2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/falloc.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
32 #include <linux/security.h>
36 #endif /* CONFIG_NFSD_V3 */
41 #endif /* CONFIG_NFSD_V4 */
46 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
50 * This is a cache of readahead params that help us choose the proper
51 * readahead strategy. Initially, we set all readahead parameters to 0
52 * and let the VFS handle things.
53 * If you increase the number of cached files very much, you'll need to
54 * add a hash table here.
57 struct raparms *p_next;
62 struct file_ra_state p_ra;
63 unsigned int p_hindex;
66 struct raparm_hbucket {
67 struct raparms *pb_head;
69 } ____cacheline_aligned_in_smp;
71 #define RAPARM_HASH_BITS 4
72 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
73 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
74 static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
77 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
79 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
80 * or nfs_ok having possibly changed *dpp and *expp
83 nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
84 struct svc_export **expp)
86 struct svc_export *exp = *expp, *exp2 = NULL;
87 struct dentry *dentry = *dpp;
88 struct path path = {.mnt = mntget(exp->ex_path.mnt),
89 .dentry = dget(dentry)};
92 err = follow_down(&path);
96 exp2 = rqst_exp_get_by_name(rqstp, &path);
100 * We normally allow NFS clients to continue
101 * "underneath" a mountpoint that is not exported.
102 * The exception is V4ROOT, where no traversal is ever
103 * allowed without an explicit export of the new
106 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
111 if (nfsd_v4client(rqstp) ||
112 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
113 /* successfully crossed mount point */
115 * This is subtle: path.dentry is *not* on path.mnt
116 * at this point. The only reason we are safe is that
117 * original mnt is pinned down by exp, so we should
118 * put path *before* putting exp
121 path.dentry = dentry;
131 static void follow_to_parent(struct path *path)
135 while (path->dentry == path->mnt->mnt_root && follow_up(path))
137 dp = dget_parent(path->dentry);
142 static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
144 struct svc_export *exp2;
145 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
146 .dentry = dget(dparent)};
148 follow_to_parent(&path);
150 exp2 = rqst_exp_parent(rqstp, &path);
151 if (PTR_ERR(exp2) == -ENOENT) {
152 *dentryp = dget(dparent);
153 } else if (IS_ERR(exp2)) {
155 return PTR_ERR(exp2);
157 *dentryp = dget(path.dentry);
166 * For nfsd purposes, we treat V4ROOT exports as though there was an
167 * export at *every* directory.
169 int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
171 if (d_mountpoint(dentry))
173 if (nfsd4_is_junction(dentry))
175 if (!(exp->ex_flags & NFSEXP_V4ROOT))
177 return d_inode(dentry) != NULL;
181 nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
182 const char *name, unsigned int len,
183 struct svc_export **exp_ret, struct dentry **dentry_ret)
185 struct svc_export *exp;
186 struct dentry *dparent;
187 struct dentry *dentry;
190 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
192 dparent = fhp->fh_dentry;
193 exp = exp_get(fhp->fh_export);
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name, len)) {
198 dentry = dget(dparent);
199 else if (dparent != exp->ex_path.dentry)
200 dentry = dget_parent(dparent);
201 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
202 dentry = dget(dparent); /* .. == . just like at / */
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
211 * In the nfsd4_open() case, this may be held across
212 * subsequent open and delegation acquisition which may
213 * need to take the child's i_mutex:
215 fh_lock_nested(fhp, I_MUTEX_PARENT);
216 dentry = lookup_one_len(name, dparent, len);
217 host_err = PTR_ERR(dentry);
221 * check if we have crossed a mount point ...
223 if (nfsd_mountpoint(dentry, exp)) {
224 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
230 *dentry_ret = dentry;
236 return nfserrno(host_err);
240 * Look up one component of a pathname.
241 * N.B. After this call _both_ fhp and resfh need an fh_put
243 * If the lookup would cross a mountpoint, and the mounted filesystem
244 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
245 * accepted as it stands and the mounted directory is
246 * returned. Otherwise the covered directory is returned.
247 * NOTE: this mountpoint crossing is not supported properly by all
248 * clients and is explicitly disallowed for NFSv3
249 * NeilBrown <neilb@cse.unsw.edu.au>
252 nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
253 unsigned int len, struct svc_fh *resfh)
255 struct svc_export *exp;
256 struct dentry *dentry;
259 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
262 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
265 err = check_nfsd_access(exp, rqstp);
269 * Note: we compose the file handle now, but as the
270 * dentry may be negative, it may need to be updated.
272 err = fh_compose(resfh, exp, dentry, fhp);
273 if (!err && d_really_is_negative(dentry))
282 * Commit metadata changes to stable storage.
