blob: f501a9b5c9df18f5ec7cc360d385aa0c9dfb8344 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * 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.
5 *
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.
11 *
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
14 */
15
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/fs.h>
17#include <linux/file.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020018#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/fcntl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/delay.h>
Robert Love0eeca282005-07-12 17:06:03 -040022#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/xattr.h>
Boaz Harrosh9a74af22009-12-03 20:30:56 +020025#include <linux/jhash.h>
26#include <linux/ima.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
Boaz Harrosh9a74af22009-12-03 20:30:56 +020028#include <asm/uaccess.h>
Ben Myersf5019122010-02-17 14:05:11 -060029#include <linux/exportfs.h>
30#include <linux/writeback.h>
David Quigley18032ca2013-05-02 13:19:10 -040031#include <linux/security.h>
Boaz Harrosh9a74af22009-12-03 20:30:56 +020032
33#ifdef CONFIG_NFSD_V3
34#include "xdr3.h"
35#endif /* CONFIG_NFSD_V3 */
36
Christoph Hellwig5be196e2006-01-09 20:51:55 -080037#ifdef CONFIG_NFSD_V4
J. Bruce Fields2ca72e12011-01-04 17:37:15 -050038#include "acl.h"
39#include "idmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#endif /* CONFIG_NFSD_V4 */
41
Boaz Harrosh9a74af22009-12-03 20:30:56 +020042#include "nfsd.h"
43#include "vfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#define NFSDDBG_FACILITY NFSDDBG_FILEOP
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
54 */
55struct raparms {
56 struct raparms *p_next;
57 unsigned int p_count;
58 ino_t p_ino;
59 dev_t p_dev;
60 int p_set;
61 struct file_ra_state p_ra;
Greg Banksfce14562006-10-04 02:15:49 -070062 unsigned int p_hindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063};
64
Greg Banksfce14562006-10-04 02:15:49 -070065struct raparm_hbucket {
66 struct raparms *pb_head;
67 spinlock_t pb_lock;
68} ____cacheline_aligned_in_smp;
69
Greg Banksfce14562006-10-04 02:15:49 -070070#define RAPARM_HASH_BITS 4
71#define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72#define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73static struct raparm_hbucket raparm_hash[RAPARM_HASH_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75/*
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * a mount point.
J.Bruce Fieldse0bb89e2006-12-13 00:35:25 -080078 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * or nfs_ok having possibly changed *dpp and *expp
80 */
81int
82nfsd_cross_mnt(struct svc_rqst *rqstp, struct dentry **dpp,
83 struct svc_export **expp)
84{
85 struct svc_export *exp = *expp, *exp2 = NULL;
86 struct dentry *dentry = *dpp;
Al Viro91c9fa82009-04-18 02:42:05 -040087 struct path path = {.mnt = mntget(exp->ex_path.mnt),
88 .dentry = dget(dentry)};
Al Viro6264d692006-10-19 23:28:58 -070089 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070090
Al Viro7cc90cc2011-03-18 09:04:20 -040091 err = follow_down(&path);
David Howellscc53ce52011-01-14 18:45:26 +000092 if (err < 0)
93 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Al Viro91c9fa82009-04-18 02:42:05 -040095 exp2 = rqst_exp_get_by_name(rqstp, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 if (IS_ERR(exp2)) {
J. Bruce Fields3b6cee7bc2009-10-25 21:18:19 -040097 err = PTR_ERR(exp2);
98 /*
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
103 * directory.
104 */
105 if (err == -ENOENT && !(exp->ex_flags & NFSEXP_V4ROOT))
106 err = 0;
Al Viro91c9fa82009-04-18 02:42:05 -0400107 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 goto out;
109 }
Steve Dickson3c394dd2009-09-09 15:02:40 -0400110 if (nfsd_v4client(rqstp) ||
111 (exp->ex_flags & NFSEXP_CROSSMOUNT) || EX_NOHIDE(exp2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 /* successfully crossed mount point */
Al Viro1644ccc2009-04-18 02:32:31 -0400113 /*
Al Viro91c9fa82009-04-18 02:42:05 -0400114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
Al Viro1644ccc2009-04-18 02:32:31 -0400118 */
Al Viro91c9fa82009-04-18 02:42:05 -0400119 *dpp = path.dentry;
120 path.dentry = dentry;
Al Viro1644ccc2009-04-18 02:32:31 -0400121 *expp = exp2;
Al Viro91c9fa82009-04-18 02:42:05 -0400122 exp2 = exp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 }
Al Viro91c9fa82009-04-18 02:42:05 -0400124 path_put(&path);
125 exp_put(exp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126out:
127 return err;
128}
129
J. Bruce Fields289ede42009-09-26 20:32:24 -0400130static void follow_to_parent(struct path *path)
131{
132 struct dentry *dp;
133
134 while (path->dentry == path->mnt->mnt_root && follow_up(path))
135 ;
136 dp = dget_parent(path->dentry);
137 dput(path->dentry);
138 path->dentry = dp;
139}
140
141static int nfsd_lookup_parent(struct svc_rqst *rqstp, struct dentry *dparent, struct svc_export **exp, struct dentry **dentryp)
142{
143 struct svc_export *exp2;
144 struct path path = {.mnt = mntget((*exp)->ex_path.mnt),
145 .dentry = dget(dparent)};
146
147 follow_to_parent(&path);
148
149 exp2 = rqst_exp_parent(rqstp, &path);
150 if (PTR_ERR(exp2) == -ENOENT) {
151 *dentryp = dget(dparent);
152 } else if (IS_ERR(exp2)) {
153 path_put(&path);
154 return PTR_ERR(exp2);
155 } else {
156 *dentryp = dget(path.dentry);
157 exp_put(*exp);
158 *exp = exp2;
159 }
160 path_put(&path);
161 return 0;
162}
163
J. Bruce Fields82ead7f2009-10-25 21:33:15 -0400164/*
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
167 */
J. Bruce Fields3227fa42009-10-25 21:43:01 -0400168int nfsd_mountpoint(struct dentry *dentry, struct svc_export *exp)
J. Bruce Fields82ead7f2009-10-25 21:33:15 -0400169{
170 if (d_mountpoint(dentry))
171 return 1;
Trond Myklebust11fcee022011-09-12 19:37:26 -0400172 if (nfsd4_is_junction(dentry))
173 return 1;
J. Bruce Fields82ead7f2009-10-25 21:33:15 -0400174 if (!(exp->ex_flags & NFSEXP_V4ROOT))
175 return 0;
176 return dentry->d_inode != NULL;
177}
178
Al Viro6264d692006-10-19 23:28:58 -0700179__be32
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700180nfsd_lookup_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400181 const char *name, unsigned int len,
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700182 struct svc_export **exp_ret, struct dentry **dentry_ret)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183{
184 struct svc_export *exp;
185 struct dentry *dparent;
186 struct dentry *dentry;
Al Viro6264d692006-10-19 23:28:58 -0700187 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp), len,name);
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 dparent = fhp->fh_dentry;
Kinglong Meebf18f162014-06-10 22:06:44 +0800192 exp = exp_get(fhp->fh_export);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 /* Lookup the name, but don't follow links */
195 if (isdotent(name, len)) {
196 if (len==1)
197 dentry = dget(dparent);
Jan Blunck54775492008-02-14 19:38:39 -0800198 else if (dparent != exp->ex_path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 dentry = dget_parent(dparent);
J. Bruce Fieldsfed83812009-09-26 16:53:01 -0400200 else if (!EX_NOHIDE(exp) && !nfsd_v4client(rqstp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 dentry = dget(dparent); /* .. == . just like at / */
202 else {
203 /* checking mountpoint crossing is very different when stepping up */
J. Bruce Fields289ede42009-09-26 20:32:24 -0400204 host_err = nfsd_lookup_parent(rqstp, dparent, &exp, &dentry);
205 if (host_err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto out_nfserr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
208 } else {
J. Bruce Fields43357232014-01-24 18:04:40 -0500209 /*
210 * In the nfsd4_open() case, this may be held across
211 * subsequent open and delegation acquisition which may
212 * need to take the child's i_mutex:
213 */
214 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 dentry = lookup_one_len(name, dparent, len);
Al Viro6264d692006-10-19 23:28:58 -0700216 host_err = PTR_ERR(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 if (IS_ERR(dentry))
218 goto out_nfserr;
219 /*
220 * check if we have crossed a mount point ...
221 */
J. Bruce Fields82ead7f2009-10-25 21:33:15 -0400222 if (nfsd_mountpoint(dentry, exp)) {
Al Viro6264d692006-10-19 23:28:58 -0700223 if ((host_err = nfsd_cross_mnt(rqstp, &dentry, &exp))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 dput(dentry);
225 goto out_nfserr;
226 }
227 }
228 }
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700229 *dentry_ret = dentry;
230 *exp_ret = exp;
231 return 0;
232
233out_nfserr:
234 exp_put(exp);
235 return nfserrno(host_err);
236}
237
238/*
239 * Look up one component of a pathname.
240 * N.B. After this call _both_ fhp and resfh need an fh_put
241 *
242 * If the lookup would cross a mountpoint, and the mounted filesystem
243 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
244 * accepted as it stands and the mounted directory is
245 * returned. Otherwise the covered directory is returned.
246 * NOTE: this mountpoint crossing is not supported properly by all
247 * clients and is explicitly disallowed for NFSv3
248 * NeilBrown <neilb@cse.unsw.edu.au>
249 */
250__be32
251nfsd_lookup(struct svc_rqst *rqstp, struct svc_fh *fhp, const char *name,
Chuck Lever5a022fc2007-11-01 16:57:09 -0400252 unsigned int len, struct svc_fh *resfh)
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700253{
254 struct svc_export *exp;
255 struct dentry *dentry;
256 __be32 err;
257
J. Bruce Fields29a78a32011-04-09 11:28:53 -0400258 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
259 if (err)
260 return err;
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700261 err = nfsd_lookup_dentry(rqstp, fhp, name, len, &exp, &dentry);
262 if (err)
263 return err;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700264 err = check_nfsd_access(exp, rqstp);
265 if (err)
266 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 /*
268 * Note: we compose the file handle now, but as the
269 * dentry may be negative, it may need to be updated.
270 */
271 err = fh_compose(resfh, exp, dentry, fhp);
272 if (!err && !dentry->d_inode)
273 err = nfserr_noent;
Andy Adamson32c1eb02007-07-17 04:04:48 -0700274out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 exp_put(exp);
277 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
Ben Myersf5019122010-02-17 14:05:11 -0600280/*
281 * Commit metadata changes to stable storage.
282 */
283static int
284commit_metadata(struct svc_fh *fhp)
285{
286 struct inode *inode = fhp->fh_dentry->d_inode;
287 const struct export_operations *export_ops = inode->i_sb->s_export_op;
Ben Myersf5019122010-02-17 14:05:11 -0600288
289 if (!EX_ISSYNC(fhp->fh_export))
290 return 0;
291
Christoph Hellwigc37650162010-10-06 10:48:20 +0200292 if (export_ops->commit_metadata)
293 return export_ops->commit_metadata(inode);
294 return sync_inode_metadata(inode, 1);
Ben Myersf5019122010-02-17 14:05:11 -0600295}
J. Bruce Fields6c0a6542007-07-17 04:04:47 -0700296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297/*
Christoph Hellwig818e5a22013-11-18 05:07:30 -0800298 * Go over the attributes and take care of the small differences between
299 * NFS semantics and what Linux expects.
