dentry = dget_parent(dparent);
169 } else if (!EX_NOHIDE(exp))
170 dentry = dget(dparent); /* .. == . just like at / */
172 /* checking mountpoint crossing is very different when stepping up */
173 struct svc_export *exp2 = NULL;
175 struct vfsmount *mnt = mntget(exp->ex_mnt);
176 dentry = dget(dparent);
177 while(dentry == mnt->mnt_root && follow_up(&mnt, &dentry))
179 dp = dget_parent(dentry);
183 exp2 = exp_parent(exp->ex_client, mnt, dentry,
193 dentry = dget(dparent);
202 dentry = lookup_one_len(name, dparent, len);
203 err = PTR_ERR(dentry);
207 * check if we have crossed a mount point ...
209 if (d_mountpoint(dentry)) {
210 if ((err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
217 * Note: we compose the file handle now, but as the
218 * dentry may be negative, it may need to be updated.
220 err = fh_compose(resfh, exp, dentry, fhp);
221 if (!err && !dentry->d_inode)
234 * Set various file attributes.
235 * N.B. After this call fhp needs an fh_put
238 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
239 int check_guard, time_t guardtime)
241 struct dentry *dentry;
243 int accmode = MAY_SATTR;
249 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
250 accmode |= MAY_WRITE|MAY_OWNER_OVERRIDE;
251 if (iap->ia_valid & ATTR_SIZE)
255 err = fh_verify(rqstp, fhp, ftype, accmode);
259 dentry = fhp->fh_dentry;
260 inode = dentry->d_inode;
262 /* Ignore any mode updates on symlinks */
263 if (S_ISLNK(inode->i_mode))
264 iap->ia_valid &= ~ATTR_MODE;
269 /* NFSv2 does not differentiate between "set-[ac]time-to-now"
270 * which only requires access, and "set-[ac]time-to-X" which
271 * requires ownership.
272 * So if it looks like it might be "set both to the same time which
273 * is close to now", and if inode_change_ok fails, then we
274 * convert to "set to now" instead of "set to explicit time"
276 * We only call inode_change_ok as the last test as technically
277 * it is not an interface that we should be using. It is only
278 * valid if the filesystem does not define it's own i_op->setattr.
280 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
281 #define MAX_TOUCH_TIME_ERROR (30*60)
282 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET
283 && iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec
285 /* Looks probable. Now just make sure time is in the right ballpark.
286 * Solaris, at least, doesn't seem to care what the time request is.
287 * We require it be within 30 minutes of now.
289 time_t delta = iap->ia_atime.tv_sec - get_seconds();
290 if (delta<0) delta = -delta;
291 if (delta < MAX_TOUCH_TIME_ERROR &&
292 inode_change_ok(inode, iap) != 0) {
293 /* turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME
294 * this will cause notify_change to set these times to "now"
296 iap->ia_valid &= ~BOTH_TIME_SET;
300 /* The size case is special. It changes the file as well as the attributes. */
301 if (iap->ia_valid & ATTR_SIZE) {
302 if (iap->ia_size < inode->i_size) {
303 err = nfsd_permission(fhp->fh_export, dentry, MAY_TRUNC|MAY_OWNER_OVERRIDE);
309 * If we are changing the size of the file, then
310 * we need to break all leases.
312 err = break_lease(inode, FMODE_WRITE | O_NONBLOCK);
313 if (err == -EWOULDBLOCK)
315 if (err) /* ENOMEM or EWOULDBLOCK */
318 err = get_write_access(inode);
323 err = locks_verify_truncate(inode, NULL, iap->ia_size);
325 put_write_access(inode);
331 imode = inode->i_mode;
332 if (iap->ia_valid & ATTR_MODE) {
333 iap->ia_mode &= S_IALLUGO;
334 imode = iap->ia_mode |= (imode & ~S_IALLUGO);
337 /* Revoke setuid/setgid bit on chown/chgrp */
338 if ((iap->ia_valid & ATTR_UID) && iap->ia_uid != inode->i_uid)
339 iap->ia_valid |= ATTR_KILL_SUID;
340 if ((iap->ia_valid & ATTR_GID) && iap->ia_gid != inode->i_gid)
341 iap->ia_valid |= ATTR_KILL_SGID;
343 /* Change the attributes. */
345 iap->ia_valid |= ATTR_CTIME;
347 err = nfserr_notsync;
348 if (!check_guard || guardtime == inode->i_ctime.tv_sec) {
350 err = notify_change(dentry, iap);
355 put_write_access(inode);
357 if (EX_ISSYNC(fhp->fh_export))
358 write_inode_now(inode, 1);
367 #if defined(CONFIG_NFSD_V2_ACL) || \
368 defined(CONFIG_NFSD_V3_ACL) || \
369 defined(CONFIG_NFSD_V4)
370 static ssize_t nfsd_getxattr(struct dentry *dentry, char *key, void **buf)