285 commit_metadata(struct svc_fh *fhp)
287 struct inode *inode = d_inode(fhp->fh_dentry);
288 const struct export_operations *export_ops = inode->i_sb->s_export_op;
290 if (!EX_ISSYNC(fhp->fh_export))
293 if (export_ops->commit_metadata)
294 return export_ops->commit_metadata(inode);
295 return sync_inode_metadata(inode, 1);
299 * Go over the attributes and take care of the small differences between
300 * NFS semantics and what Linux expects.
303 nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
305 /* sanitize the mode change */
306 if (iap->ia_valid & ATTR_MODE) {
307 iap->ia_mode &= S_IALLUGO;
308 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
311 /* Revoke setuid/setgid on chown */
312 if (!S_ISDIR(inode->i_mode) &&
313 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
314 iap->ia_valid |= ATTR_KILL_PRIV;
315 if (iap->ia_valid & ATTR_MODE) {
316 /* we're setting mode too, just clear the s*id bits */
317 iap->ia_mode &= ~S_ISUID;
318 if (iap->ia_mode & S_IXGRP)
319 iap->ia_mode &= ~S_ISGID;
321 /* set ATTR_KILL_* bits and let VFS handle it */
322 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
328 nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
331 struct inode *inode = d_inode(fhp->fh_dentry);
334 if (iap->ia_size < inode->i_size) {
337 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
338 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
343 host_err = get_write_access(inode);
347 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
349 goto out_put_write_access;
352 out_put_write_access:
353 put_write_access(inode);
355 return nfserrno(host_err);
359 * Set various file attributes. After this call fhp needs an fh_put.
362 nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
363 int check_guard, time_t guardtime)
365 struct dentry *dentry;
367 int accmode = NFSD_MAY_SATTR;
371 bool get_write_count;
372 bool size_change = (iap->ia_valid & ATTR_SIZE);
374 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
375 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
376 if (iap->ia_valid & ATTR_SIZE)
379 /* Callers that do fh_verify should do the fh_want_write: */
380 get_write_count = !fhp->fh_dentry;
383 err = fh_verify(rqstp, fhp, ftype, accmode);
386 if (get_write_count) {
387 host_err = fh_want_write(fhp);
392 dentry = fhp->fh_dentry;
393 inode = d_inode(dentry);
395 /* Ignore any mode updates on symlinks */
396 if (S_ISLNK(inode->i_mode))
397 iap->ia_valid &= ~ATTR_MODE;
402 nfsd_sanitize_attrs(inode, iap);
404 if (check_guard && guardtime != inode->i_ctime.tv_sec)
405 return nfserr_notsync;
408 * The size case is special, it changes the file in addition to the
409 * attributes, and file systems don't expect it to be mixed with
410 * "random" attribute changes. We thus split out the size change
411 * into a separate call to ->setattr, and do the rest as a separate
415 err = nfsd_get_write_access(rqstp, fhp, iap);
423 * RFC5661, Section 18.30.4:
424 * Changing the size of a file with SETATTR indirectly
425 * changes the time_modify and change attributes.
427 * (and similar for the older RFCs)
429 struct iattr size_attr = {
430 .ia_valid = ATTR_SIZE | ATTR_CTIME | ATTR_MTIME,
431 .ia_size = iap->ia_size,
434 host_err = notify_change(dentry, &size_attr, NULL);
437 iap->ia_valid &= ~ATTR_SIZE;
440 * Avoid the additional setattr call below if the only other
441 * attribute that the client sends is the mtime, as we update
442 * it as part of the size change above.
444 if ((iap->ia_valid & ~ATTR_MTIME) == 0)
448 iap->ia_valid |= ATTR_CTIME;
449 host_err = notify_change(dentry, iap, NULL);
454 put_write_access(inode);
457 host_err = commit_metadata(fhp);
458 return nfserrno(host_err);
461 #if defined(CONFIG_NFSD_V4)
463 * NFS junction information is stored in an extended attribute.
465 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
468 * nfsd4_is_junction - Test if an object could be an NFS junction
470 * @dentry: object to test
472 * Returns 1 if "dentry" appears to contain NFS junction information.
473 * Otherwise 0 is returned.