300 */
301static void
302nfsd_sanitize_attrs(struct inode *inode, struct iattr *iap)
303{
304 /*
305 * NFSv2 does not differentiate between "set-[ac]time-to-now"
306 * which only requires access, and "set-[ac]time-to-X" which
307 * requires ownership.
308 * So if it looks like it might be "set both to the same time which
309 * is close to now", and if inode_change_ok fails, then we
310 * convert to "set to now" instead of "set to explicit time"
311 *
312 * We only call inode_change_ok as the last test as technically
313 * it is not an interface that we should be using.
314 */
315#define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
316#define MAX_TOUCH_TIME_ERROR (30*60)
317 if ((iap->ia_valid & BOTH_TIME_SET) == BOTH_TIME_SET &&
318 iap->ia_mtime.tv_sec == iap->ia_atime.tv_sec) {
319 /*
320 * Looks probable.
321 *
322 * Now just make sure time is in the right ballpark.
323 * Solaris, at least, doesn't seem to care what the time
324 * request is. We require it be within 30 minutes of now.
325 */
326 time_t delta = iap->ia_atime.tv_sec - get_seconds();
327 if (delta < 0)
328 delta = -delta;
329 if (delta < MAX_TOUCH_TIME_ERROR &&
330 inode_change_ok(inode, iap) != 0) {
331 /*
332 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
333 * This will cause notify_change to set these times
334 * to "now"
335 */
336 iap->ia_valid &= ~BOTH_TIME_SET;
337 }
338 }
339
340 /* sanitize the mode change */
341 if (iap->ia_valid & ATTR_MODE) {
342 iap->ia_mode &= S_IALLUGO;
343 iap->ia_mode |= (inode->i_mode & ~S_IALLUGO);
344 }
345
346 /* Revoke setuid/setgid on chown */
347 if (!S_ISDIR(inode->i_mode) &&
Stanislav Kholmanskikhc4fa6d72013-12-11 14:16:36 +0400348 ((iap->ia_valid & ATTR_UID) || (iap->ia_valid & ATTR_GID))) {
Christoph Hellwig818e5a22013-11-18 05:07:30 -0800349 iap->ia_valid |= ATTR_KILL_PRIV;
350 if (iap->ia_valid & ATTR_MODE) {
351 /* we're setting mode too, just clear the s*id bits */
352 iap->ia_mode &= ~S_ISUID;
353 if (iap->ia_mode & S_IXGRP)
354 iap->ia_mode &= ~S_ISGID;
355 } else {
356 /* set ATTR_KILL_* bits and let VFS handle it */
357 iap->ia_valid |= (ATTR_KILL_SUID | ATTR_KILL_SGID);
358 }
359 }
360}
361
362static __be32
363nfsd_get_write_access(struct svc_rqst *rqstp, struct svc_fh *fhp,
364 struct iattr *iap)
365{
366 struct inode *inode = fhp->fh_dentry->d_inode;
367 int host_err;
368
369 if (iap->ia_size < inode->i_size) {
370 __be32 err;
371
372 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
373 NFSD_MAY_TRUNC | NFSD_MAY_OWNER_OVERRIDE);
374 if (err)
375 return err;
376 }
377
378 host_err = get_write_access(inode);
379 if (host_err)
380 goto out_nfserrno;
381
382 host_err = locks_verify_truncate(inode, NULL, iap->ia_size);
383 if (host_err)
384 goto out_put_write_access;
385 return 0;
386
387out_put_write_access:
388 put_write_access(inode);
389out_nfserrno:
390 return nfserrno(host_err);
391}
392
393/*
394 * Set various file attributes. After this call fhp needs an fh_put.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 */
Al Viro6264d692006-10-19 23:28:58 -0700396__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397nfsd_setattr(struct svc_rqst *rqstp, struct svc_fh *fhp, struct iattr *iap,
398 int check_guard, time_t guardtime)
399{
400 struct dentry *dentry;
401 struct inode *inode;
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200402 int accmode = NFSD_MAY_SATTR;
Al Viro175a4eb2011-07-26 03:30:54 -0400403 umode_t ftype = 0;
Al Viro6264d692006-10-19 23:28:58 -0700404 __be32 err;
405 int host_err;
J. Bruce Fields9f67f182014-02-24 14:59:47 -0500406 bool get_write_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 int size_change = 0;
408
409 if (iap->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_SIZE))
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200410 accmode |= NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (iap->ia_valid & ATTR_SIZE)
412 ftype = S_IFREG;
413
J. Bruce Fields9f67f182014-02-24 14:59:47 -0500414 /* Callers that do fh_verify should do the fh_want_write: */
415 get_write_count = !fhp->fh_dentry;
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 /* Get inode */
418 err = fh_verify(rqstp, fhp, ftype, accmode);
NeilBrown15b7a1b2005-11-07 01:00:23 -0800419 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 goto out;
J. Bruce Fields9f67f182014-02-24 14:59:47 -0500421 if (get_write_count) {
422 host_err = fh_want_write(fhp);
423 if (host_err)
424 return nfserrno(host_err);
425 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 dentry = fhp->fh_dentry;
428 inode = dentry->d_inode;
429
NeilBrown15b7a1b2005-11-07 01:00:23 -0800430 /* Ignore any mode updates on symlinks */
431 if (S_ISLNK(inode->i_mode))
432 iap->ia_valid &= ~ATTR_MODE;
433
434 if (!iap->ia_valid)
435 goto out;
436
Christoph Hellwig818e5a22013-11-18 05:07:30 -0800437 nfsd_sanitize_attrs(inode, iap);
438
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000439 /*
Christoph Hellwig818e5a22013-11-18 05:07:30 -0800440 * The size case is special, it changes the file in addition to the
441 * attributes.
Christoph Hellwig9c85fca2007-03-07 15:26:25 +0000442 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (iap->ia_valid & ATTR_SIZE) {
Christoph Hellwig818e5a22013-11-18 05:07:30 -0800444 err = nfsd_get_write_access(rqstp, fhp, iap);
445 if (err)
446 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 size_change = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 }
449
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 iap->ia_valid |= ATTR_CTIME;
451
Christoph Hellwig987da472013-11-18 05:07:47 -0800452 if (check_guard && guardtime != inode->i_ctime.tv_sec) {
453 err = nfserr_notsync;
454 goto out_put_write_access;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 }
Christoph Hellwig987da472013-11-18 05:07:47 -0800456
Christoph Hellwig987da472013-11-18 05:07:47 -0800457 fh_lock(fhp);
458 host_err = notify_change(dentry, iap, NULL);
459 fh_unlock(fhp);
J. R. Okajima1406b912014-02-19 00:27:53 +0900460 err = nfserrno(host_err);
Christoph Hellwig987da472013-11-18 05:07:47 -0800461
Christoph Hellwig987da472013-11-18 05:07:47 -0800462out_put_write_access:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 if (size_change)
464 put_write_access(inode);
465 if (!err)
Jeff Layton722b6202014-07-03 07:54:19 -0400466 err = nfserrno(commit_metadata(fhp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467out:
468 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
Christoph Hellwig5be196e2006-01-09 20:51:55 -0800471#if defined(CONFIG_NFSD_V4)
Chuck Lever9b4146e2012-01-04 16:26:43 -0500472/*
473 * NFS junction information is stored in an extended attribute.
474 */
475#define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
476
477/**
478 * nfsd4_is_junction - Test if an object could be an NFS junction
479 *
480 * @dentry: object to test
481 *
482 * Returns 1 if "dentry" appears to contain NFS junction information.
483 * Otherwise 0 is returned.
484 */
Trond Myklebust11fcee022011-09-12 19:37:26 -0400485int nfsd4_is_junction(struct dentry *dentry)
486{
487 struct inode *inode = dentry->d_inode;
488
489 if (inode == NULL)
490 return 0;
491 if (inode->i_mode & S_IXUGO)
492 return 0;
493 if (!(inode->i_mode & S_ISVTX))
494 return 0;
Chuck Lever9b4146e2012-01-04 16:26:43 -0500495 if (vfs_getxattr(dentry, NFSD_JUNCTION_XATTR_NAME, NULL, 0) <= 0)
Trond Myklebust11fcee022011-09-12 19:37:26 -0400496 return 0;
497 return 1;
498}
David Quigley18032ca2013-05-02 13:19:10 -0400499#ifdef CONFIG_NFSD_V4_SECURITY_LABEL
500__be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
501 struct xdr_netobj *label)
502{
503 __be32 error;
504 int host_error;
505 struct dentry *dentry;
506
507 error = fh_verify(rqstp, fhp, 0 /* S_IFREG */, NFSD_MAY_SATTR);
508 if (error)
509 return error;
510
511 dentry = fhp->fh_dentry;
512
513 mutex_lock(&dentry->d_inode->i_mutex);
514 host_error = security_inode_setsecctx(dentry, label->data, label->len);
515 mutex_unlock(&dentry->d_inode->i_mutex);
516 return nfserrno(host_error);
517}
518#else
519__be32 nfsd4_set_nfs4_label(struct svc_rqst *rqstp, struct svc_fh *fhp,
520 struct xdr_netobj *label)
521{
522 return nfserr_notsupp;
523}
524#endif
525
J. Bruce Fields6a85d6c2010-07-06 12:39:12 -0400526#endif /* defined(CONFIG_NFSD_V4) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528#ifdef CONFIG_NFSD_V3
529/*
530 * Check server access rights to a file system object
531 */
532struct accessmap {
533 u32 access;
534 int how;
535};
536static struct accessmap nfs3_regaccess[] = {
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200537 { NFS3_ACCESS_READ, NFSD_MAY_READ },
538 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
539 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_TRUNC },
540 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 { 0, 0 }
543};
544
545static struct accessmap nfs3_diraccess[] = {
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200546 { NFS3_ACCESS_READ, NFSD_MAY_READ },
547 { NFS3_ACCESS_LOOKUP, NFSD_MAY_EXEC },
548 { NFS3_ACCESS_MODIFY, NFSD_MAY_EXEC|NFSD_MAY_WRITE|NFSD_MAY_TRUNC},
549 { NFS3_ACCESS_EXTEND, NFSD_MAY_EXEC|NFSD_MAY_WRITE },
550 { NFS3_ACCESS_DELETE, NFSD_MAY_REMOVE },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552 { 0, 0 }
553};
554
555static struct accessmap nfs3_anyaccess[] = {
556 /* Some clients - Solaris 2.6 at least, make an access call
557 * to the server to check for access for things like /dev/null
558 * (which really, the server doesn't care about). So
559 * We provide simple access checking for them, looking
560 * mainly at mode bits, and we make sure to ignore read-only
561 * filesystem checks
562 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200563 { NFS3_ACCESS_READ, NFSD_MAY_READ },
564 { NFS3_ACCESS_EXECUTE, NFSD_MAY_EXEC },
565 { NFS3_ACCESS_MODIFY, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
566 { NFS3_ACCESS_EXTEND, NFSD_MAY_WRITE|NFSD_MAY_LOCAL_ACCESS },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
568 { 0, 0 }
569};
570
Al Viro6264d692006-10-19 23:28:58 -0700571__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572nfsd_access(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *access, u32 *supported)
573{
574 struct accessmap *map;
575 struct svc_export *export;
576 struct dentry *dentry;
577 u32 query, result = 0, sresult = 0;
Al Viro6264d692006-10-19 23:28:58 -0700578 __be32 error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
Miklos Szeredi8837abc2008-06-16 13:20:29 +0200580 error = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (error)
582 goto out;
583
584 export = fhp->fh_export;
585 dentry = fhp->fh_dentry;
586
587 if (S_ISREG(dentry->d_inode->i_mode))
588 map = nfs3_regaccess;
589 else if (S_ISDIR(dentry->d_inode->i_mode))
590 map = nfs3_diraccess;
591 else
592 map = nfs3_anyaccess;
593
594
595 query = *access;
596 for (; map->access; map++) {
597 if (map->access & query) {
Al Viro6264d692006-10-19 23:28:58 -0700598 __be32 err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600 sresult |= map->access;
601
J. Bruce Fields0ec757d2007-07-17 04:04:48 -0700602 err2 = nfsd_permission(rqstp, export, dentry, map->how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 switch (err2) {
604 case nfs_ok:
605 result |= map->access;
606 break;
607
608 /* the following error codes just mean the access was not allowed,
609 * rather than an error occurred */
610 case nfserr_rofs:
611 case nfserr_acces:
612 case nfserr_perm:
613 /* simply don't "or" in the access bit. */
614 break;
615 default:
616 error = err2;
617 goto out;
618 }
619 }
620 }
621 *access = result;
622 if (supported)
623 *supported = sresult;
624
625 out:
626 return error;
627}
628#endif /* CONFIG_NFSD_V3 */
629
J. Bruce Fields105f4622011-06-07 11:50:23 -0400630static int nfsd_open_break_lease(struct inode *inode, int access)
631{
632 unsigned int mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
J. Bruce Fields105f4622011-06-07 11:50:23 -0400634 if (access & NFSD_MAY_NOT_BREAK_LEASE)
635 return 0;
636 mode = (access & NFSD_MAY_WRITE) ? O_WRONLY : O_RDONLY;
637 return break_lease(inode, mode | O_NONBLOCK);
638}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
640/*
641 * Open an existing file or directory.