374 buflen = vfs_getxattr(dentry, key, NULL, 0);
378 *buf = kmalloc(buflen, GFP_KERNEL);
382 return vfs_getxattr(dentry, key, *buf, buflen);
386 #if defined(CONFIG_NFSD_V4)
388 set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
395 buflen = posix_acl_xattr_size(pacl->a_count);
396 buf = kmalloc(buflen, GFP_KERNEL);
401 len = posix_acl_to_xattr(pacl, buf, buflen);
407 error = vfs_setxattr(dentry, key, buf, len, 0);
414 nfsd4_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
415 struct nfs4_acl *acl)
418 struct dentry *dentry;
420 struct posix_acl *pacl = NULL, *dpacl = NULL;
421 unsigned int flags = 0;
424 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, MAY_SATTR);
428 dentry = fhp->fh_dentry;
429 inode = dentry->d_inode;
430 if (S_ISDIR(inode->i_mode))
431 flags = NFS4_ACL_DIR;
433 error = nfs4_acl_nfsv4_to_posix(acl, &pacl, &dpacl, flags);
434 if (error == -EINVAL) {
435 error = nfserr_attrnotsupp;
437 } else if (error < 0)
441 error = set_nfsv4_acl_one(dentry, pacl, POSIX_ACL_XATTR_ACCESS);
447 error = set_nfsv4_acl_one(dentry, dpacl, POSIX_ACL_XATTR_DEFAULT);
455 posix_acl_release(pacl);
456 posix_acl_release(dpacl);
459 error = nfserrno(error);
463 static struct posix_acl *
464 _get_posix_acl(struct dentry *dentry, char *key)
467 struct posix_acl *pacl = NULL;
470 buflen = nfsd_getxattr(dentry, key, &buf);
474 return ERR_PTR(buflen);
476 pacl = posix_acl_from_xattr(buf, buflen);
482 nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry, struct nfs4_acl **acl)
484 struct inode *inode = dentry->d_inode;
486 struct posix_acl *pacl = NULL, *dpacl = NULL;
487 unsigned int flags = 0;
489 pacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_ACCESS);
490 if (IS_ERR(pacl) && PTR_ERR(pacl) == -ENODATA)
491 pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
493 error = PTR_ERR(pacl);
498 if (S_ISDIR(inode->i_mode)) {
499 dpacl = _get_posix_acl(dentry, POSIX_ACL_XATTR_DEFAULT);
500 if (IS_ERR(dpacl) && PTR_ERR(dpacl) == -ENODATA)
502 else if (IS_ERR(dpacl)) {
503 error = PTR_ERR(dpacl);
507 flags = NFS4_ACL_DIR;
510 *acl = nfs4_acl_posix_to_nfsv4(pacl, dpacl, flags);
512 error = PTR_ERR(*acl);
516 posix_acl_release(pacl);
517 posix_acl_release(dpacl);
521 #endif /* defined(CONFIG_NFS_V4) */
523 #ifdef CONFIG_NFSD_V3
525 * Check server access rights to a file system object
531 static struct accessmap nfs3_regaccess[] = {
532 { NFS3_ACCESS_READ, MAY_READ },
533 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
534 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_TRUNC },
535 { NFS3_ACCESS_EXTEND, MAY_WRITE },
540 static struct accessmap nfs3_diraccess[] = {
541 { NFS3_ACCESS_READ, MAY_READ },
542 { NFS3_ACCESS_LOOKUP, MAY_EXEC },
543 { NFS3_ACCESS_MODIFY, MAY_EXEC|MAY_WRITE|MAY_TRUNC },
544 { NFS3_ACCESS_EXTEND, MAY_EXEC|MAY_WRITE },
545 { NFS3_ACCESS_DELETE, MAY_REMOVE },
550 static struct accessmap nfs3_anyaccess[] = {
551 /* Some clients - Solaris 2.6 at least, make an access call
552 * to the server to check for access for things like /dev/null
553 * (which really, the server doesn't care about). So
554 * We provide simple access checking for them, looking
555 * mainly at mode bits, and we make sure to ignore read-only
558 { NFS3_ACCESS_READ, MAY_READ },
559 { NFS3_ACCESS_EXECUTE, MAY_EXEC },
560 { NFS3_ACCESS_MODIFY, MAY_WRITE|MAY_LOCAL_ACCESS },
561 { NFS3_ACCESS_EXTEND, MAY_WRITE|MAY_LOCAL_ACCESS },
567 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
569 struct accessmap *map;
570 struct svc_export *export;
571 struct dentry *dentry;
572 u32 query, result = 0, sresult = 0;
575 error = fh_verify(rqstp, fhp, 0, MAY_NOP);
579 export = fhp->fh_export;
580 dentry = fhp->fh_dentry;
582 if (S_ISREG(dentry->d_inode->i_mode))
583 map = nfs3_regaccess;
584 else if (S_ISDIR(dentry->d_inode->i_mode))
585 map = nfs3_diraccess;
587 map = nfs3_anyaccess;
591 for (; map->access; map++) {
592 if (map->access & query) {
595 sresult |= map->access;
597 err2 = nfsd_permission(export, dentry, map->how);
600 result |= map->access;
603 /* the following error codes just mean the access was not allowed,
604 * rather than an error occurred */
608 /* simply don't "or" in the access bit. */
618 *supported = sresult;
623 #endif /* CONFIG_NFSD_V3 */
628 * Open an existing file or directory.