475 int nfsd4_is_junction(struct dentry *dentry)
477 struct inode *inode = d_inode(dentry);
481 if (inode->i_mode & S_IXUGO)
483 if (!(inode->i_mode & S_ISVTX))
485 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
489 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
490 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
491 struct xdr_netobj *label)
495 struct dentry *dentry;
497 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
501 dentry = fhp->fh_dentry;
503 mutex_lock(&d_inode(dentry)->i_mutex);
504 host_error = security_inode_setsecctx(dentry, label->data, label->len);
505 mutex_unlock(&d_inode(dentry)->i_mutex);
506 return nfserrno(host_error);
509 __be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
510 struct xdr_netobj *label)
512 return nfserr_notsupp;
516 __be32 nfsd4_vfs_fallocate(struct svc_rqst *rqstp, struct svc_fh *fhp,
517 struct file *file, loff_t offset, loff_t len,
522 if (!S_ISREG(file_inode(file)->i_mode))
525 error = vfs_fallocate(file, flags, offset, len);
527 error = commit_metadata(fhp);
529 return nfserrno(error);
531 #endif /* defined(CONFIG_NFSD_V4) */
533 #ifdef CONFIG_NFSD_V3
535 * Check server access rights to a file system object
541 static struct accessmap nfs3_regaccess[] = {
542 { NFS3_ACCESS_READ, NFSD_MAY_READ },
543 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
544 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
545 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
550 static struct accessmap nfs3_diraccess[] = {
551 { NFS3_ACCESS_READ, NFSD_MAY_READ },
552 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
553 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
554 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
555 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
560 static struct accessmap nfs3_anyaccess[] = {
561 /* Some clients - Solaris 2.6 at least, make an access call
562 * to the server to check for access for things like /dev/null
563 * (which really, the server doesn't care about). So
564 * We provide simple access checking for them, looking
565 * mainly at mode bits, and we make sure to ignore read-only
568 { NFS3_ACCESS_READ, NFSD_MAY_READ },
569 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
570 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
571 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
577 nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
579 struct accessmap *map;
580 struct svc_export *export;
581 struct dentry *dentry;
582 u32 query, result = 0, sresult = 0;
585 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
589 export = fhp->fh_export;
590 dentry = fhp->fh_dentry;
592 if (d_is_reg(dentry))
593 map = nfs3_regaccess;
594 else if (d_is_dir(dentry))
595 map = nfs3_diraccess;
597 map = nfs3_anyaccess;
601 for (; map->access; map++) {
602 if (map->access & query) {
605 sresult |= map->access;
607 err2 = nfsd_permission(rqstp, export, dentry, map->how);
610 result |= map->access;
613 /* the following error codes just mean the access was not allowed,
614 * rather than an error occurred */
618 /* simply don't "or" in the access bit. */
628 *supported = sresult;
633 #endif /* CONFIG_NFSD_V3 */
635 static int nfsd_open_break_lease(struct inode *inode, int access)
639 if (access & NFSD_MAY_NOT_BREAK_LEASE)
641 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
642 return break_lease(inode, mode | O_NONBLOCK);
646 * Open an existing file or directory.
647 * The may_flags argument indicates the type of open (read/write/lock)
648 * and additional flags.
649 * N.B. After this call fhp needs an fh_put
652 nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
653 int may_flags, struct file **filp)
658 int flags = O_RDONLY|O_LARGEFILE;
662 validate_process_creds();
665 * If we get here, then the client has already done an "open",
666 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
667 * in case a chmod has now revoked permission.
669 * Arguably we should also allow the owner override for
670 * directories, but we never have and it doesn't seem to have
671 * caused anyone a problem. If we were to change this, note
672 * also that our filldir callbacks would need a variant of
673 * lookup_one_len that doesn't check permissions.
676 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
677 err = fh_verify(rqstp, fhp, type, may_flags);
681 path.mnt = fhp->fh_export->ex_path.mnt;
682 path.dentry = fhp->fh_dentry;
683 inode = d_inode(path.dentry);
685 /* Disallow write access to files with the append-only bit set
686 * or any access when mandatory locking enabled
689 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
692 * We must ignore files (but only files) which might have mandatory
693 * locks on them because there is no way to know if the accesser has
696 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
702 host_err = nfsd_open_break_lease(inode, may_flags);
703 if (host_err) /* NOMEM or WOULDBLOCK */
706 if (may_flags & NFSD_MAY_WRITE) {
707 if (may_flags & NFSD_MAY_READ)
708 flags = O_RDWR|O_LARGEFILE;
710 flags = O_WRONLY|O_LARGEFILE;
713 file = dentry_open(&path, flags, current_cred());
715 host_err = PTR_ERR(file);
719 host_err = ima_file_check(file, may_flags, 0);
725 if (may_flags & NFSD_MAY_64BIT_COOKIE)
726 file->f_mode |= FMODE_64BITHASH;
728 file->f_mode |= FMODE_32BITHASH;
732 err = nfserrno(host_err);
734 validate_process_creds();
739 nfsd_init_raparms(struct file *file)
741 struct inode *inode = file_inode(file);
742 dev_t dev = inode->i_sb->s_dev;
743 ino_t ino = inode->i_ino;
744 struct raparms *ra, **rap, **frap = NULL;
747 struct raparm_hbucket *rab;
749 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
750 rab = &raparm_hash[hash];
752 spin_lock(&rab->pb_lock);
753 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
754 if (ra->p_ino == ino && ra->p_dev == dev)
757 if (ra->p_count == 0)
760 depth = nfsdstats.ra_size;
762 spin_unlock(&rab->pb_lock);
772 if (rap != &rab->pb_head) {
774 ra->p_next = rab->pb_head;
778 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
779 spin_unlock(&rab->pb_lock);
782 file->f_ra = ra->p_ra;
786 void nfsd_put_raparams(struct file *file, struct raparms *ra)
788 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
790 spin_lock(&rab->pb_lock);
791 ra->p_ra = file->f_ra;
794 spin_unlock(&rab->pb_lock);
798 * Grab and keep cached pages associated with a file in the svc_rqst
799 * so that they can be passed to the network sendmsg/sendpage routines
800 * directly. They will be released after the sending has completed.