Bernd Schubert999448a2012-03-18 22:44:49 -0400642 * The may_flags argument indicates the type of open (read/write/lock)
643 * and additional flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * N.B. After this call fhp needs an fh_put
645 */
Al Viro6264d692006-10-19 23:28:58 -0700646__be32
Al Viro175a4eb2011-07-26 03:30:54 -0400647nfsd_open(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type,
Bernd Schubert999448a2012-03-18 22:44:49 -0400648 int may_flags, struct file **filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Al Viro765927b2012-06-26 21:58:53 +0400650 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 struct inode *inode;
Al Viro6264d692006-10-19 23:28:58 -0700652 int flags = O_RDONLY|O_LARGEFILE;
653 __be32 err;
Jeff Layton91885252010-03-19 08:06:28 -0400654 int host_err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
David Howellse0e81732009-09-02 09:13:40 +0100656 validate_process_creds();
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 /*
659 * If we get here, then the client has already done an "open",
660 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
661 * in case a chmod has now revoked permission.
J. Bruce Fieldsd91d0b52012-06-06 12:12:57 -0400662 *
663 * Arguably we should also allow the owner override for
664 * directories, but we never have and it doesn't seem to have
665 * caused anyone a problem. If we were to change this, note
666 * also that our filldir callbacks would need a variant of
667 * lookup_one_len that doesn't check permissions.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 */
J. Bruce Fieldsd91d0b52012-06-06 12:12:57 -0400669 if (type == S_IFREG)
670 may_flags |= NFSD_MAY_OWNER_OVERRIDE;
671 err = fh_verify(rqstp, fhp, type, may_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (err)
673 goto out;
674
Al Viro765927b2012-06-26 21:58:53 +0400675 path.mnt = fhp->fh_export->ex_path.mnt;
676 path.dentry = fhp->fh_dentry;
677 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679 /* Disallow write access to files with the append-only bit set
680 * or any access when mandatory locking enabled
681 */
682 err = nfserr_perm;
Bernd Schubert999448a2012-03-18 22:44:49 -0400683 if (IS_APPEND(inode) && (may_flags & NFSD_MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 goto out;
J. Bruce Fields5e7fc432007-10-02 14:18:12 -0400685 /*
686 * We must ignore files (but only files) which might have mandatory
687 * locks on them because there is no way to know if the accesser has
688 * the lock.
689 */
690 if (S_ISREG((inode)->i_mode) && mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 goto out;
692
693 if (!inode->i_fop)
694 goto out;
695
Bernd Schubert999448a2012-03-18 22:44:49 -0400696 host_err = nfsd_open_break_lease(inode, may_flags);
Al Viro6264d692006-10-19 23:28:58 -0700697 if (host_err) /* NOMEM or WOULDBLOCK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 goto out_nfserr;
699
Bernd Schubert999448a2012-03-18 22:44:49 -0400700 if (may_flags & NFSD_MAY_WRITE) {
701 if (may_flags & NFSD_MAY_READ)
J. Bruce Fields9ecb6a02006-06-30 01:56:17 -0700702 flags = O_RDWR|O_LARGEFILE;
703 else
704 flags = O_WRONLY|O_LARGEFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 }
Al Viro765927b2012-06-26 21:58:53 +0400706 *filp = dentry_open(&path, flags, current_cred());
Harshula Jayasuriyae4daf1f2013-07-23 14:21:35 +1000707 if (IS_ERR(*filp)) {
Al Viro6264d692006-10-19 23:28:58 -0700708 host_err = PTR_ERR(*filp);
Harshula Jayasuriyae4daf1f2013-07-23 14:21:35 +1000709 *filp = NULL;
710 } else {
Bernd Schubert999448a2012-03-18 22:44:49 -0400711 host_err = ima_file_check(*filp, may_flags);
712
Bernd Schubert06effdb2012-03-18 22:44:50 -0400713 if (may_flags & NFSD_MAY_64BIT_COOKIE)
714 (*filp)->f_mode |= FMODE_64BITHASH;
715 else
716 (*filp)->f_mode |= FMODE_32BITHASH;
717 }
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700720 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721out:
David Howellse0e81732009-09-02 09:13:40 +0100722 validate_process_creds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 return err;
724}
725
726/*
727 * Close a file.
728 */
729void
730nfsd_close(struct file *filp)
731{
732 fput(filp);
733}
734
J. Bruce Fields9a8d2482009-01-06 13:37:03 -0500735/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 * Obtain the readahead parameters for the file
737 * specified by (dev, ino).
738 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
740static inline struct raparms *
741nfsd_get_raparms(dev_t dev, ino_t ino)
742{
743 struct raparms *ra, **rap, **frap = NULL;
744 int depth = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700745 unsigned int hash;
746 struct raparm_hbucket *rab;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Greg Banksfce14562006-10-04 02:15:49 -0700748 hash = jhash_2words(dev, ino, 0xfeedbeef) & RAPARM_HASH_MASK;
749 rab = &raparm_hash[hash];
750
751 spin_lock(&rab->pb_lock);
752 for (rap = &rab->pb_head; (ra = *rap); rap = &ra->p_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (ra->p_ino == ino && ra->p_dev == dev)
754 goto found;
755 depth++;
756 if (ra->p_count == 0)
757 frap = rap;
758 }
Konstantin Khorenko3aa6e0a2011-02-01 17:16:29 +0300759 depth = nfsdstats.ra_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 if (!frap) {
Greg Banksfce14562006-10-04 02:15:49 -0700761 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 return NULL;
763 }
764 rap = frap;
765 ra = *frap;
766 ra->p_dev = dev;
767 ra->p_ino = ino;
768 ra->p_set = 0;
Greg Banksfce14562006-10-04 02:15:49 -0700769 ra->p_hindex = hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770found:
Greg Banksfce14562006-10-04 02:15:49 -0700771 if (rap != &rab->pb_head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 *rap = ra->p_next;
Greg Banksfce14562006-10-04 02:15:49 -0700773 ra->p_next = rab->pb_head;
774 rab->pb_head = ra;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 }
776 ra->p_count++;
777 nfsdstats.ra_depth[depth*10/nfsdstats.ra_size]++;
Greg Banksfce14562006-10-04 02:15:49 -0700778 spin_unlock(&rab->pb_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 return ra;
780}
781
782/*
Jens Axboecf8208d2007-06-12 21:22:14 +0200783 * Grab and keep cached pages associated with a file in the svc_rqst
784 * so that they can be passed to the network sendmsg/sendpage routines
785 * directly. They will be released after the sending has completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 */
787static int
Jens Axboecf8208d2007-06-12 21:22:14 +0200788nfsd_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
789 struct splice_desc *sd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
Jens Axboecf8208d2007-06-12 21:22:14 +0200791 struct svc_rqst *rqstp = sd->u.data;
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500792 struct page **pp = rqstp->rq_next_page;
Jens Axboecf8208d2007-06-12 21:22:14 +0200793 struct page *page = buf->page;
794 size_t size;
Jens Axboecf8208d2007-06-12 21:22:14 +0200795
796 size = sd->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798 if (rqstp->rq_res.page_len == 0) {
799 get_page(page);
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500800 put_page(*rqstp->rq_next_page);
801 *(rqstp->rq_next_page++) = page;
Jens Axboecf8208d2007-06-12 21:22:14 +0200802 rqstp->rq_res.page_base = buf->offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 rqstp->rq_res.page_len = size;
NeilBrown44524352006-10-04 02:15:46 -0700804 } else if (page != pp[-1]) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 get_page(page);
J. Bruce Fieldsafc59402012-12-10 18:01:37 -0500806 if (*rqstp->rq_next_page)
807 put_page(*rqstp->rq_next_page);
808 *(rqstp->rq_next_page++) = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 rqstp->rq_res.page_len += size;
NeilBrown44524352006-10-04 02:15:46 -0700810 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 rqstp->rq_res.page_len += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return size;
814}
815
Jens Axboecf8208d2007-06-12 21:22:14 +0200816static int nfsd_direct_splice_actor(struct pipe_inode_info *pipe,
817 struct splice_desc *sd)
818{
819 return __splice_from_pipe(pipe, sd, nfsd_splice_actor);
820}
821
Jeff Laytone2afc812014-06-17 07:44:13 -0400822static __be32
823nfsd_finish_read(struct file *file, unsigned long *count, int host_err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824{
Al Viro6264d692006-10-19 23:28:58 -0700825 if (host_err >= 0) {
826 nfsdstats.io_read += host_err;
827 *count = host_err;
Eric Paris2a12a9d2009-12-17 21:24:21 -0500828 fsnotify_access(file);
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400829 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 } else
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400831 return nfserrno(host_err);
832}
833
Jeff Laytone2afc812014-06-17 07:44:13 -0400834__be32 nfsd_splice_read(struct svc_rqst *rqstp,
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400835 struct file *file, loff_t offset, unsigned long *count)
836{
837 struct splice_desc sd = {
838 .len = 0,
839 .total_len = *count,
840 .pos = offset,
841 .u.data = rqstp,
842 };
843 int host_err;
844
845 rqstp->rq_next_page = rqstp->rq_respages + 1;
846 host_err = splice_direct_to_actor(file, &sd, nfsd_direct_splice_actor);
847 return nfsd_finish_read(file, count, host_err);
848}
849
Jeff Laytone2afc812014-06-17 07:44:13 -0400850__be32 nfsd_readv(struct file *file, loff_t offset, struct kvec *vec, int vlen,
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400851 unsigned long *count)
852{
853 mm_segment_t oldfs;
854 int host_err;
855
856 oldfs = get_fs();
857 set_fs(KERNEL_DS);
858 host_err = vfs_readv(file, (struct iovec __user *)vec, vlen, &offset);
859 set_fs(oldfs);
860 return nfsd_finish_read(file, count, host_err);
861}
862
863static __be32
864nfsd_vfs_read(struct svc_rqst *rqstp, struct file *file,
865 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
866{
867 if (file->f_op->splice_read && rqstp->rq_splice_ok)
868 return nfsd_splice_read(rqstp, file, offset, count);
869 else
870 return nfsd_readv(file, offset, vec, vlen, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871}
872
J. Bruce Fieldsd911df72009-06-15 16:03:53 -0700873/*
874 * Gathered writes: If another process is currently writing to the file,
875 * there's a high chance this is another nfsd (triggered by a bulk write
876 * from a client's biod). Rather than syncing the file with each write
877 * request, we sleep for 10 msec.