629 * The access argument indicates the type of open (read/write/lock)
630 * N.B. After this call fhp needs an fh_put
633 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
634 int access, struct file **filp)
636 struct dentry *dentry;
638 int flags = O_RDONLY|O_LARGEFILE, err;
641 * If we get here, then the client has already done an "open",
642 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
643 * in case a chmod has now revoked permission.
645 err = fh_verify(rqstp, fhp, type, access | MAY_OWNER_OVERRIDE);
649 dentry = fhp->fh_dentry;
650 inode = dentry->d_inode;
652 /* Disallow write access to files with the append-only bit set
653 * or any access when mandatory locking enabled
656 if (IS_APPEND(inode) && (access & MAY_WRITE))
658 if (IS_ISMNDLK(inode))
665 * Check to see if there are any leases on this file.
666 * This may block while leases are broken.
668 err = break_lease(inode, O_NONBLOCK | ((access & MAY_WRITE) ? FMODE_WRITE : 0));
669 if (err == -EWOULDBLOCK)
671 if (err) /* NOMEM or WOULDBLOCK */
674 if (access & MAY_WRITE) {
675 flags = O_WRONLY|O_LARGEFILE;
679 *filp = dentry_open(dget(dentry), mntget(fhp->fh_export->ex_mnt), flags);
681 err = PTR_ERR(*filp);
693 nfsd_close(struct file *filp)
700 * As this calls fsync (not fdatasync) there is no need for a write_inode
703 static inline int nfsd_dosync(struct file *filp, struct dentry *dp,
704 const struct file_operations *fop)
706 struct inode *inode = dp->d_inode;
707 int (*fsync) (struct file *, struct dentry *, int);
710 err = filemap_fdatawrite(inode->i_mapping);
711 if (err == 0 && fop && (fsync = fop->fsync))
712 err = fsync(filp, dp, 0);
714 err = filemap_fdatawait(inode->i_mapping);
721 nfsd_sync(struct file *filp)
724 struct inode *inode = filp->f_dentry->d_inode;
725 dprintk("nfsd: sync file %s\n", filp->f_dentry->d_name.name);
726 mutex_lock(&inode->i_mutex);
727 err=nfsd_dosync(filp, filp->f_dentry, filp->f_op);
728 mutex_unlock(&inode->i_mutex);
734 nfsd_sync_dir(struct dentry *dp)
736 return nfsd_dosync(NULL, dp, dp->d_inode->i_fop);
740 * Obtain the readahead parameters for the file
741 * specified by (dev, ino).
743 static DEFINE_SPINLOCK(ra_lock);
745 static inline struct raparms *
746 nfsd_get_raparms(dev_t dev, ino_t ino)
748 struct raparms *ra, **rap, **frap = NULL;
752 for (rap = &raparm_cache; (ra = *rap); rap = &ra->p_next) {
753 if (ra->p_ino == ino && ra->p_dev == dev)
756 if (ra->p_count == 0)
759 depth = nfsdstats.ra_size*11/10;
761 spin_unlock(&ra_lock);
770 if (rap != &raparm_cache) {
772 ra->p_next = raparm_cache;
776 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
777 spin_unlock(&ra_lock);
782 * Grab and keep cached pages assosiated with a file in the svc_rqst
783 * so that they can be passed to the netowork sendmsg/sendpage routines
784 * directrly. They will be released after the sending has completed.