803 nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
804 struct splice_desc *sd)
806 struct svc_rqst *rqstp = sd->u.data;
807 struct page **pp = rqstp->rq_next_page;
808 struct page *page = buf->page;
813 if (rqstp->rq_res.page_len == 0) {
815 put_page(*rqstp->rq_next_page);
816 *(rqstp->rq_next_page++) = page;
817 rqstp->rq_res.page_base = buf->offset;
818 rqstp->rq_res.page_len = size;
819 } else if (page != pp[-1]) {
821 if (*rqstp->rq_next_page)
822 put_page(*rqstp->rq_next_page);
823 *(rqstp->rq_next_page++) = page;
824 rqstp->rq_res.page_len += size;
826 rqstp->rq_res.page_len += size;
831 static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
832 struct splice_desc *sd)
834 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
838 nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
841 nfsdstats.io_read += host_err;
843 fsnotify_access(file);
846 return nfserrno(host_err);
849 __be32 nfsd_splice_read(struct svc_rqst *rqstp,
850 struct file *file, loff_t offset, unsigned long *count)
852 struct splice_desc sd = {
860 rqstp->rq_next_page = rqstp->rq_respages + 1;
861 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
862 return nfsd_finish_read(file, count, host_err);
865 __be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
866 unsigned long *count)
873 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
875 return nfsd_finish_read(file, count, host_err);
879 nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
880 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
882 if (file->f_op->splice_read && test_bit(RQ_SPLICE_OK, &rqstp->rq_flags))
883 return nfsd_splice_read(rqstp, file, offset, count);
885 return nfsd_readv(file, offset, vec, vlen, count);
889 * Gathered writes: If another process is currently writing to the file,
890 * there's a high chance this is another nfsd (triggered by a bulk write
891 * from a client's biod). Rather than syncing the file with each write
892 * request, we sleep for 10 msec.
894 * I don't know if this roughly approximates C. Juszak's idea of
895 * gathered writes, but it's a nice and simple solution (IMHO), and it
898 * Note: we do this only in the NFSv2 case, since v3 and higher have a
899 * better tool (separate unstable writes and commits) for solving this
902 static int wait_for_concurrent_writes(struct file *file)
904 struct inode *inode = file_inode(file);
905 static ino_t last_ino;
906 static dev_t last_dev;
909 if (atomic_read(&inode->i_writecount) > 1
910 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
911 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
913 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
916 if (inode->i_state & I_DIRTY) {
917 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
918 err = vfs_fsync(file, 0);
920 last_ino = inode->i_ino;
921 last_dev = inode->i_sb->s_dev;
926 nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
927 loff_t offset, struct kvec *vec, int vlen,
928 unsigned long *cnt, int *stablep)
930 struct svc_export *exp;
935 int stable = *stablep;
938 loff_t end = LLONG_MAX;
939 unsigned int pflags = current->flags;
941 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
943 * We want less throttling in balance_dirty_pages()
944 * and shrink_inactive_list() so that nfs to
945 * localhost doesn't cause nfsd to lock up due to all
946 * the client's dirty pages or its congested queue.
948 current->flags |= PF_LESS_THROTTLE;
950 inode = file_inode(file);
951 exp = fhp->fh_export;
953 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
958 /* Write the data. */
959 oldfs = get_fs(); set_fs(KERNEL_DS);
960 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
965 nfsdstats.io_write += host_err;
966 fsnotify_modify(file);
970 host_err = wait_for_concurrent_writes(file);
973 end = offset + *cnt - 1;
974 host_err = vfs_fsync_range(file, offset, end, 0);
979 dprintk("nfsd: write complete host_err=%d\n", host_err);
983 err = nfserrno(host_err);
984 if (test_bit(RQ_LOCAL, &rqstp->rq_flags))
985 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
990 * Read data from a file. count must contain the requested read count
991 * on entry. On return, *count contains the number of bytes actually read.
992 * N.B. After this call fhp needs an fh_put
994 __be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
995 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1001 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
1005 ra = nfsd_init_raparms(file);
1006 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1008 nfsd_put_raparams(file, ra);
1015 * Write data to a file.
1016 * The stable flag requests synchronous writes.
1017 * N.B. After this call fhp needs an fh_put
1020 nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
1021 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
1027 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
1028 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
1031 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1034 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
1039 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1047 #ifdef CONFIG_NFSD_V3
1049 * Commit all pending writes to stable storage.
1051 * Note: we only guarantee that data that lies within the range specified
1052 * by the 'offset' and 'count' parameters will be synced.