878 *
879 * I don't know if this roughly approximates C. Juszak's idea of
880 * gathered writes, but it's a nice and simple solution (IMHO), and it
881 * seems to work:-)
882 *
883 * Note: we do this only in the NFSv2 case, since v3 and higher have a
884 * better tool (separate unstable writes and commits) for solving this
885 * problem.
886 */
887static int wait_for_concurrent_writes(struct file *file)
888{
Al Viro496ad9a2013-01-23 17:07:38 -0500889 struct inode *inode = file_inode(file);
J. Bruce Fieldsd911df72009-06-15 16:03:53 -0700890 static ino_t last_ino;
891 static dev_t last_dev;
892 int err = 0;
893
894 if (atomic_read(&inode->i_writecount) > 1
895 || (last_ino == inode->i_ino && last_dev == inode->i_sb->s_dev)) {
896 dprintk("nfsd: write defer %d\n", task_pid_nr(current));
897 msleep(10);
898 dprintk("nfsd: write resume %d\n", task_pid_nr(current));
899 }
900
901 if (inode->i_state & I_DIRTY) {
902 dprintk("nfsd: write sync %d\n", task_pid_nr(current));
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100903 err = vfs_fsync(file, 0);
J. Bruce Fieldsd911df72009-06-15 16:03:53 -0700904 }
905 last_ino = inode->i_ino;
906 last_dev = inode->i_sb->s_dev;
907 return err;
908}
909
Al Viro6264d692006-10-19 23:28:58 -0700910static __be32
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
912 loff_t offset, struct kvec *vec, int vlen,
David Shaw31dec252009-03-05 20:16:14 -0500913 unsigned long *cnt, int *stablep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
915 struct svc_export *exp;
916 struct dentry *dentry;
917 struct inode *inode;
918 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -0700919 __be32 err = 0;
920 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 int stable = *stablep;
Trond Myklebust48e03bc2009-06-05 12:35:15 -0400922 int use_wgather;
Kent Overstreete49dbbf2013-03-22 11:18:24 -0700923 loff_t pos = offset;
NeilBrown86584522014-05-12 11:22:47 +1000924 unsigned int pflags = current->flags;
925
926 if (rqstp->rq_local)
927 /*
928 * We want less throttling in balance_dirty_pages()
929 * and shrink_inactive_list() so that nfs to
930 * localhost doesn't cause nfsd to lock up due to all
931 * the client's dirty pages or its congested queue.
932 */
933 current->flags |= PF_LESS_THROTTLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Josef "Jeff" Sipek7eaa36e2006-12-08 02:36:41 -0800935 dentry = file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 inode = dentry->d_inode;
937 exp = fhp->fh_export;
938
Trond Myklebust48e03bc2009-06-05 12:35:15 -0400939 use_wgather = (rqstp->rq_vers == 2) && EX_WGATHER(exp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 if (!EX_ISSYNC(exp))
942 stable = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 /* Write the data. */
945 oldfs = get_fs(); set_fs(KERNEL_DS);
Kent Overstreete49dbbf2013-03-22 11:18:24 -0700946 host_err = vfs_writev(file, (struct iovec __user *)vec, vlen, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 set_fs(oldfs);
J. Bruce Fieldse4636d52009-06-15 19:07:13 -0700948 if (host_err < 0)
949 goto out_nfserr;
950 *cnt = host_err;
951 nfsdstats.io_write += host_err;
Eric Paris2a12a9d2009-12-17 21:24:21 -0500952 fsnotify_modify(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
J. Bruce Fieldsface1502012-10-26 16:12:31 -0400954 if (stable) {
955 if (use_wgather)
956 host_err = wait_for_concurrent_writes(file);
957 else
958 host_err = vfs_fsync_range(file, offset, offset+*cnt, 0);
959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
J. Bruce Fieldse4636d52009-06-15 19:07:13 -0700961out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -0700962 dprintk("nfsd: write complete host_err=%d\n", host_err);
Wei Yongjuna0d24b22009-05-19 12:03:15 +0800963 if (host_err >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 err = 0;
Wei Yongjuna0d24b22009-05-19 12:03:15 +0800965 else
Al Viro6264d692006-10-19 23:28:58 -0700966 err = nfserrno(host_err);
NeilBrown86584522014-05-12 11:22:47 +1000967 if (rqstp->rq_local)
968 tsk_restore_flags(current, pflags, PF_LESS_THROTTLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 return err;
970}
971
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400972__be32 nfsd_get_tmp_read_open(struct svc_rqst *rqstp, struct svc_fh *fhp,
973 struct file **file, struct raparms **ra)
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400974{
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400975 struct inode *inode;
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400976 __be32 err;
977
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400978 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, file);
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400979 if (err)
980 return err;
981
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400982 inode = file_inode(*file);
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400983
984 /* Get readahead parameters */
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400985 *ra = nfsd_get_raparms(inode->i_sb->s_dev, inode->i_ino);
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400986
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400987 if (*ra && (*ra)->p_set)
988 (*file)->f_ra = (*ra)->p_ra;
989 return nfs_ok;
990}
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400991
J. Bruce Fieldsdc976182014-03-18 17:01:51 -0400992void nfsd_put_tmp_read_open(struct file *file, struct raparms *ra)
993{
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -0400994 /* Write back readahead params */
995 if (ra) {
996 struct raparm_hbucket *rab = &raparm_hash[ra->p_hindex];
997 spin_lock(&rab->pb_lock);
998 ra->p_ra = file->f_ra;
999 ra->p_set = 1;
1000 ra->p_count--;
1001 spin_unlock(&rab->pb_lock);
1002 }
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -04001003 nfsd_close(file);
J. Bruce Fieldsdc976182014-03-18 17:01:51 -04001004}
1005
1006/*
1007 * Read data from a file. count must contain the requested read count
1008 * on entry. On return, *count contains the number of bytes actually read.
1009 * N.B. After this call fhp needs an fh_put
1010 */
1011__be32 nfsd_read(struct svc_rqst *rqstp, struct svc_fh *fhp,
1012 loff_t offset, struct kvec *vec, int vlen, unsigned long *count)
1013{
1014 struct file *file;
1015 struct raparms *ra;
1016 __be32 err;
1017
1018 err = nfsd_get_tmp_read_open(rqstp, fhp, &file, &ra);
1019 if (err)
1020 return err;
1021
1022 err = nfsd_vfs_read(rqstp, file, offset, vec, vlen, count);
1023
1024 nfsd_put_tmp_read_open(file, ra);
1025
J. Bruce Fieldsfa0a2122010-07-27 16:48:54 -04001026 return err;
1027}
1028
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029/*
1030 * Write data to a file.
1031 * The stable flag requests synchronous writes.
1032 * N.B. After this call fhp needs an fh_put
1033 */
Al Viro6264d692006-10-19 23:28:58 -07001034__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035nfsd_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
David Shaw31dec252009-03-05 20:16:14 -05001036 loff_t offset, struct kvec *vec, int vlen, unsigned long *cnt,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 int *stablep)
1038{
Al Viro6264d692006-10-19 23:28:58 -07001039 __be32 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 if (file) {
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001042 err = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001043 NFSD_MAY_WRITE|NFSD_MAY_OWNER_OVERRIDE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 if (err)
1045 goto out;
1046 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen, cnt,
1047 stablep);
1048 } else {
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001049 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_WRITE, &file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (err)
1051 goto out;
1052
1053 if (cnt)
1054 err = nfsd_vfs_write(rqstp, fhp, file, offset, vec, vlen,
1055 cnt, stablep);
1056 nfsd_close(file);
1057 }
1058out:
1059 return err;
1060}
1061
1062#ifdef CONFIG_NFSD_V3
1063/*
1064 * Commit all pending writes to stable storage.
Trond Myklebustaa696a62010-01-29 16:44:25 -05001065 *
1066 * Note: we only guarantee that data that lies within the range specified
1067 * by the 'offset' and 'count' parameters will be synced.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 *
1069 * Unfortunately we cannot lock the file to make sure we return full WCC
1070 * data to the client, as locking happens lower down in the filesystem.
1071 */
Al Viro6264d692006-10-19 23:28:58 -07001072__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073nfsd_commit(struct svc_rqst *rqstp, struct svc_fh *fhp,
1074 loff_t offset, unsigned long count)
1075{
1076 struct file *file;
Trond Myklebustaa696a62010-01-29 16:44:25 -05001077 loff_t end = LLONG_MAX;
1078 __be32 err = nfserr_inval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
Trond Myklebustaa696a62010-01-29 16:44:25 -05001080 if (offset < 0)
1081 goto out;
1082 if (count != 0) {
1083 end = offset + (loff_t)count - 1;
1084 if (end < offset)
1085 goto out;
1086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
Jeff Layton91885252010-03-19 08:06:28 -04001088 err = nfsd_open(rqstp, fhp, S_IFREG,
1089 NFSD_MAY_WRITE|NFSD_MAY_NOT_BREAK_LEASE, &file);
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001090 if (err)
Trond Myklebustaa696a62010-01-29 16:44:25 -05001091 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 if (EX_ISSYNC(fhp->fh_export)) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +01001093 int err2 = vfs_fsync_range(file, offset, end, 0);
Trond Myklebustaa696a62010-01-29 16:44:25 -05001094
1095 if (err2 != -EINVAL)
1096 err = nfserrno(err2);
1097 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 err = nfserr_notsupp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100
1101 nfsd_close(file);
Trond Myklebustaa696a62010-01-29 16:44:25 -05001102out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 return err;
1104}
1105#endif /* CONFIG_NFSD_V3 */
1106
Adrian Bunkf2b0dee2008-02-13 23:30:26 +02001107static __be32
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001108nfsd_create_setattr(struct svc_rqst *rqstp, struct svc_fh *resfhp,
1109 struct iattr *iap)
1110{
1111 /*
1112 * Mode has already been set earlier in create:
1113 */
1114 iap->ia_valid &= ~ATTR_MODE;
1115 /*
1116 * Setting uid/gid works only for root. Irix appears to
1117 * send along the gid on create when it tries to implement
1118 * setgid directories via NFS:
1119 */
Eric W. Biederman6fab8772013-02-02 06:53:11 -08001120 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001121 iap->ia_valid &= ~(ATTR_UID|ATTR_GID);
1122 if (iap->ia_valid)
1123 return nfsd_setattr(rqstp, resfhp, iap, 0, (time_t)0);
Trond Myklebust0f3a24b2014-07-01 18:27:53 -04001124 /* Callers expect file metadata to be committed here */
Jeff Layton722b6202014-07-03 07:54:19 -04001125 return nfserrno(commit_metadata(resfhp));
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001126}
1127
wengang wang4ac35c22009-02-10 11:27:51 +08001128/* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1129 * setting size to 0 may fail for some specific file systems by the permission
1130 * checking which requires WRITE permission but the mode is 000.