787 nfsd_read_actor(read_descriptor_t *desc, struct page *page, unsigned long offset , unsigned long size)
789 unsigned long count = desc->count;
790 struct svc_rqst *rqstp = desc->arg.data;
795 if (rqstp->rq_res.page_len == 0) {
797 rqstp->rq_respages[rqstp->rq_resused++] = page;
798 rqstp->rq_res.page_base = offset;
799 rqstp->rq_res.page_len = size;
800 } else if (page != rqstp->rq_respages[rqstp->rq_resused-1]) {
802 rqstp->rq_respages[rqstp->rq_resused++] = page;
803 rqstp->rq_res.page_len += size;
805 rqstp->rq_res.page_len += size;
808 desc->count = count - size;
809 desc->written += size;
814 nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
815 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
823 inode = file->f_dentry->d_inode;
825 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
826 (!lock_may_read(inode, offset, *count)))
830 /* Get readahead parameters */
831 ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
834 file->f_ra = ra->p_ra;
836 if (file->f_op->sendfile) {
837 svc_pushback_unused_pages(rqstp);
838 err = file->f_op->sendfile(file, &offset, *count,
839 nfsd_read_actor, rqstp);
843 err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
847 /* Write back readahead params */
850 ra->p_ra = file->f_ra;
853 spin_unlock(&ra_lock);
857 nfsdstats.io_read += err;
860 fsnotify_access(file->f_dentry);
867 static void kill_suid(struct dentry *dentry)
870 ia.ia_valid = ATTR_KILL_SUID | ATTR_KILL_SGID;
872 mutex_lock(&dentry->d_inode->i_mutex);
873 notify_change(dentry, &ia);
874 mutex_unlock(&dentry->d_inode->i_mutex);
878 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
879 loff_t offset, struct kvec *vec, int vlen,
880 unsigned long cnt, int *stablep)
882 struct svc_export *exp;
883 struct dentry *dentry;
887 int stable = *stablep;
892 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
893 (!lock_may_write(file->f_dentry->d_inode, offset, cnt)))
897 dentry = file->f_dentry;
898 inode = dentry->d_inode;
899 exp = fhp->fh_export;
902 * Request sync writes if
903 * - the sync export option has been set, or
904 * - the client requested O_SYNC behavior (NFSv3 feature).
905 * - The file system doesn't support fsync().
906 * When gathered writes have been configured for this volume,
907 * flushing the data to disk is handled separately below.
910 if (file->f_op->fsync == 0) {/* COMMIT3 cannot work */
912 *stablep = 2; /* FILE_SYNC */
917 if (stable && !EX_WGATHER(exp))
918 file->f_flags |= O_SYNC;
920 /* Write the data. */
921 oldfs = get_fs(); set_fs(KERNEL_DS);
922 err = vfs_writev(file, (struct iovec __user *)vec, vlen, &offset);
925 nfsdstats.io_write += cnt;
926 fsnotify_modify(file->f_dentry);
929 /* clear setuid/setgid flag after write */
930 if (err >= 0 && (inode->i_mode & (S_ISUID | S_ISGID)))
933 if (err >= 0 && stable) {
934 static ino_t last_ino;
935 static dev_t last_dev;
938 * Gathered writes: If another process is currently
939 * writing to the file, there's a high chance
940 * this is another nfsd (triggered by a bulk write
941 * from a client's biod). Rather than syncing the
942 * file with each write request, we sleep for 10 msec.
944 * I don't know if this roughly approximates
945 * C. Juszak's idea of gathered writes, but it's a
946 * nice and simple solution (IMHO), and it seems to
949 if (EX_WGATHER(exp)) {
950 if (atomic_read(&inode->i_writecount) > 1
951 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
952 dprintk("nfsd: write defer %d\n", current->pid);
954 dprintk("nfsd: write resume %d\n", current->pid);
957 if (inode->i_state & I_DIRTY) {
958 dprintk("nfsd: write sync %d\n", current->pid);
962 wake_up(&inode->i_wait);
965 last_ino = inode->i_ino;
966 last_dev = inode->i_sb->s_dev;
969 dprintk("nfsd: write complete err=%d\n", err);
979 * Read data from a file. count must contain the requested read count
980 * on entry. On return, *count contains the number of bytes actually read.
981 * N.B. After this call fhp needs an fh_put
984 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
985 loff_t offset, struct kvec *vec, int vlen,
986 unsigned long *count)
991 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
992 MAY_READ|MAY_OWNER_OVERRIDE);
995 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
997 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_READ, &file);
1000 err = nfsd_vfs_read(rqstp, fhp, file, offset, vec, vlen, count);
1008 * Write data to a file.
1009 * The stable flag requests synchronous writes.
1010 * N.B. After this call fhp needs an fh_put
1013 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1014 loff_t offset, struct kvec *vec, int vlen, unsigned long cnt,
1020 err = nfsd_permission(fhp->fh_export, fhp->fh_dentry,
1021 MAY_WRITE|MAY_OWNER_OVERRIDE);
1024 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1027 err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file);
1032 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1040 #ifdef CONFIG_NFSD_V3
1042 * Commit all pending writes to stable storage.
1043 * Strictly speaking, we could sync just the indicated file region here,
1044 * but there's currently no way we can ask the VFS to do so.
1046 * Unfortunately we cannot lock the file to make sure we return full WCC
1047 * data to the client, as locking happens lower down in the filesystem.
1050 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1051 loff_t offset, unsigned long count)
1056 if ((u64)count > ~(u64)offset)
1057 return nfserr_inval;
1059 if ((err = nfsd_open(rqstp, fhp, S_IFREG, MAY_WRITE, &file)) != 0)
1061 if (EX_ISSYNC(fhp->fh_export)) {
1062 if (file->f_op && file->f_op->fsync) {
1063 err = nfserrno(nfsd_sync(file));
1065 err = nfserr_notsupp;
1072 #endif /* CONFIG_NFSD_V3 */
1075 * Create a file (regular, directory, device, fifo); UNIX sockets
1076 * not yet implemented.