1054 * Unfortunately we cannot lock the file to make sure we return full WCC
1055 * data to the client, as locking happens lower down in the filesystem.
1058 nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1059 loff_t offset, unsigned long count)
1062 loff_t end = LLONG_MAX;
1063 __be32 err = nfserr_inval;
1068 end = offset + (loff_t)count - 1;
1073 err = nfsd_open(rqstp, fhp, S_IFREG,
1074 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
1077 if (EX_ISSYNC(fhp->fh_export)) {
1078 int err2 = vfs_fsync_range(file, offset, end, 0);
1080 if (err2 != -EINVAL)
1081 err = nfserrno(err2);
1083 err = nfserr_notsupp;
1090 #endif /* CONFIG_NFSD_V3 */
1093 nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1097 * Mode has already been set earlier in create:
1099 iap->ia_valid &= ~ATTR_MODE;
1101 * Setting uid/gid works only for root. Irix appears to
1102 * send along the gid on create when it tries to implement
1103 * setgid directories via NFS:
1105 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
1106 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1108 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
1109 /* Callers expect file metadata to be committed here */
1110 return nfserrno(commit_metadata(resfhp));
1113 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1114 * setting size to 0 may fail for some specific file systems by the permission
1115 * checking which requires WRITE permission but the mode is 000.
1116 * we ignore the resizing(to 0) on the just new created file, since the size is
1117 * 0 after file created.
1119 * call this only after vfs_create() is called.
1122 nfsd_check_ignore_resizing(struct iattr *iap)
1124 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1125 iap->ia_valid &= ~ATTR_SIZE;
1129 * Create a file (regular, directory, device, fifo); UNIX sockets
1130 * not yet implemented.
1131 * If the response fh has been verified, the parent directory should
1132 * already be locked. Note that the parent directory is left locked.
1134 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1137 nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1138 char *fname, int flen, struct iattr *iap,
1139 int type, dev_t rdev, struct svc_fh *resfhp)
1141 struct dentry *dentry, *dchild = NULL;
1151 if (isdotent(fname, flen))
1154 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1158 dentry = fhp->fh_dentry;
1159 dirp = d_inode(dentry);
1161 err = nfserr_notdir;
1162 if (!dirp->i_op->lookup)
1165 * Check whether the response file handle has been verified yet.
1166 * If it has, the parent directory should already be locked.
1168 if (!resfhp->fh_dentry) {
1169 host_err = fh_want_write(fhp);
1173 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1174 fh_lock_nested(fhp, I_MUTEX_PARENT);
1175 dchild = lookup_one_len(fname, dentry, flen);
1176 host_err = PTR_ERR(dchild);
1179 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1183 /* called from nfsd_proc_create */
1184 dchild = dget(resfhp->fh_dentry);
1185 if (!fhp->fh_locked) {
1186 /* not actually possible */
1188 "nfsd_create: parent %pd2 not locked!\n",
1195 * Make sure the child dentry is still negative ...
1198 if (d_really_is_positive(dchild)) {
1199 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1204 if (!(iap->ia_valid & ATTR_MODE))
1206 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1209 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1210 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1216 * Get the dir op function pointer.
1222 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1224 nfsd_check_ignore_resizing(iap);
1227 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
1233 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
1239 err = nfsd_create_setattr(rqstp, resfhp, iap);
1242 * nfsd_create_setattr already committed the child. Transactional
1243 * filesystems had a chance to commit changes for both parent and
1244 * child * simultaneously making the following commit_metadata a
1247 err2 = nfserrno(commit_metadata(fhp));
1251 * Update the file handle to get the new inode info.
1254 err = fh_update(resfhp);
1256 if (dchild && !IS_ERR(dchild))
1261 err = nfserrno(host_err);
1265 #ifdef CONFIG_NFSD_V3
1268 * NFSv3 and NFSv4 version of nfsd_create
1271 do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1272 char *fname, int flen, struct iattr *iap,
1273 struct svc_fh *resfhp, int createmode, u32 *verifier,
1274 bool *truncp, bool *created)
1276 struct dentry *dentry, *dchild = NULL;
1280 __u32 v_mtime=0, v_atime=0;
1286 if (isdotent(fname, flen))
1288 if (!(iap->ia_valid & ATTR_MODE))
1290 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
1294 dentry = fhp->fh_dentry;
1295 dirp = d_inode(dentry);
1297 /* Get all the sanity checks out of the way before
1298 * we lock the parent. */
1299 err = nfserr_notdir;
1300 if (!dirp->i_op->lookup)
1303 host_err = fh_want_write(fhp);
1307 fh_lock_nested(fhp, I_MUTEX_PARENT);
1310 * Compose the response file handle.