1131 * we ignore the resizing(to 0) on the just new created file, since the size is
1132 * 0 after file created.
1133 *
1134 * call this only after vfs_create() is called.
1135 * */
1136static void
1137nfsd_check_ignore_resizing(struct iattr *iap)
1138{
1139 if ((iap->ia_valid & ATTR_SIZE) && (iap->ia_size == 0))
1140 iap->ia_valid &= ~ATTR_SIZE;
1141}
1142
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143/*
1144 * Create a file (regular, directory, device, fifo); UNIX sockets
1145 * not yet implemented.
1146 * If the response fh has been verified, the parent directory should
1147 * already be locked. Note that the parent directory is left locked.
1148 *
1149 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1150 */
Al Viro6264d692006-10-19 23:28:58 -07001151__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
1153 char *fname, int flen, struct iattr *iap,
1154 int type, dev_t rdev, struct svc_fh *resfhp)
1155{
1156 struct dentry *dentry, *dchild = NULL;
1157 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001158 __be32 err;
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001159 __be32 err2;
Al Viro6264d692006-10-19 23:28:58 -07001160 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
1162 err = nfserr_perm;
1163 if (!flen)
1164 goto out;
1165 err = nfserr_exist;
1166 if (isdotent(fname, flen))
1167 goto out;
1168
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001169 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (err)
1171 goto out;
1172
1173 dentry = fhp->fh_dentry;
1174 dirp = dentry->d_inode;
1175
1176 err = nfserr_notdir;
Al Viroacfa4382008-12-04 10:06:33 -05001177 if (!dirp->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 goto out;
1179 /*
1180 * Check whether the response file handle has been verified yet.
1181 * If it has, the parent directory should already be locked.
1182 */
1183 if (!resfhp->fh_dentry) {
Jan Kara4a55c102012-06-12 16:20:33 +02001184 host_err = fh_want_write(fhp);
1185 if (host_err)
1186 goto out_nfserr;
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001189 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001191 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 if (IS_ERR(dchild))
1193 goto out_nfserr;
1194 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1195 if (err)
1196 goto out;
1197 } else {
1198 /* called from nfsd_proc_create */
1199 dchild = dget(resfhp->fh_dentry);
1200 if (!fhp->fh_locked) {
1201 /* not actually possible */
1202 printk(KERN_ERR
Al Viroa6a9f182013-09-16 10:57:01 -04001203 "nfsd_create: parent %pd2 not locked!\n",
1204 dentry);
Al Virod75f2b92006-01-18 17:43:44 -08001205 err = nfserr_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 goto out;
1207 }
1208 }
1209 /*
1210 * Make sure the child dentry is still negative ...
1211 */
1212 err = nfserr_exist;
1213 if (dchild->d_inode) {
Al Viroa6a9f182013-09-16 10:57:01 -04001214 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1215 dentry, dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 goto out;
1217 }
1218
1219 if (!(iap->ia_valid & ATTR_MODE))
1220 iap->ia_mode = 0;
1221 iap->ia_mode = (iap->ia_mode & S_IALLUGO) | type;
1222
Miklos Szeredi07cad1d2008-07-01 15:38:35 +02001223 err = nfserr_inval;
1224 if (!S_ISREG(type) && !S_ISDIR(type) && !special_file(type)) {
1225 printk(KERN_WARNING "nfsd: bad file type %o in nfsd_create\n",
1226 type);
1227 goto out;
1228 }
1229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 /*
1231 * Get the dir op function pointer.
1232 */
J. Bruce Fields088406b2006-11-08 17:44:59 -08001233 err = 0;
Jan Kara4a55c102012-06-12 16:20:33 +02001234 host_err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 switch (type) {
1236 case S_IFREG:
Al Viro312b63f2012-06-10 18:09:36 -04001237 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
wengang wang4ac35c22009-02-10 11:27:51 +08001238 if (!host_err)
1239 nfsd_check_ignore_resizing(iap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 break;
1241 case S_IFDIR:
Al Viro6264d692006-10-19 23:28:58 -07001242 host_err = vfs_mkdir(dirp, dchild, iap->ia_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 break;
1244 case S_IFCHR:
1245 case S_IFBLK:
1246 case S_IFIFO:
1247 case S_IFSOCK:
Al Viro6264d692006-10-19 23:28:58 -07001248 host_err = vfs_mknod(dirp, dchild, iap->ia_mode, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 break;
Dave Hansen463c3192008-02-15 14:37:57 -08001250 }
Jan Kara4a55c102012-06-12 16:20:33 +02001251 if (host_err < 0)
Dave Hansen463c3192008-02-15 14:37:57 -08001252 goto out_nfserr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Ben Myersf5019122010-02-17 14:05:11 -06001254 err = nfsd_create_setattr(rqstp, resfhp, iap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Ben Myersf5019122010-02-17 14:05:11 -06001256 /*
Trond Myklebust0f3a24b2014-07-01 18:27:53 -04001257 * nfsd_create_setattr already committed the child. Transactional
1258 * filesystems had a chance to commit changes for both parent and
1259 * child * simultaneously making the following commit_metadata a
1260 * noop.
Ben Myersf5019122010-02-17 14:05:11 -06001261 */
1262 err2 = nfserrno(commit_metadata(fhp));
J. Bruce Fields5c002b32007-11-30 16:55:23 -05001263 if (err2)
1264 err = err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 /*
1266 * Update the file handle to get the new inode info.
1267 */
1268 if (!err)
1269 err = fh_update(resfhp);
1270out:
1271 if (dchild && !IS_ERR(dchild))
1272 dput(dchild);
1273 return err;
1274
1275out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001276 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 goto out;
1278}
1279
1280#ifdef CONFIG_NFSD_V3
Mi Jinlongac6721a2011-04-20 17:06:25 +08001281
1282static inline int nfsd_create_is_exclusive(int createmode)
1283{
1284 return createmode == NFS3_CREATE_EXCLUSIVE
1285 || createmode == NFS4_CREATE_EXCLUSIVE4_1;
1286}
1287
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288/*
Mi Jinlongac6721a2011-04-20 17:06:25 +08001289 * NFSv3 and NFSv4 version of nfsd_create
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 */
Al Viro6264d692006-10-19 23:28:58 -07001291__be32
Mi Jinlongac6721a2011-04-20 17:06:25 +08001292do_nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 char *fname, int flen, struct iattr *iap,
1294 struct svc_fh *resfhp, int createmode, u32 *verifier,
J. Bruce Fields856121b2011-10-13 11:37:11 -04001295 bool *truncp, bool *created)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
1297 struct dentry *dentry, *dchild = NULL;
1298 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001299 __be32 err;
1300 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 __u32 v_mtime=0, v_atime=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 err = nfserr_perm;
1304 if (!flen)
1305 goto out;
1306 err = nfserr_exist;
1307 if (isdotent(fname, flen))
1308 goto out;
1309 if (!(iap->ia_valid & ATTR_MODE))
1310 iap->ia_mode = 0;
Sachin Prabhu1574dff2011-04-20 13:09:35 +01001311 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (err)
1313 goto out;
1314
1315 dentry = fhp->fh_dentry;
1316 dirp = dentry->d_inode;
1317
1318 /* Get all the sanity checks out of the way before
1319 * we lock the parent. */
1320 err = nfserr_notdir;
Al Viroacfa4382008-12-04 10:06:33 -05001321 if (!dirp->i_op->lookup)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 goto out;
Jan Kara4a55c102012-06-12 16:20:33 +02001323
1324 host_err = fh_want_write(fhp);
1325 if (host_err)
1326 goto out_nfserr;
1327
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001328 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
1330 /*
1331 * Compose the response file handle.
1332 */
1333 dchild = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001334 host_err = PTR_ERR(dchild);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 if (IS_ERR(dchild))
1336 goto out_nfserr;
1337
Sachin Prabhu1574dff2011-04-20 13:09:35 +01001338 /* If file doesn't exist, check for permissions to create one */
1339 if (!dchild->d_inode) {
1340 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
1341 if (err)
1342 goto out;
1343 }
1344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
1346 if (err)
1347 goto out;
1348
Mi Jinlongac6721a2011-04-20 17:06:25 +08001349 if (nfsd_create_is_exclusive(createmode)) {
Peter Staubachc3978522007-01-26 00:57:00 -08001350 /* solaris7 gets confused (bugid 4218508) if these have
Jeff Layton749997e2007-07-31 00:37:51 -07001351 * the high bit set, so just clear the high bits. If this is
1352 * ever changed to use different attrs for storing the
1353 * verifier, then do_open_lookup() will also need to be fixed
1354 * accordingly.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 */
1356 v_mtime = verifier[0]&0x7fffffff;
1357 v_atime = verifier[1]&0x7fffffff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
1359
1360 if (dchild->d_inode) {
1361 err = 0;
1362
1363 switch (createmode) {
1364 case NFS3_CREATE_UNCHECKED:
1365 if (! S_ISREG(dchild->d_inode->i_mode))
J. Bruce Fields9dc4e6c2012-04-09 18:06:49 -04001366 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 else if (truncp) {
1368 /* in nfsv4, we need to treat this case a little
1369 * differently. we don't want to truncate the
1370 * file now; this would be wrong if the OPEN
1371 * fails for some other reason. furthermore,
1372 * if the size is nonzero, we should ignore it
1373 * according to spec!