1077 * If the response fh has been verified, the parent directory should
1078 * already be locked. Note that the parent directory is left locked.
1080 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1083 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1084 char *fname, int flen, struct iattr *iap,
1085 int type, dev_t rdev, struct svc_fh *resfhp)
1087 struct dentry *dentry, *dchild = NULL;
1095 if (isdotent(fname, flen))
1098 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1102 dentry = fhp->fh_dentry;
1103 dirp = dentry->d_inode;
1105 err = nfserr_notdir;
1106 if(!dirp->i_op || !dirp->i_op->lookup)
1109 * Check whether the response file handle has been verified yet.
1110 * If it has, the parent directory should already be locked.
1112 if (!resfhp->fh_dentry) {
1113 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1115 dchild = lookup_one_len(fname, dentry, flen);
1116 err = PTR_ERR(dchild);
1119 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1123 /* called from nfsd_proc_create */
1124 dchild = dget(resfhp->fh_dentry);
1125 if (!fhp->fh_locked) {
1126 /* not actually possible */
1128 "nfsd_create: parent %s/%s not locked!\n",
1129 dentry->d_parent->d_name.name,
1130 dentry->d_name.name);
1136 * Make sure the child dentry is still negative ...
1139 if (dchild->d_inode) {
1140 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1141 dentry->d_name.name, dchild->d_name.name);
1145 if (!(iap->ia_valid & ATTR_MODE))
1147 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1150 * Get the dir op function pointer.
1155 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1158 err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1164 err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1167 printk("nfsd: bad file type %o in nfsd_create\n", type);
1173 if (EX_ISSYNC(fhp->fh_export)) {
1174 err = nfserrno(nfsd_sync_dir(dentry));
1175 write_inode_now(dchild->d_inode, 1);
1179 /* Set file attributes. Mode has already been set and
1180 * setting uid/gid works only for root. Irix appears to
1181 * send along the gid when it tries to implement setgid
1182 * directories via NFS.
1184 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID|ATTR_MODE)) != 0) {
1185 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1190 * Update the file handle to get the new inode info.
1193 err = fh_update(resfhp);
1195 if (dchild && !IS_ERR(dchild))
1200 err = nfserrno(err);
1204 #ifdef CONFIG_NFSD_V3
1206 * NFSv3 version of nfsd_create
1209 nfsd_create_v3(struct svc_rqst *rqstp, struct svc_fh *fhp,
1210 char *fname, int flen, struct iattr *iap,
1211 struct svc_fh *resfhp, int createmode, u32 *verifier,
1214 struct dentry *dentry, *dchild = NULL;
1217 __u32 v_mtime=0, v_atime=0;
1224 if (isdotent(fname, flen))
1226 if (!(iap->ia_valid & ATTR_MODE))
1228 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1232 dentry = fhp->fh_dentry;
1233 dirp = dentry->d_inode;
1235 /* Get all the sanity checks out of the way before
1236 * we lock the parent. */
1237 err = nfserr_notdir;
1238 if(!dirp->i_op || !dirp->i_op->lookup)
1243 * Compose the response file handle.
1245 dchild = lookup_one_len(fname, dentry, flen);
1246 err = PTR_ERR(dchild);
1250 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1254 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1255 /* while the verifier would fit in mtime+atime,
1256 * solaris7 gets confused (bugid 4218508) if these have
1257 * the high bit set, so we use the mode as well
1259 v_mtime = verifier[0]&0x7fffffff;
1260 v_atime = verifier[1]&0x7fffffff;
1262 | ((verifier[0]&0x80000000) >> (32-7)) /* u+x */
1263 | ((verifier[1]&0x80000000) >> (32-9)) /* u+r */
1267 if (dchild->d_inode) {
1270 switch (createmode) {
1271 case NFS3_CREATE_UNCHECKED:
1272 if (! S_ISREG(dchild->d_inode->i_mode))
1275 /* in nfsv4, we need to treat this case a little
1276 * differently. we don't want to truncate the
1277 * file now; this would be wrong if the OPEN
1278 * fails for some other reason. furthermore,
1279 * if the size is nonzero, we should ignore it
1280 * according to spec!
1282 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1285 iap->ia_valid &= ATTR_SIZE;
1289 case NFS3_CREATE_EXCLUSIVE:
1290 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1291 && dchild->d_inode->i_atime.tv_sec == v_atime
1292 && dchild->d_inode->i_mode == v_mode
1293 && dchild->d_inode->i_size == 0 )
1296 case NFS3_CREATE_GUARDED:
1302 err = vfs_create(dirp, dchild, iap->ia_mode, NULL);
1306 if (EX_ISSYNC(fhp->fh_export)) {
1307 err = nfserrno(nfsd_sync_dir(dentry));
1308 /* setattr will sync the child (or not) */
1311 if (createmode == NFS3_CREATE_EXCLUSIVE) {
1312 /* Cram the verifier into atime/mtime/mode */
1313 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1314 | ATTR_MTIME_SET|ATTR_ATIME_SET
1316 /* XXX someone who knows this better please fix it for nsec */
1317 iap->ia_mtime.tv_sec = v_mtime;
1318 iap->ia_atime.tv_sec = v_atime;
1319 iap->ia_mtime.tv_nsec = 0;
1320 iap->ia_atime.tv_nsec = 0;
1321 iap->ia_mode = v_mode;
1324 /* Set file attributes.