1312 dchild = lookup_one_len(fname, dentry, flen);
1313 host_err = PTR_ERR(dchild);
1317 /* If file doesn't exist, check for permissions to create one */
1318 if (d_really_is_negative(dchild)) {
1319 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1324 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1328 if (nfsd_create_is_exclusive(createmode)) {
1329 /* solaris7 gets confused (bugid 4218508) if these have
1330 * the high bit set, so just clear the high bits. If this is
1331 * ever changed to use different attrs for storing the
1332 * verifier, then do_open_lookup() will also need to be fixed
1335 v_mtime = verifier[0]&0x7fffffff;
1336 v_atime = verifier[1]&0x7fffffff;
1339 if (d_really_is_positive(dchild)) {
1342 switch (createmode) {
1343 case NFS3_CREATE_UNCHECKED:
1344 if (! d_is_reg(dchild))
1347 /* in nfsv4, we need to treat this case a little
1348 * differently. we don't want to truncate the
1349 * file now; this would be wrong if the OPEN
1350 * fails for some other reason. furthermore,
1351 * if the size is nonzero, we should ignore it
1352 * according to spec!
1354 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1357 iap->ia_valid &= ATTR_SIZE;
1361 case NFS3_CREATE_EXCLUSIVE:
1362 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1363 && d_inode(dchild)->i_atime.tv_sec == v_atime
1364 && d_inode(dchild)->i_size == 0 ) {
1369 case NFS4_CREATE_EXCLUSIVE4_1:
1370 if ( d_inode(dchild)->i_mtime.tv_sec == v_mtime
1371 && d_inode(dchild)->i_atime.tv_sec == v_atime
1372 && d_inode(dchild)->i_size == 0 ) {
1378 case NFS3_CREATE_GUARDED:
1385 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
1393 nfsd_check_ignore_resizing(iap);
1395 if (nfsd_create_is_exclusive(createmode)) {
1396 /* Cram the verifier into atime/mtime */
1397 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
1398 | ATTR_MTIME_SET|ATTR_ATIME_SET;
1399 /* XXX someone who knows this better please fix it for nsec */
1400 iap->ia_mtime.tv_sec = v_mtime;
1401 iap->ia_atime.tv_sec = v_atime;
1402 iap->ia_mtime.tv_nsec = 0;
1403 iap->ia_atime.tv_nsec = 0;
1407 err = nfsd_create_setattr(rqstp, resfhp, iap);
1410 * nfsd_create_setattr already committed the child
1411 * (and possibly also the parent).
1414 err = nfserrno(commit_metadata(fhp));
1417 * Update the filehandle to get the new inode info.
1420 err = fh_update(resfhp);
1424 if (dchild && !IS_ERR(dchild))
1430 err = nfserrno(host_err);
1433 #endif /* CONFIG_NFSD_V3 */
1436 * Read a symlink. On entry, *lenp must contain the maximum path length that
1437 * fits into the buffer. On return, it contains the true length.
1438 * N.B. After this call fhp needs an fh_put
1441 nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1443 struct inode *inode;
1449 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
1453 path.mnt = fhp->fh_export->ex_path.mnt;
1454 path.dentry = fhp->fh_dentry;
1455 inode = d_inode(path.dentry);
1458 if (!inode->i_op->readlink)
1462 /* N.B. Why does this call need a get_fs()??
1463 * Remove the set_fs and watch the fireworks:-) --okir
1466 oldfs = get_fs(); set_fs(KERNEL_DS);
1467 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
1478 err = nfserrno(host_err);
1483 * Create a symlink and look up its inode
1484 * N.B. After this call _both_ fhp and resfhp need an fh_put
1487 nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1488 char *fname, int flen,
1490 struct svc_fh *resfhp)
1492 struct dentry *dentry, *dnew;
1497 if (!flen || path[0] == '\0')
1500 if (isdotent(fname, flen))
1503 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1507 host_err = fh_want_write(fhp);
1512 dentry = fhp->fh_dentry;
1513 dnew = lookup_one_len(fname, dentry, flen);
1514 host_err = PTR_ERR(dnew);
1518 host_err = vfs_symlink(d_inode(dentry), dnew, path);
1519 err = nfserrno(host_err);
1521 err = nfserrno(commit_metadata(fhp));
1526 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1528 if (err==0) err = cerr;
1533 err = nfserrno(host_err);
1539 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1542 nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1543 char *name, int len, struct svc_fh *tfhp)
1545 struct dentry *ddir, *dnew, *dold;
1550 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
1553 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
1557 if (d_is_dir(tfhp->fh_dentry))
1563 if (isdotent(name, len))
1566 host_err = fh_want_write(tfhp);
1568 err = nfserrno(host_err);
1572 fh_lock_nested(ffhp, I_MUTEX_PARENT);
1573 ddir = ffhp->fh_dentry;
1574 dirp = d_inode(ddir);
1576 dnew = lookup_one_len(name, ddir, len);
1577 host_err = PTR_ERR(dnew);