1374 */
1375 *truncp = (iap->ia_valid & ATTR_SIZE) && !iap->ia_size;
1376 }
1377 else {
1378 iap->ia_valid &= ATTR_SIZE;
1379 goto set_attr;
1380 }
1381 break;
1382 case NFS3_CREATE_EXCLUSIVE:
1383 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1384 && dchild->d_inode->i_atime.tv_sec == v_atime
Neil Brown7007c902012-12-07 15:40:55 -05001385 && dchild->d_inode->i_size == 0 ) {
1386 if (created)
1387 *created = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 break;
Neil Brown7007c902012-12-07 15:40:55 -05001389 }
Mi Jinlongac6721a2011-04-20 17:06:25 +08001390 case NFS4_CREATE_EXCLUSIVE4_1:
1391 if ( dchild->d_inode->i_mtime.tv_sec == v_mtime
1392 && dchild->d_inode->i_atime.tv_sec == v_atime
Neil Brown7007c902012-12-07 15:40:55 -05001393 && dchild->d_inode->i_size == 0 ) {
1394 if (created)
1395 *created = 1;
Mi Jinlongac6721a2011-04-20 17:06:25 +08001396 goto set_attr;
Neil Brown7007c902012-12-07 15:40:55 -05001397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /* fallthru */
1399 case NFS3_CREATE_GUARDED:
1400 err = nfserr_exist;
1401 }
Al Virobad0dcf2011-11-23 12:03:18 -05001402 fh_drop_write(fhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 goto out;
1404 }
1405
Al Viro312b63f2012-06-10 18:09:36 -04001406 host_err = vfs_create(dirp, dchild, iap->ia_mode, true);
Dave Hansen463c3192008-02-15 14:37:57 -08001407 if (host_err < 0) {
Al Virobad0dcf2011-11-23 12:03:18 -05001408 fh_drop_write(fhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 goto out_nfserr;
Dave Hansen463c3192008-02-15 14:37:57 -08001410 }
J. Bruce Fields81ac95c2006-11-08 17:44:40 -08001411 if (created)
1412 *created = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
wengang wang4ac35c22009-02-10 11:27:51 +08001414 nfsd_check_ignore_resizing(iap);
1415
Mi Jinlongac6721a2011-04-20 17:06:25 +08001416 if (nfsd_create_is_exclusive(createmode)) {
Peter Staubachc3978522007-01-26 00:57:00 -08001417 /* Cram the verifier into atime/mtime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 iap->ia_valid = ATTR_MTIME|ATTR_ATIME
Peter Staubachc3978522007-01-26 00:57:00 -08001419 | ATTR_MTIME_SET|ATTR_ATIME_SET;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 /* XXX someone who knows this better please fix it for nsec */
1421 iap->ia_mtime.tv_sec = v_mtime;
1422 iap->ia_atime.tv_sec = v_atime;
1423 iap->ia_mtime.tv_nsec = 0;
1424 iap->ia_atime.tv_nsec = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 }
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 set_attr:
Ben Myersf5019122010-02-17 14:05:11 -06001428 err = nfsd_create_setattr(rqstp, resfhp, iap);
1429
1430 /*
Trond Myklebust0f3a24b2014-07-01 18:27:53 -04001431 * nfsd_create_setattr already committed the child
1432 * (and possibly also the parent).
Ben Myersf5019122010-02-17 14:05:11 -06001433 */
1434 if (!err)
1435 err = nfserrno(commit_metadata(fhp));
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001436
1437 /*
1438 * Update the filehandle to get the new inode info.
1439 */
1440 if (!err)
1441 err = fh_update(resfhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 out:
1444 fh_unlock(fhp);
1445 if (dchild && !IS_ERR(dchild))
1446 dput(dchild);
Jan Kara4a55c102012-06-12 16:20:33 +02001447 fh_drop_write(fhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 return err;
1449
1450 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001451 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 goto out;
1453}
1454#endif /* CONFIG_NFSD_V3 */
1455
1456/*
1457 * Read a symlink. On entry, *lenp must contain the maximum path length that
1458 * fits into the buffer. On return, it contains the true length.
1459 * N.B. After this call fhp needs an fh_put
1460 */
Al Viro6264d692006-10-19 23:28:58 -07001461__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462nfsd_readlink(struct svc_rqst *rqstp, struct svc_fh *fhp, char *buf, int *lenp)
1463{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 struct inode *inode;
1465 mm_segment_t oldfs;
Al Viro6264d692006-10-19 23:28:58 -07001466 __be32 err;
1467 int host_err;
Al Viro68ac1232012-03-15 08:21:57 -04001468 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001470 err = fh_verify(rqstp, fhp, S_IFLNK, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 if (err)
1472 goto out;
1473
Al Viro68ac1232012-03-15 08:21:57 -04001474 path.mnt = fhp->fh_export->ex_path.mnt;
1475 path.dentry = fhp->fh_dentry;
1476 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
1478 err = nfserr_inval;
Al Viroacfa4382008-12-04 10:06:33 -05001479 if (!inode->i_op->readlink)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 goto out;
1481
Al Viro68ac1232012-03-15 08:21:57 -04001482 touch_atime(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 /* N.B. Why does this call need a get_fs()??
1484 * Remove the set_fs and watch the fireworks:-) --okir
1485 */
1486
1487 oldfs = get_fs(); set_fs(KERNEL_DS);
J. Bruce Fieldsfac7a172012-07-27 14:16:55 -04001488 host_err = inode->i_op->readlink(path.dentry, (char __user *)buf, *lenp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 set_fs(oldfs);
1490
Al Viro6264d692006-10-19 23:28:58 -07001491 if (host_err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 goto out_nfserr;
Al Viro6264d692006-10-19 23:28:58 -07001493 *lenp = host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 err = 0;
1495out:
1496 return err;
1497
1498out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001499 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 goto out;
1501}
1502
1503/*
1504 * Create a symlink and look up its inode
1505 * N.B. After this call _both_ fhp and resfhp need an fh_put
1506 */
Al Viro6264d692006-10-19 23:28:58 -07001507__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508nfsd_symlink(struct svc_rqst *rqstp, struct svc_fh *fhp,
1509 char *fname, int flen,
J. Bruce Fields52ee0432014-06-20 11:52:21 -04001510 char *path,
Kinglong Mee1e444f52014-07-01 17:48:02 +08001511 struct svc_fh *resfhp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 struct dentry *dentry, *dnew;
Al Viro6264d692006-10-19 23:28:58 -07001514 __be32 err, cerr;
1515 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
1517 err = nfserr_noent;
J. Bruce Fields52ee0432014-06-20 11:52:21 -04001518 if (!flen || path[0] == '\0')
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 goto out;
1520 err = nfserr_exist;
1521 if (isdotent(fname, flen))
1522 goto out;
1523
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001524 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 if (err)
1526 goto out;
Jan Kara4a55c102012-06-12 16:20:33 +02001527
1528 host_err = fh_want_write(fhp);
1529 if (host_err)
1530 goto out_nfserr;
1531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 fh_lock(fhp);
1533 dentry = fhp->fh_dentry;
1534 dnew = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001535 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 if (IS_ERR(dnew))
1537 goto out_nfserr;
1538
J. Bruce Fields52ee0432014-06-20 11:52:21 -04001539 host_err = vfs_symlink(dentry->d_inode, dnew, path);
Al Viro6264d692006-10-19 23:28:58 -07001540 err = nfserrno(host_err);
Ben Myersf5019122010-02-17 14:05:11 -06001541 if (!err)
1542 err = nfserrno(commit_metadata(fhp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 fh_unlock(fhp);
1544
Al Virobad0dcf2011-11-23 12:03:18 -05001545 fh_drop_write(fhp);
Dave Hansen75c3f292008-02-15 14:37:45 -08001546
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 cerr = fh_compose(resfhp, fhp->fh_export, dnew, fhp);
1548 dput(dnew);
1549 if (err==0) err = cerr;
1550out:
1551 return err;
1552
1553out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001554 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 goto out;
1556}
1557
1558/*
1559 * Create a hardlink
1560 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1561 */
Al Viro6264d692006-10-19 23:28:58 -07001562__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563nfsd_link(struct svc_rqst *rqstp, struct svc_fh *ffhp,
1564 char *name, int len, struct svc_fh *tfhp)
1565{
1566 struct dentry *ddir, *dnew, *dold;
J. Bruce Fields55b13352010-07-19 16:38:24 -04001567 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001568 __be32 err;
1569 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001571 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572 if (err)
1573 goto out;
J. Bruce Fields7d818a72011-08-15 16:59:55 -04001574 err = fh_verify(rqstp, tfhp, 0, NFSD_MAY_NOP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 if (err)
1576 goto out;
J. Bruce Fields7d818a72011-08-15 16:59:55 -04001577 err = nfserr_isdir;
1578 if (S_ISDIR(tfhp->fh_dentry->d_inode->i_mode))
1579 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 err = nfserr_perm;
1581 if (!len)
1582 goto out;
1583 err = nfserr_exist;
1584 if (isdotent(name, len))
1585 goto out;
1586
Jan Kara4a55c102012-06-12 16:20:33 +02001587 host_err = fh_want_write(tfhp);
1588 if (host_err) {
1589 err = nfserrno(host_err);
1590 goto out;
1591 }
1592
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001593 fh_lock_nested(ffhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 ddir = ffhp->fh_dentry;
1595 dirp = ddir->d_inode;
1596
1597 dnew = lookup_one_len(name, ddir, len);
Al Viro6264d692006-10-19 23:28:58 -07001598 host_err = PTR_ERR(dnew);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 if (IS_ERR(dnew))
1600 goto out_nfserr;
1601
1602 dold = tfhp->fh_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603
J. Bruce Fields4795bb32011-01-11 13:55:46 -05001604 err = nfserr_noent;
1605 if (!dold->d_inode)
Jan Kara4a55c102012-06-12 16:20:33 +02001606 goto out_dput;
J. Bruce Fields146a8592011-09-20 17:14:31 -04001607 host_err = vfs_link(dold, dirp, dnew, NULL);
Al Viro6264d692006-10-19 23:28:58 -07001608 if (!host_err) {
Ben Myersf5019122010-02-17 14:05:11 -06001609 err = nfserrno(commit_metadata(ffhp));
1610 if (!err)
1611 err = nfserrno(commit_metadata(tfhp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 } else {
Al Viro6264d692006-10-19 23:28:58 -07001613 if (host_err == -EXDEV && rqstp->rq_vers == 2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 err = nfserr_acces;
1615 else
Al Viro6264d692006-10-19 23:28:58 -07001616 err = nfserrno(host_err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 }
Dave Hansen75c3f292008-02-15 14:37:45 -08001618out_dput:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 dput(dnew);