1325 * Mode has already been set but we might need to reset it
1326 * for CREATE_EXCLUSIVE
1327 * Irix appears to send along the gid when it tries to
1328 * implement setgid directories via NFS. Clear out all that cruft.
1331 if ((iap->ia_valid &= ~(ATTR_UID|ATTR_GID)) != 0) {
1332 int err2 = nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1338 * Update the filehandle to get the new inode info.
1341 err = fh_update(resfhp);
1345 if (dchild && !IS_ERR(dchild))
1350 err = nfserrno(err);
1353 #endif /* CONFIG_NFSD_V3 */
1356 * Read a symlink. On entry, *lenp must contain the maximum path length that
1357 * fits into the buffer. On return, it contains the true length.
1358 * N.B. After this call fhp needs an fh_put
1361 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1363 struct dentry *dentry;
1364 struct inode *inode;
1368 err = fh_verify(rqstp, fhp, S_IFLNK, MAY_NOP);
1372 dentry = fhp->fh_dentry;
1373 inode = dentry->d_inode;
1376 if (!inode->i_op || !inode->i_op->readlink)
1379 touch_atime(fhp->fh_export->ex_mnt, dentry);
1380 /* N.B. Why does this call need a get_fs()??
1381 * Remove the set_fs and watch the fireworks:-) --okir
1384 oldfs = get_fs(); set_fs(KERNEL_DS);
1385 err = inode->i_op->readlink(dentry, buf, *lenp);
1396 err = nfserrno(err);
1401 * Create a symlink and look up its inode
1402 * N.B. After this call _both_ fhp and resfhp need an fh_put
1405 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1406 char *fname, int flen,
1407 char *path, int plen,
1408 struct svc_fh *resfhp,
1411 struct dentry *dentry, *dnew;
1419 if (isdotent(fname, flen))
1422 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_CREATE);
1426 dentry = fhp->fh_dentry;
1427 dnew = lookup_one_len(fname, dentry, flen);
1428 err = PTR_ERR(dnew);
1433 /* Only the MODE ATTRibute is even vaguely meaningful */
1434 if (iap && (iap->ia_valid & ATTR_MODE))
1435 mode = iap->ia_mode & S_IALLUGO;
1437 if (unlikely(path[plen] != 0)) {
1438 char *path_alloced = kmalloc(plen+1, GFP_KERNEL);
1439 if (path_alloced == NULL)
1442 strncpy(path_alloced, path, plen);
1443 path_alloced[plen] = 0;
1444 err = vfs_symlink(dentry->d_inode, dnew, path_alloced, mode);
1445 kfree(path_alloced);
1448 err = vfs_symlink(dentry->d_inode, dnew, path, mode);
1451 if (EX_ISSYNC(fhp->fh_export))
1452 err = nfsd_sync_dir(dentry);
1454 err = nfserrno(err);
1457 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1459 if (err==0) err = cerr;
1464 err = nfserrno(err);
1470 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1473 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1474 char *name, int len, struct svc_fh *tfhp)
1476 struct dentry *ddir, *dnew, *dold;
1477 struct inode *dirp, *dest;
1480 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_CREATE);
1483 err = fh_verify(rqstp, tfhp, -S_IFDIR, MAY_NOP);
1491 if (isdotent(name, len))
1495 ddir = ffhp->fh_dentry;
1496 dirp = ddir->d_inode;
1498 dnew = lookup_one_len(name, ddir, len);
1499 err = PTR_ERR(dnew);
1503 dold = tfhp->fh_dentry;
1504 dest = dold->d_inode;
1506 err = vfs_link(dold, dirp, dnew);
1508 if (EX_ISSYNC(ffhp->fh_export)) {
1509 err = nfserrno(nfsd_sync_dir(ddir));
1510 write_inode_now(dest, 1);
1513 if (err == -EXDEV && rqstp->rq_vers == 2)
1516 err = nfserrno(err);
1525 err = nfserrno(err);
1531 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1534 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1535 struct svc_fh *tfhp, char *tname, int tlen)
1537 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1538 struct inode *fdir, *tdir;
1541 err = fh_verify(rqstp, ffhp, S_IFDIR, MAY_REMOVE);
1544 err = fh_verify(rqstp, tfhp, S_IFDIR, MAY_CREATE);
1548 fdentry = ffhp->fh_dentry;
1549 fdir = fdentry->d_inode;
1551 tdentry = tfhp->fh_dentry;
1552 tdir = tdentry->d_inode;
1554 err = (rqstp->rq_vers == 2) ? nfserr_acces : nfserr_xdev;
1555 if (fdir->i_sb != tdir->i_sb)
1559 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1562 /* cannot use fh_lock as we need deadlock protective ordering