1581 dold = tfhp->fh_dentry;
1584 if (d_really_is_negative(dold))
1586 host_err = vfs_link(dold, dirp, dnew, NULL);
1588 err = nfserrno(commit_metadata(ffhp));
1590 err = nfserrno(commit_metadata(tfhp));
1592 if (host_err == -EXDEV && rqstp->rq_vers == 2)
1595 err = nfserrno(host_err);
1601 fh_drop_write(tfhp);
1606 err = nfserrno(host_err);
1612 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1615 nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1616 struct svc_fh *tfhp, char *tname, int tlen)
1618 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1619 struct inode *fdir, *tdir;
1623 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
1626 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
1630 fdentry = ffhp->fh_dentry;
1631 fdir = d_inode(fdentry);
1633 tdentry = tfhp->fh_dentry;
1634 tdir = d_inode(tdentry);
1637 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1640 host_err = fh_want_write(ffhp);
1642 err = nfserrno(host_err);
1646 /* cannot use fh_lock as we need deadlock protective ordering
1647 * so do it by hand */
1648 trap = lock_rename(tdentry, fdentry);
1649 ffhp->fh_locked = tfhp->fh_locked = true;
1653 odentry = lookup_one_len(fname, fdentry, flen);
1654 host_err = PTR_ERR(odentry);
1655 if (IS_ERR(odentry))
1659 if (d_really_is_negative(odentry))
1662 if (odentry == trap)
1665 ndentry = lookup_one_len(tname, tdentry, tlen);
1666 host_err = PTR_ERR(ndentry);
1667 if (IS_ERR(ndentry))
1669 host_err = -ENOTEMPTY;
1670 if (ndentry == trap)
1674 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1676 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1679 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
1681 host_err = commit_metadata(tfhp);
1683 host_err = commit_metadata(ffhp);
1690 err = nfserrno(host_err);
1692 * We cannot rely on fh_unlock on the two filehandles,
1693 * as that would do the wrong thing if the two directories
1694 * were the same, so again we do it by hand.
1696 fill_post_wcc(ffhp);
1697 fill_post_wcc(tfhp);
1698 unlock_rename(tdentry, fdentry);
1699 ffhp->fh_locked = tfhp->fh_locked = false;
1700 fh_drop_write(ffhp);
1707 * Unlink a file or directory
1708 * N.B. After this call fhp needs an fh_put
1711 nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1712 char *fname, int flen)
1714 struct dentry *dentry, *rdentry;
1720 if (!flen || isdotent(fname, flen))
1722 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
1726 host_err = fh_want_write(fhp);
1730 fh_lock_nested(fhp, I_MUTEX_PARENT);
1731 dentry = fhp->fh_dentry;
1732 dirp = d_inode(dentry);
1734 rdentry = lookup_one_len(fname, dentry, flen);
1735 host_err = PTR_ERR(rdentry);
1736 if (IS_ERR(rdentry))
1739 if (d_really_is_negative(rdentry)) {
1746 type = d_inode(rdentry)->i_mode & S_IFMT;
1748 if (type != S_IFDIR)
1749 host_err = vfs_unlink(dirp, rdentry, NULL);
1751 host_err = vfs_rmdir(dirp, rdentry);
1753 host_err = commit_metadata(fhp);
1757 err = nfserrno(host_err);
1763 * We do this buffering because we must not call back into the file
1764 * system's ->lookup() method from the filldir callback. That may well
1765 * deadlock a number of file systems.
1767 * This is based heavily on the implementation of same in XFS.
1769 struct buffered_dirent {
1773 unsigned int d_type;
1777 struct readdir_data {
1778 struct dir_context ctx;
1784 static int nfsd_buffered_filldir(struct dir_context *ctx, const char *name,
1785 int namlen, loff_t offset, u64 ino,
1786 unsigned int d_type)
1788 struct readdir_data *buf =
1789 container_of(ctx, struct readdir_data, ctx);
1790 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1791 unsigned int reclen;
1793 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
1794 if (buf->used + reclen > PAGE_SIZE) {
1799 de->namlen = namlen;
1800 de->offset = offset;
1802 de->d_type = d_type;
1803 memcpy(de->name, name, namlen);
1804 buf->used += reclen;
1809 static __be32 nfsd_buffered_readdir(struct file *file, nfsd_filldir_t func,
1810 struct readdir_cd *cdp, loff_t *offsetp)
1812 struct buffered_dirent *de;
1816 struct readdir_data buf = {
1817 .ctx.actor = nfsd_buffered_filldir,
1818 .dirent = (void *)__get_free_page(GFP_KERNEL)
1822 return nfserrno(-ENOMEM);
1827 struct inode *dir_inode = file_inode(file);
1828 unsigned int reclen;
1830 cdp->err = nfserr_eof; /* will be cleared on successful read */
1834 host_err = iterate_dir(file, &buf.ctx);
1847 * Various filldir functions may end up calling back into
1848 * lookup_one_len() and the file system's ->lookup() method.
1849 * These expect i_mutex to be held, as it would within readdir.