David M. Richter270d56e2006-06-30 01:56:15 -07001620out_unlock:
1621 fh_unlock(ffhp);
Jan Kara4a55c102012-06-12 16:20:33 +02001622 fh_drop_write(tfhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623out:
1624 return err;
1625
1626out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001627 err = nfserrno(host_err);
David M. Richter270d56e2006-06-30 01:56:15 -07001628 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629}
1630
1631/*
1632 * Rename a file
1633 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1634 */
Al Viro6264d692006-10-19 23:28:58 -07001635__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636nfsd_rename(struct svc_rqst *rqstp, struct svc_fh *ffhp, char *fname, int flen,
1637 struct svc_fh *tfhp, char *tname, int tlen)
1638{
1639 struct dentry *fdentry, *tdentry, *odentry, *ndentry, *trap;
1640 struct inode *fdir, *tdir;
Al Viro6264d692006-10-19 23:28:58 -07001641 __be32 err;
1642 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001644 err = fh_verify(rqstp, ffhp, S_IFDIR, NFSD_MAY_REMOVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 if (err)
1646 goto out;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001647 err = fh_verify(rqstp, tfhp, S_IFDIR, NFSD_MAY_CREATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 if (err)
1649 goto out;
1650
1651 fdentry = ffhp->fh_dentry;
1652 fdir = fdentry->d_inode;
1653
1654 tdentry = tfhp->fh_dentry;
1655 tdir = tdentry->d_inode;
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 err = nfserr_perm;
1658 if (!flen || isdotent(fname, flen) || !tlen || isdotent(tname, tlen))
1659 goto out;
1660
Jan Kara4a55c102012-06-12 16:20:33 +02001661 host_err = fh_want_write(ffhp);
1662 if (host_err) {
1663 err = nfserrno(host_err);
1664 goto out;
1665 }
1666
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 /* cannot use fh_lock as we need deadlock protective ordering
1668 * so do it by hand */
1669 trap = lock_rename(tdentry, fdentry);
1670 ffhp->fh_locked = tfhp->fh_locked = 1;
1671 fill_pre_wcc(ffhp);
1672 fill_pre_wcc(tfhp);
1673
1674 odentry = lookup_one_len(fname, fdentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001675 host_err = PTR_ERR(odentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (IS_ERR(odentry))
1677 goto out_nfserr;
1678
Al Viro6264d692006-10-19 23:28:58 -07001679 host_err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 if (!odentry->d_inode)
1681 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001682 host_err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 if (odentry == trap)
1684 goto out_dput_old;
1685
1686 ndentry = lookup_one_len(tname, tdentry, tlen);
Al Viro6264d692006-10-19 23:28:58 -07001687 host_err = PTR_ERR(ndentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 if (IS_ERR(ndentry))
1689 goto out_dput_old;
Al Viro6264d692006-10-19 23:28:58 -07001690 host_err = -ENOTEMPTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 if (ndentry == trap)
1692 goto out_dput_new;
1693
Dave Hansen9079b1e2008-02-15 14:37:49 -08001694 host_err = -EXDEV;
1695 if (ffhp->fh_export->ex_path.mnt != tfhp->fh_export->ex_path.mnt)
1696 goto out_dput_new;
J. Bruce Fieldsaa387d6c2013-04-15 16:03:46 -04001697 if (ffhp->fh_export->ex_path.dentry != tfhp->fh_export->ex_path.dentry)
1698 goto out_dput_new;
Dave Hansen9079b1e2008-02-15 14:37:49 -08001699
Miklos Szeredi520c8b12014-04-01 17:08:42 +02001700 host_err = vfs_rename(fdir, odentry, tdir, ndentry, NULL, 0);
Ben Myersf5019122010-02-17 14:05:11 -06001701 if (!host_err) {
1702 host_err = commit_metadata(tfhp);
Al Viro6264d692006-10-19 23:28:58 -07001703 if (!host_err)
Ben Myersf5019122010-02-17 14:05:11 -06001704 host_err = commit_metadata(ffhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 out_dput_new:
1707 dput(ndentry);
1708 out_dput_old:
1709 dput(odentry);
1710 out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001711 err = nfserrno(host_err);
J. Bruce Fieldsfbb74a32014-03-28 16:43:17 -04001712 /*
1713 * We cannot rely on fh_unlock on the two filehandles,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 * as that would do the wrong thing if the two directories
J. Bruce Fieldsfbb74a32014-03-28 16:43:17 -04001715 * were the same, so again we do it by hand.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 */
1717 fill_post_wcc(ffhp);
1718 fill_post_wcc(tfhp);
1719 unlock_rename(tdentry, fdentry);
1720 ffhp->fh_locked = tfhp->fh_locked = 0;
Jan Kara4a55c102012-06-12 16:20:33 +02001721 fh_drop_write(ffhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723out:
1724 return err;
1725}
1726
1727/*
1728 * Unlink a file or directory
1729 * N.B. After this call fhp needs an fh_put
1730 */
Al Viro6264d692006-10-19 23:28:58 -07001731__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
1733 char *fname, int flen)
1734{
1735 struct dentry *dentry, *rdentry;
1736 struct inode *dirp;
Al Viro6264d692006-10-19 23:28:58 -07001737 __be32 err;
1738 int host_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740 err = nfserr_acces;
1741 if (!flen || isdotent(fname, flen))
1742 goto out;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001743 err = fh_verify(rqstp, fhp, S_IFDIR, NFSD_MAY_REMOVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if (err)
1745 goto out;
1746
Jan Kara4a55c102012-06-12 16:20:33 +02001747 host_err = fh_want_write(fhp);
1748 if (host_err)
1749 goto out_nfserr;
1750
Peter Zijlstra12fd3522006-10-02 02:18:03 -07001751 fh_lock_nested(fhp, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 dentry = fhp->fh_dentry;
1753 dirp = dentry->d_inode;
1754
1755 rdentry = lookup_one_len(fname, dentry, flen);
Al Viro6264d692006-10-19 23:28:58 -07001756 host_err = PTR_ERR(rdentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (IS_ERR(rdentry))
1758 goto out_nfserr;
1759
1760 if (!rdentry->d_inode) {
1761 dput(rdentry);
1762 err = nfserr_noent;
1763 goto out;
1764 }
1765
1766 if (!type)
1767 type = rdentry->d_inode->i_mode & S_IFMT;
1768
J. Bruce Fields9ce137e2011-01-11 14:07:12 -05001769 if (type != S_IFDIR)
J. Bruce Fieldsb21996e2011-09-20 09:14:34 -04001770 host_err = vfs_unlink(dirp, rdentry, NULL);
J. Bruce Fields9ce137e2011-01-11 14:07:12 -05001771 else
Al Viro6264d692006-10-19 23:28:58 -07001772 host_err = vfs_rmdir(dirp, rdentry);
J. Bruce Fields541ce982011-01-14 20:00:02 -05001773 if (!host_err)
1774 host_err = commit_metadata(fhp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 dput(rdentry);
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777out_nfserr:
Al Viro6264d692006-10-19 23:28:58 -07001778 err = nfserrno(host_err);
YAMAMOTO Takashif193fba2006-01-18 17:43:13 -08001779out:
1780 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781}
1782
1783/*
David Woodhouse14f7dd62008-07-31 20:29:12 +01001784 * We do this buffering because we must not call back into the file
1785 * system's ->lookup() method from the filldir callback. That may well
1786 * deadlock a number of file systems.
1787 *
1788 * This is based heavily on the implementation of same in XFS.
1789 */
1790struct buffered_dirent {
1791 u64 ino;
1792 loff_t offset;
1793 int namlen;
1794 unsigned int d_type;
1795 char name[];
1796};
1797
1798struct readdir_data {
Al Viro5c0ba4e2013-05-15 13:52:59 -04001799 struct dir_context ctx;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001800 char *dirent;
1801 size_t used;
Al Viro53c9c5c2008-08-24 07:29:52 -04001802 int full;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001803};
1804
1805static int nfsd_buffered_filldir(void *__buf, const char *name, int namlen,
1806 loff_t offset, u64 ino, unsigned int d_type)
David Woodhouse2628b762008-07-31 17:16:51 +01001807{
David Woodhouse14f7dd62008-07-31 20:29:12 +01001808 struct readdir_data *buf = __buf;
1809 struct buffered_dirent *de = (void *)(buf->dirent + buf->used);
1810 unsigned int reclen;
1811
1812 reclen = ALIGN(sizeof(struct buffered_dirent) + namlen, sizeof(u64));
Al Viro53c9c5c2008-08-24 07:29:52 -04001813 if (buf->used + reclen > PAGE_SIZE) {
1814 buf->full = 1;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001815 return -EINVAL;
Al Viro53c9c5c2008-08-24 07:29:52 -04001816 }
David Woodhouse14f7dd62008-07-31 20:29:12 +01001817
1818 de->namlen = namlen;
1819 de->offset = offset;
1820 de->ino = ino;
1821 de->d_type = d_type;
1822 memcpy(de->name, name, namlen);
1823 buf->used += reclen;
1824
1825 return 0;
1826}
1827
David Woodhouse2f9092e2009-04-20 23:18:37 +01001828static __be32 nfsd_buffered_readdir(struct file *file, filldir_t func,
1829 struct readdir_cd *cdp, loff_t *offsetp)
David Woodhouse14f7dd62008-07-31 20:29:12 +01001830{
David Woodhouse14f7dd62008-07-31 20:29:12 +01001831 struct buffered_dirent *de;
David Woodhouse2628b762008-07-31 17:16:51 +01001832 int host_err;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001833 int size;
1834 loff_t offset;
Al Viroac6614b2013-05-22 22:22:04 -04001835 struct readdir_data buf = {
1836 .ctx.actor = nfsd_buffered_filldir,
1837 .dirent = (void *)__get_free_page(GFP_KERNEL)
1838 };
David Woodhouse2628b762008-07-31 17:16:51 +01001839
David Woodhouse14f7dd62008-07-31 20:29:12 +01001840 if (!buf.dirent)
David Woodhouse2f9092e2009-04-20 23:18:37 +01001841 return nfserrno(-ENOMEM);
David Woodhouse2628b762008-07-31 17:16:51 +01001842
David Woodhouse14f7dd62008-07-31 20:29:12 +01001843 offset = *offsetp;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001844
1845 while (1) {
Al Viro496ad9a2013-01-23 17:07:38 -05001846 struct inode *dir_inode = file_inode(file);
David Woodhouse14f7dd62008-07-31 20:29:12 +01001847 unsigned int reclen;
1848
Doug Nazarb726e922008-11-05 06:16:28 -05001849 cdp->err = nfserr_eof; /* will be cleared on successful read */
David Woodhouse14f7dd62008-07-31 20:29:12 +01001850 buf.used = 0;
Al Viro53c9c5c2008-08-24 07:29:52 -04001851 buf.full = 0;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001852
Al Viro5c0ba4e2013-05-15 13:52:59 -04001853 host_err = iterate_dir(file, &buf.ctx);
Al Viro53c9c5c2008-08-24 07:29:52 -04001854 if (buf.full)
1855 host_err = 0;
1856
1857 if (host_err < 0)
David Woodhouse14f7dd62008-07-31 20:29:12 +01001858 break;
1859
1860 size = buf.used;
1861
1862 if (!size)
1863 break;
1864
David Woodhouse2f9092e2009-04-20 23:18:37 +01001865 /*
1866 * Various filldir functions may end up calling back into
1867 * lookup_one_len() and the file system's ->lookup() method.
1868 * These expect i_mutex to be held, as it would within readdir.