1563 * so do it by hand */
1564 trap = lock_rename(tdentry, fdentry);
1565 ffhp->fh_locked = tfhp->fh_locked = 1;
1569 odentry = lookup_one_len(fname, fdentry, flen);
1570 err = PTR_ERR(odentry);
1571 if (IS_ERR(odentry))
1575 if (!odentry->d_inode)
1578 if (odentry == trap)
1581 ndentry = lookup_one_len(tname, tdentry, tlen);
1582 err = PTR_ERR(ndentry);
1583 if (IS_ERR(ndentry))
1586 if (ndentry == trap)
1590 if ((ffhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1591 ((atomic_read(&odentry->d_count) > 1)
1592 || (atomic_read(&ndentry->d_count) > 1))) {
1596 err = vfs_rename(fdir, odentry, tdir, ndentry);
1597 if (!err && EX_ISSYNC(tfhp->fh_export)) {
1598 err = nfsd_sync_dir(tdentry);
1600 err = nfsd_sync_dir(fdentry);
1609 err = nfserrno(err);
1611 /* we cannot reply on fh_unlock on the two filehandles,
1612 * as that would do the wrong thing if the two directories
1613 * were the same, so again we do it by hand
1615 fill_post_wcc(ffhp);
1616 fill_post_wcc(tfhp);
1617 unlock_rename(tdentry, fdentry);
1618 ffhp->fh_locked = tfhp->fh_locked = 0;
1625 * Unlink a file or directory
1626 * N.B. After this call fhp needs an fh_put
1629 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1630 char *fname, int flen)
1632 struct dentry *dentry, *rdentry;
1637 if (!flen || isdotent(fname, flen))
1639 err = fh_verify(rqstp, fhp, S_IFDIR, MAY_REMOVE);
1644 dentry = fhp->fh_dentry;
1645 dirp = dentry->d_inode;
1647 rdentry = lookup_one_len(fname, dentry, flen);
1648 err = PTR_ERR(rdentry);
1649 if (IS_ERR(rdentry))
1652 if (!rdentry->d_inode) {
1659 type = rdentry->d_inode->i_mode & S_IFMT;
1661 if (type != S_IFDIR) { /* It's UNLINK */
1663 if ((fhp->fh_export->ex_flags & NFSEXP_MSNFS) &&
1664 (atomic_read(&rdentry->d_count) > 1)) {
1668 err = vfs_unlink(dirp, rdentry);
1669 } else { /* It's RMDIR */
1670 err = vfs_rmdir(dirp, rdentry);
1676 EX_ISSYNC(fhp->fh_export))
1677 err = nfsd_sync_dir(dentry);
1680 err = nfserrno(err);
1686 * Read entries from a directory.
1687 * The NFSv3/4 verifier we ignore for now.
1690 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1691 struct readdir_cd *cdp, encode_dent_fn func)
1695 loff_t offset = *offsetp;
1697 err = nfsd_open(rqstp, fhp, S_IFDIR, MAY_READ, &file);
1701 offset = vfs_llseek(file, offset, 0);
1703 err = nfserrno((int)offset);
1708 * Read the directory entries. This silly loop is necessary because
1709 * readdir() is not guaranteed to fill up the entire buffer, but
1710 * may choose to do less.
1714 cdp->err = nfserr_eof; /* will be cleared on successful read */
1715 err = vfs_readdir(file, (filldir_t) func, cdp);
1716 } while (err >=0 && cdp->err == nfs_ok);
1718 err = nfserrno(err);
1721 *offsetp = vfs_llseek(file, 0, 1);
1723 if (err == nfserr_eof || err == nfserr_toosmall)
1724 err = nfs_ok; /* can still be found in ->err */
1732 * Get file system stats
1733 * N.B. After this call fhp needs an fh_put
1736 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat)
1738 int err = fh_verify(rqstp, fhp, 0, MAY_NOP);
1739 if (!err && vfs_statfs(fhp->fh_dentry,stat))
1745 * Check for a user's access permissions to this inode.
1748 nfsd_permission(struct svc_export *exp, struct dentry *dentry, int acc)
1750 struct inode *inode = dentry->d_inode;
1756 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1758 (acc & MAY_READ)? " read" : "",
1759 (acc & MAY_WRITE)? " write" : "",
1760 (acc & MAY_EXEC)? " exec" : "",
1761 (acc & MAY_SATTR)? " sattr" : "",
1762 (acc & MAY_TRUNC)? " trunc" : "",
1763 (acc & MAY_LOCK)? " lock" : "",
1764 (acc & MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1766 IS_IMMUTABLE(inode)? " immut" : "",
1767 IS_APPEND(inode)? " append" : "",
1768 IS_RDONLY(inode)? " ro" : "");
1769 dprintk(" owner %d/%d user %d/%d\n",
1770 inode->i_uid, inode->i_gid, current->fsuid, current->fsgid);
1773 /* Normally we reject any write/sattr etc access on a read-only file
1774 * system. But if it is IRIX doing check on write-access for a
1775 * device special file, we ignore rofs.