1851 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1855 de = (struct buffered_dirent *)buf.dirent;
1857 offset = de->offset;
1859 if (func(cdp, de->name, de->namlen, de->offset,
1860 de->ino, de->d_type))
1863 if (cdp->err != nfs_ok)
1866 reclen = ALIGN(sizeof(*de) + de->namlen,
1869 de = (struct buffered_dirent *)((char *)de + reclen);
1871 mutex_unlock(&dir_inode->i_mutex);
1872 if (size > 0) /* We bailed out early */
1875 offset = vfs_llseek(file, 0, SEEK_CUR);
1878 free_page((unsigned long)(buf.dirent));
1881 return nfserrno(host_err);
1888 * Read entries from a directory.
1889 * The NFSv3/4 verifier we ignore for now.
1892 nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
1893 struct readdir_cd *cdp, nfsd_filldir_t func)
1897 loff_t offset = *offsetp;
1898 int may_flags = NFSD_MAY_READ;
1900 /* NFSv2 only supports 32 bit cookies */
1901 if (rqstp->rq_vers > 2)
1902 may_flags |= NFSD_MAY_64BIT_COOKIE;
1904 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
1908 offset = vfs_llseek(file, offset, SEEK_SET);
1910 err = nfserrno((int)offset);
1914 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
1916 if (err == nfserr_eof || err == nfserr_toosmall)
1917 err = nfs_ok; /* can still be found in ->err */
1925 * Get file system stats
1926 * N.B. After this call fhp needs an fh_put
1929 nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
1933 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
1935 struct path path = {
1936 .mnt = fhp->fh_export->ex_path.mnt,
1937 .dentry = fhp->fh_dentry,
1939 if (vfs_statfs(&path, stat))
1945 static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
1947 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
1951 * Check for a user's access permissions to this inode.
1954 nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1955 struct dentry *dentry, int acc)
1957 struct inode *inode = d_inode(dentry);
1960 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
1963 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1965 (acc & NFSD_MAY_READ)? " read" : "",
1966 (acc & NFSD_MAY_WRITE)? " write" : "",
1967 (acc & NFSD_MAY_EXEC)? " exec" : "",
1968 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1969 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1970 (acc & NFSD_MAY_LOCK)? " lock" : "",
1971 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
1973 IS_IMMUTABLE(inode)? " immut" : "",
1974 IS_APPEND(inode)? " append" : "",
1975 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
1976 dprintk(" owner %d/%d user %d/%d\n",
1977 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
1980 /* Normally we reject any write/sattr etc access on a read-only file
1981 * system. But if it is IRIX doing check on write-access for a
1982 * device special file, we ignore rofs.
1984 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
1985 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
1986 if (exp_rdonly(rqstp, exp) ||
1987 __mnt_is_readonly(exp->ex_path.mnt))
1989 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
1992 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
1995 if (acc & NFSD_MAY_LOCK) {
1996 /* If we cannot rely on authentication in NLM requests,
1997 * just allow locks, otherwise require read permission, or
2000 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2003 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
2006 * The file owner always gets access permission for accesses that
2007 * would normally be checked at open time. This is to make
2008 * file access work even when the client has done a fchmod(fd, 0).
2010 * However, `cp foo bar' should fail nevertheless when bar is
2011 * readonly. A sensible way to do this might be to reject all
2012 * attempts to truncate a read-only file, because a creat() call
2013 * always implies file truncation.
2014 * ... but this isn't really fair. A process may reasonably call
2015 * ftruncate on an open file descriptor on a file with perm 000.
2016 * We must trust the client to do permission checking - using "ACCESS"
2019 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
2020 uid_eq(inode->i_uid, current_fsuid()))
2023 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2024 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
2026 /* Allow read access to binaries even when mode 111 */
2027 if (err == -EACCES && S_ISREG(inode->i_mode) &&
2028 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2029 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
2030 err = inode_permission(inode, MAY_EXEC);
2032 return err? nfserrno(err) : 0;
2036 nfsd_racache_shutdown(void)
2038 struct raparms *raparm, *last_raparm;
2041 dprintk("nfsd: freeing readahead buffers.\n");
2043 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2044 raparm = raparm_hash[i].pb_head;
2046 last_raparm = raparm;
2047 raparm = raparm->p_next;
2050 raparm_hash[i].pb_head = NULL;
2054 * Initialize readahead param cache
2057 nfsd_racache_init(int cache_size)
2062 struct raparms **raparm = NULL;
2065 if (raparm_hash[0].pb_head)
2067 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
2068 nperbucket = max(2, nperbucket);
2069 cache_size = nperbucket * RAPARM_HASH_SIZE;
2071 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
2073 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2074 spin_lock_init(&raparm_hash[i].pb_lock);
2076 raparm = &raparm_hash[i].pb_head;
2077 for (j = 0; j < nperbucket; j++) {
2078 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2081 raparm = &(*raparm)->p_next;
2086 nfsdstats.ra_size = cache_size;
2090 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2091 nfsd_racache_shutdown();