1869 */
1870 host_err = mutex_lock_killable(&dir_inode->i_mutex);
1871 if (host_err)
1872 break;
1873
David Woodhouse14f7dd62008-07-31 20:29:12 +01001874 de = (struct buffered_dirent *)buf.dirent;
1875 while (size > 0) {
1876 offset = de->offset;
1877
1878 if (func(cdp, de->name, de->namlen, de->offset,
1879 de->ino, de->d_type))
David Woodhouse2f9092e2009-04-20 23:18:37 +01001880 break;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001881
1882 if (cdp->err != nfs_ok)
David Woodhouse2f9092e2009-04-20 23:18:37 +01001883 break;
David Woodhouse14f7dd62008-07-31 20:29:12 +01001884
1885 reclen = ALIGN(sizeof(*de) + de->namlen,
1886 sizeof(u64));
1887 size -= reclen;
1888 de = (struct buffered_dirent *)((char *)de + reclen);
1889 }
David Woodhouse2f9092e2009-04-20 23:18:37 +01001890 mutex_unlock(&dir_inode->i_mutex);
1891 if (size > 0) /* We bailed out early */
1892 break;
1893
David Woodhousec002a6c2008-08-17 17:21:18 +01001894 offset = vfs_llseek(file, 0, SEEK_CUR);
David Woodhouse14f7dd62008-07-31 20:29:12 +01001895 }
1896
David Woodhouse14f7dd62008-07-31 20:29:12 +01001897 free_page((unsigned long)(buf.dirent));
David Woodhouse2628b762008-07-31 17:16:51 +01001898
1899 if (host_err)
1900 return nfserrno(host_err);
David Woodhouse14f7dd62008-07-31 20:29:12 +01001901
1902 *offsetp = offset;
1903 return cdp->err;
David Woodhouse2628b762008-07-31 17:16:51 +01001904}
1905
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906/*
1907 * Read entries from a directory.
1908 * The NFSv3/4 verifier we ignore for now.
1909 */
Al Viro6264d692006-10-19 23:28:58 -07001910__be32
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911nfsd_readdir(struct svc_rqst *rqstp, struct svc_fh *fhp, loff_t *offsetp,
NeilBrowna0ad13e2007-01-26 00:57:10 -08001912 struct readdir_cd *cdp, filldir_t func)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913{
Al Viro6264d692006-10-19 23:28:58 -07001914 __be32 err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 struct file *file;
1916 loff_t offset = *offsetp;
Bernd Schubert06effdb2012-03-18 22:44:50 -04001917 int may_flags = NFSD_MAY_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Bernd Schubert06effdb2012-03-18 22:44:50 -04001919 /* NFSv2 only supports 32 bit cookies */
1920 if (rqstp->rq_vers > 2)
1921 may_flags |= NFSD_MAY_64BIT_COOKIE;
1922
1923 err = nfsd_open(rqstp, fhp, S_IFDIR, may_flags, &file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 if (err)
1925 goto out;
1926
Jeff Laytonb108fe62012-04-25 15:30:00 -04001927 offset = vfs_llseek(file, offset, SEEK_SET);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 if (offset < 0) {
1929 err = nfserrno((int)offset);
1930 goto out_close;
1931 }
1932
David Woodhouse14f7dd62008-07-31 20:29:12 +01001933 err = nfsd_buffered_readdir(file, func, cdp, offsetp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935 if (err == nfserr_eof || err == nfserr_toosmall)
1936 err = nfs_ok; /* can still be found in ->err */
1937out_close:
1938 nfsd_close(file);
1939out:
1940 return err;
1941}
1942
1943/*
1944 * Get file system stats
1945 * N.B. After this call fhp needs an fh_put
1946 */
Al Viro6264d692006-10-19 23:28:58 -07001947__be32
J. Bruce Fields04716e62008-08-07 13:00:20 -04001948nfsd_statfs(struct svc_rqst *rqstp, struct svc_fh *fhp, struct kstatfs *stat, int access)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
Christoph Hellwigebabe9a2010-07-07 18:53:11 +02001950 __be32 err;
1951
1952 err = fh_verify(rqstp, fhp, 0, NFSD_MAY_NOP | access);
Takashi Iwaif6360ef2010-08-13 15:53:49 +02001953 if (!err) {
1954 struct path path = {
1955 .mnt = fhp->fh_export->ex_path.mnt,
1956 .dentry = fhp->fh_dentry,
1957 };
1958 if (vfs_statfs(&path, stat))
1959 err = nfserr_io;
1960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 return err;
1962}
1963
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001964static int exp_rdonly(struct svc_rqst *rqstp, struct svc_export *exp)
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001965{
J. Bruce Fieldsc7d51402007-07-19 01:49:20 -07001966 return nfsexp_flags(rqstp, exp) & NFSEXP_READONLY;
J. Bruce Fieldse22841c2007-07-19 01:49:20 -07001967}
1968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969/*
1970 * Check for a user's access permissions to this inode.
1971 */
Al Viro6264d692006-10-19 23:28:58 -07001972__be32
J. Bruce Fields0ec757d2007-07-17 04:04:48 -07001973nfsd_permission(struct svc_rqst *rqstp, struct svc_export *exp,
1974 struct dentry *dentry, int acc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975{
1976 struct inode *inode = dentry->d_inode;
1977 int err;
1978
J. Bruce Fieldsaea93392011-04-10 10:35:12 -04001979 if ((acc & NFSD_MAY_MASK) == NFSD_MAY_NOP)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 return 0;
1981#if 0
1982 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1983 acc,
Miklos Szeredi8837abc2008-06-16 13:20:29 +02001984 (acc & NFSD_MAY_READ)? " read" : "",
1985 (acc & NFSD_MAY_WRITE)? " write" : "",
1986 (acc & NFSD_MAY_EXEC)? " exec" : "",
1987 (acc & NFSD_MAY_SATTR)? " sattr" : "",
1988 (acc & NFSD_MAY_TRUNC)? " trunc" : "",
1989 (acc & NFSD_MAY_LOCK)? " lock" : "",
1990 (acc & NFSD_MAY_OWNER_OVERRIDE)? " owneroverride" : "",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 inode->i_mode,
1992 IS_IMMUTABLE(inode)? " immut" : "",
1993 IS_APPEND(inode)? " append" : "",
Dave Hansen2c463e92008-02-15 14:37:56 -08001994 __mnt_is_readonly(exp->ex_path.mnt)? " ro" : "");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 dprintk(" owner %d/%d user %d/%d\n",
David Howells5cc0a842008-11-14 10:38:58 +11001996 inode->i_uid, inode->i_gid, current_fsuid(), current_fsgid());
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997#endif
1998
1999 /* Normally we reject any write/sattr etc access on a read-only file
2000 * system. But if it is IRIX doing check on write-access for a
2001 * device special file, we ignore rofs.
2002 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002003 if (!(acc & NFSD_MAY_LOCAL_ACCESS))
2004 if (acc & (NFSD_MAY_WRITE | NFSD_MAY_SATTR | NFSD_MAY_TRUNC)) {
Dave Hansen2c463e92008-02-15 14:37:56 -08002005 if (exp_rdonly(rqstp, exp) ||
2006 __mnt_is_readonly(exp->ex_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 return nfserr_rofs;
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002008 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return nfserr_perm;
2010 }
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002011 if ((acc & NFSD_MAY_TRUNC) && IS_APPEND(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 return nfserr_perm;
2013
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002014 if (acc & NFSD_MAY_LOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 /* If we cannot rely on authentication in NLM requests,
2016 * just allow locks, otherwise require read permission, or
2017 * ownership
2018 */
2019 if (exp->ex_flags & NFSEXP_NOAUTHNLM)
2020 return 0;
2021 else
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002022 acc = NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023 }
2024 /*
2025 * The file owner always gets access permission for accesses that
2026 * would normally be checked at open time. This is to make
2027 * file access work even when the client has done a fchmod(fd, 0).
2028 *
2029 * However, `cp foo bar' should fail nevertheless when bar is
2030 * readonly. A sensible way to do this might be to reject all
2031 * attempts to truncate a read-only file, because a creat() call
2032 * always implies file truncation.
2033 * ... but this isn't really fair. A process may reasonably call
2034 * ftruncate on an open file descriptor on a file with perm 000.
2035 * We must trust the client to do permission checking - using "ACCESS"
2036 * with NFSv3.
2037 */
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002038 if ((acc & NFSD_MAY_OWNER_OVERRIDE) &&
Eric W. Biederman6fab8772013-02-02 06:53:11 -08002039 uid_eq(inode->i_uid, current_fsuid()))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 return 0;
2041
Miklos Szeredi8837abc2008-06-16 13:20:29 +02002042 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
Al Virof419a2e2008-07-22 00:07:17 -04002043 err = inode_permission(inode, acc & (MAY_READ|MAY_WRITE|MAY_EXEC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 /* Allow read access to binaries even when mode 111 */
2046 if (err == -EACCES && S_ISREG(inode->i_mode) &&
J. Bruce Fieldsa0432262011-08-25 10:48:39 -04002047 (acc == (NFSD_MAY_READ | NFSD_MAY_OWNER_OVERRIDE) ||
2048 acc == (NFSD_MAY_READ | NFSD_MAY_READ_IF_EXEC)))
Al Virof419a2e2008-07-22 00:07:17 -04002049 err = inode_permission(inode, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050
2051 return err? nfserrno(err) : 0;
2052}
2053
2054void
2055nfsd_racache_shutdown(void)
2056{
Jeff Layton54a66e52008-08-13 22:03:27 -04002057 struct raparms *raparm, *last_raparm;
2058 unsigned int i;
2059
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 dprintk("nfsd: freeing readahead buffers.\n");
Jeff Layton54a66e52008-08-13 22:03:27 -04002061
2062 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
2063 raparm = raparm_hash[i].pb_head;
2064 while(raparm) {
2065 last_raparm = raparm;
2066 raparm = raparm->p_next;
2067 kfree(last_raparm);
2068 }
2069 raparm_hash[i].pb_head = NULL;
2070 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071}
2072/*
2073 * Initialize readahead param cache
2074 */
2075int
2076nfsd_racache_init(int cache_size)
2077{
2078 int i;
Greg Banksfce14562006-10-04 02:15:49 -07002079 int j = 0;
2080 int nperbucket;
Jeff Layton54a66e52008-08-13 22:03:27 -04002081 struct raparms **raparm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082
Greg Banksfce14562006-10-04 02:15:49 -07002083
Jeff Layton54a66e52008-08-13 22:03:27 -04002084 if (raparm_hash[0].pb_head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 return 0;
Jeff Layton54a66e52008-08-13 22:03:27 -04002086 nperbucket = DIV_ROUND_UP(cache_size, RAPARM_HASH_SIZE);
Kinglong Mee3c7aa152014-06-10 18:08:19 +08002087 nperbucket = max(2, nperbucket);
Jeff Layton54a66e52008-08-13 22:03:27 -04002088 cache_size = nperbucket * RAPARM_HASH_SIZE;
Yan Burman4b3bb062006-12-08 02:39:41 -08002089
2090 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size);
Jeff Layton54a66e52008-08-13 22:03:27 -04002091
2092 for (i = 0; i < RAPARM_HASH_SIZE; i++) {
Yan Burman4b3bb062006-12-08 02:39:41 -08002093 spin_lock_init(&raparm_hash[i].pb_lock);
Jeff Layton54a66e52008-08-13 22:03:27 -04002094
2095 raparm = &raparm_hash[i].pb_head;
2096 for (j = 0; j < nperbucket; j++) {
2097 *raparm = kzalloc(sizeof(struct raparms), GFP_KERNEL);
2098 if (!*raparm)
2099 goto out_nomem;
2100 raparm = &(*raparm)->p_next;
2101 }
2102 *raparm = NULL;
Yan Burman4b3bb062006-12-08 02:39:41 -08002103 }
2104
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 nfsdstats.ra_size = cache_size;
2106 return 0;
Jeff Layton54a66e52008-08-13 22:03:27 -04002107
2108out_nomem:
2109 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2110 nfsd_racache_shutdown();
2111 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112}