1777 if (!(acc & MAY_LOCAL_ACCESS))
1778 if (acc & (MAY_WRITE | MAY_SATTR | MAY_TRUNC)) {
1779 if (EX_RDONLY(exp) || IS_RDONLY(inode))
1781 if (/* (acc & MAY_WRITE) && */ IS_IMMUTABLE(inode))
1784 if ((acc & MAY_TRUNC) && IS_APPEND(inode))
1787 if (acc & MAY_LOCK) {
1788 /* If we cannot rely on authentication in NLM requests,
1789 * just allow locks, otherwise require read permission, or
1792 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
1795 acc = MAY_READ | MAY_OWNER_OVERRIDE;
1798 * The file owner always gets access permission for accesses that
1799 * would normally be checked at open time. This is to make
1800 * file access work even when the client has done a fchmod(fd, 0).
1802 * However, `cp foo bar' should fail nevertheless when bar is
1803 * readonly. A sensible way to do this might be to reject all
1804 * attempts to truncate a read-only file, because a creat() call
1805 * always implies file truncation.
1806 * ... but this isn't really fair. A process may reasonably call
1807 * ftruncate on an open file descriptor on a file with perm 000.
1808 * We must trust the client to do permission checking - using "ACCESS"
1811 if ((acc & MAY_OWNER_OVERRIDE) &&
1812 inode->i_uid == current->fsuid)
1815 err = permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC), NULL);
1817 /* Allow read access to binaries even when mode 111 */
1818 if (err == -EACCES && S_ISREG(inode->i_mode) &&
1819 acc == (MAY_READ | MAY_OWNER_OVERRIDE))
1820 err = permission(inode, MAY_EXEC, NULL);
1822 return err? nfserrno(err) : 0;
1826 nfsd_racache_shutdown(void)
1830 dprintk("nfsd: freeing readahead buffers.\n");
1832 raparm_cache = raparml = NULL;
1835 * Initialize readahead param cache
1838 nfsd_racache_init(int cache_size)
1844 raparml = kmalloc(sizeof(struct raparms) * cache_size, GFP_KERNEL);
1846 if (raparml != NULL) {
1847 dprintk("nfsd: allocating %d readahead buffers.\n",
1849 memset(raparml, 0, sizeof(struct raparms) * cache_size);
1850 for (i = 0; i < cache_size - 1; i++) {
1851 raparml[i].p_next = raparml + i + 1;
1853 raparm_cache = raparml;
1856 "nfsd: Could not allocate memory read-ahead cache.\n");
1859 nfsdstats.ra_size = cache_size;
1863 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
1865 nfsd_get_posix_acl(struct svc_fh *fhp, int type)
1867 struct inode *inode = fhp->fh_dentry->d_inode;
1871 struct posix_acl *acl;
1873 if (!IS_POSIXACL(inode))
1874 return ERR_PTR(-EOPNOTSUPP);
1877 case ACL_TYPE_ACCESS:
1878 name = POSIX_ACL_XATTR_ACCESS;
1880 case ACL_TYPE_DEFAULT:
1881 name = POSIX_ACL_XATTR_DEFAULT;
1884 return ERR_PTR(-EOPNOTSUPP);
1887 size = nfsd_getxattr(fhp->fh_dentry, name, &value);
1889 return ERR_PTR(size);
1891 acl = posix_acl_from_xattr(value, size);
1897 nfsd_set_posix_acl(struct svc_fh *fhp, int type, struct posix_acl *acl)
1899 struct inode *inode = fhp->fh_dentry->d_inode;
1905 if (!IS_POSIXACL(inode) || !inode->i_op ||
1906 !inode->i_op->setxattr || !inode->i_op->removexattr)
1909 case ACL_TYPE_ACCESS:
1910 name = POSIX_ACL_XATTR_ACCESS;
1912 case ACL_TYPE_DEFAULT:
1913 name = POSIX_ACL_XATTR_DEFAULT;
1919 if (acl && acl->a_count) {
1920 size = posix_acl_xattr_size(acl->a_count);
1921 value = kmalloc(size, GFP_KERNEL);
1924 error = posix_acl_to_xattr(acl, value, size);
1932 error = vfs_setxattr(fhp->fh_dentry, name, value, size, 0);
1934 if (!S_ISDIR(inode->i_mode) && type == ACL_TYPE_DEFAULT)
1937 error = vfs_removexattr(fhp->fh_dentry, name);
1938 if (error == -ENODATA)
1947 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */