~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/nfs/dir.c

Version: ~ [ linux-5.2-rc1 ] ~ [ linux-5.1.2 ] ~ [ linux-5.0.16 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.43 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.119 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.176 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.179 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.139 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.67 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/fs/nfs/dir.c
  3  *
  4  *  Copyright (C) 1992  Rick Sladkey
  5  *
  6  *  nfs directory handling functions
  7  *
  8  * 10 Apr 1996  Added silly rename for unlink   --okir
  9  * 28 Sep 1996  Improved directory cache --okir
 10  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de 
 11  *              Re-implemented silly rename for unlink, newly implemented
 12  *              silly rename for nfs_rename() following the suggestions
 13  *              of Olaf Kirch (okir) found in this file.
 14  *              Following Linus comments on my original hack, this version
 15  *              depends only on the dcache stuff and doesn't touch the inode
 16  *              layer (iput() and friends).
 17  *  6 Jun 1999  Cache readdir lookups in the page cache. -DaveM
 18  */
 19 
 20 #include <linux/module.h>
 21 #include <linux/time.h>
 22 #include <linux/errno.h>
 23 #include <linux/stat.h>
 24 #include <linux/fcntl.h>
 25 #include <linux/string.h>
 26 #include <linux/kernel.h>
 27 #include <linux/slab.h>
 28 #include <linux/mm.h>
 29 #include <linux/sunrpc/clnt.h>
 30 #include <linux/nfs_fs.h>
 31 #include <linux/nfs_mount.h>
 32 #include <linux/pagemap.h>
 33 #include <linux/pagevec.h>
 34 #include <linux/namei.h>
 35 #include <linux/mount.h>
 36 #include <linux/swap.h>
 37 #include <linux/sched.h>
 38 #include <linux/kmemleak.h>
 39 #include <linux/xattr.h>
 40 
 41 #include "delegation.h"
 42 #include "iostat.h"
 43 #include "internal.h"
 44 #include "fscache.h"
 45 
 46 #include "nfstrace.h"
 47 
 48 /* #define NFS_DEBUG_VERBOSE 1 */
 49 
 50 static int nfs_opendir(struct inode *, struct file *);
 51 static int nfs_closedir(struct inode *, struct file *);
 52 static int nfs_readdir(struct file *, struct dir_context *);
 53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
 54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
 55 static void nfs_readdir_clear_array(struct page*);
 56 
 57 const struct file_operations nfs_dir_operations = {
 58         .llseek         = nfs_llseek_dir,
 59         .read           = generic_read_dir,
 60         .iterate        = nfs_readdir,
 61         .open           = nfs_opendir,
 62         .release        = nfs_closedir,
 63         .fsync          = nfs_fsync_dir,
 64 };
 65 
 66 const struct address_space_operations nfs_dir_aops = {
 67         .freepage = nfs_readdir_clear_array,
 68 };
 69 
 70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
 71 {
 72         struct nfs_inode *nfsi = NFS_I(dir);
 73         struct nfs_open_dir_context *ctx;
 74         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 75         if (ctx != NULL) {
 76                 ctx->duped = 0;
 77                 ctx->attr_gencount = nfsi->attr_gencount;
 78                 ctx->dir_cookie = 0;
 79                 ctx->dup_cookie = 0;
 80                 ctx->cred = get_rpccred(cred);
 81                 spin_lock(&dir->i_lock);
 82                 list_add(&ctx->list, &nfsi->open_files);
 83                 spin_unlock(&dir->i_lock);
 84                 return ctx;
 85         }
 86         return  ERR_PTR(-ENOMEM);
 87 }
 88 
 89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
 90 {
 91         spin_lock(&dir->i_lock);
 92         list_del(&ctx->list);
 93         spin_unlock(&dir->i_lock);
 94         put_rpccred(ctx->cred);
 95         kfree(ctx);
 96 }
 97 
 98 /*
 99  * Open file
100  */
101 static int
102 nfs_opendir(struct inode *inode, struct file *filp)
103 {
104         int res = 0;
105         struct nfs_open_dir_context *ctx;
106         struct rpc_cred *cred;
107 
108         dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
109 
110         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
111 
112         cred = rpc_lookup_cred();
113         if (IS_ERR(cred))
114                 return PTR_ERR(cred);
115         ctx = alloc_nfs_open_dir_context(inode, cred);
116         if (IS_ERR(ctx)) {
117                 res = PTR_ERR(ctx);
118                 goto out;
119         }
120         filp->private_data = ctx;
121         if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122                 /* This is a mountpoint, so d_revalidate will never
123                  * have been called, so we need to refresh the
124                  * inode (for close-open consistency) ourselves.
125                  */
126                 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
127         }
128 out:
129         put_rpccred(cred);
130         return res;
131 }
132 
133 static int
134 nfs_closedir(struct inode *inode, struct file *filp)
135 {
136         put_nfs_open_dir_context(file_inode(filp), filp->private_data);
137         return 0;
138 }
139 
140 struct nfs_cache_array_entry {
141         u64 cookie;
142         u64 ino;
143         struct qstr string;
144         unsigned char d_type;
145 };
146 
147 struct nfs_cache_array {
148         int size;
149         int eof_index;
150         u64 last_cookie;
151         struct nfs_cache_array_entry array[0];
152 };
153 
154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
155 typedef struct {
156         struct file     *file;
157         struct page     *page;
158         struct dir_context *ctx;
159         unsigned long   page_index;
160         u64             *dir_cookie;
161         u64             last_cookie;
162         loff_t          current_index;
163         decode_dirent_t decode;
164 
165         unsigned long   timestamp;
166         unsigned long   gencount;
167         unsigned int    cache_entry_index;
168         unsigned int    plus:1;
169         unsigned int    eof:1;
170 } nfs_readdir_descriptor_t;
171 
172 /*
173  * The caller is responsible for calling nfs_readdir_release_array(page)
174  */
175 static
176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
177 {
178         void *ptr;
179         if (page == NULL)
180                 return ERR_PTR(-EIO);
181         ptr = kmap(page);
182         if (ptr == NULL)
183                 return ERR_PTR(-ENOMEM);
184         return ptr;
185 }
186 
187 static
188 void nfs_readdir_release_array(struct page *page)
189 {
190         kunmap(page);
191 }
192 
193 /*
194  * we are freeing strings created by nfs_add_to_readdir_array()
195  */
196 static
197 void nfs_readdir_clear_array(struct page *page)
198 {
199         struct nfs_cache_array *array;
200         int i;
201 
202         array = kmap_atomic(page);
203         for (i = 0; i < array->size; i++)
204                 kfree(array->array[i].string.name);
205         kunmap_atomic(array);
206 }
207 
208 /*
209  * the caller is responsible for freeing qstr.name
210  * when called by nfs_readdir_add_to_array, the strings will be freed in
211  * nfs_clear_readdir_array()
212  */
213 static
214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
215 {
216         string->len = len;
217         string->name = kmemdup(name, len, GFP_KERNEL);
218         if (string->name == NULL)
219                 return -ENOMEM;
220         /*
221          * Avoid a kmemleak false positive. The pointer to the name is stored
222          * in a page cache page which kmemleak does not scan.
223          */
224         kmemleak_not_leak(string->name);
225         string->hash = full_name_hash(name, len);
226         return 0;
227 }
228 
229 static
230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
231 {
232         struct nfs_cache_array *array = nfs_readdir_get_array(page);
233         struct nfs_cache_array_entry *cache_entry;
234         int ret;
235 
236         if (IS_ERR(array))
237                 return PTR_ERR(array);
238 
239         cache_entry = &array->array[array->size];
240 
241         /* Check that this entry lies within the page bounds */
242         ret = -ENOSPC;
243         if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
244                 goto out;
245 
246         cache_entry->cookie = entry->prev_cookie;
247         cache_entry->ino = entry->ino;
248         cache_entry->d_type = entry->d_type;
249         ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
250         if (ret)
251                 goto out;
252         array->last_cookie = entry->cookie;
253         array->size++;
254         if (entry->eof != 0)
255                 array->eof_index = array->size;
256 out:
257         nfs_readdir_release_array(page);
258         return ret;
259 }
260 
261 static
262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
263 {
264         loff_t diff = desc->ctx->pos - desc->current_index;
265         unsigned int index;
266 
267         if (diff < 0)
268                 goto out_eof;
269         if (diff >= array->size) {
270                 if (array->eof_index >= 0)
271                         goto out_eof;
272                 return -EAGAIN;
273         }
274 
275         index = (unsigned int)diff;
276         *desc->dir_cookie = array->array[index].cookie;
277         desc->cache_entry_index = index;
278         return 0;
279 out_eof:
280         desc->eof = 1;
281         return -EBADCOOKIE;
282 }
283 
284 static bool
285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
286 {
287         if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
288                 return false;
289         smp_rmb();
290         return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
291 }
292 
293 static
294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
295 {
296         int i;
297         loff_t new_pos;
298         int status = -EAGAIN;
299 
300         for (i = 0; i < array->size; i++) {
301                 if (array->array[i].cookie == *desc->dir_cookie) {
302                         struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303                         struct nfs_open_dir_context *ctx = desc->file->private_data;
304 
305                         new_pos = desc->current_index + i;
306                         if (ctx->attr_gencount != nfsi->attr_gencount ||
307                             !nfs_readdir_inode_mapping_valid(nfsi)) {
308                                 ctx->duped = 0;
309                                 ctx->attr_gencount = nfsi->attr_gencount;
310                         } else if (new_pos < desc->ctx->pos) {
311                                 if (ctx->duped > 0
312                                     && ctx->dup_cookie == *desc->dir_cookie) {
313                                         if (printk_ratelimit()) {
314                                                 pr_notice("NFS: directory %pD2 contains a readdir loop."
315                                                                 "Please contact your server vendor.  "
316                                                                 "The file: %.*s has duplicate cookie %llu\n",
317                                                                 desc->file, array->array[i].string.len,
318                                                                 array->array[i].string.name, *desc->dir_cookie);
319                                         }
320                                         status = -ELOOP;
321                                         goto out;
322                                 }
323                                 ctx->dup_cookie = *desc->dir_cookie;
324                                 ctx->duped = -1;
325                         }
326                         desc->ctx->pos = new_pos;
327                         desc->cache_entry_index = i;
328                         return 0;
329                 }
330         }
331         if (array->eof_index >= 0) {
332                 status = -EBADCOOKIE;
333                 if (*desc->dir_cookie == array->last_cookie)
334                         desc->eof = 1;
335         }
336 out:
337         return status;
338 }
339 
340 static
341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
342 {
343         struct nfs_cache_array *array;
344         int status;
345 
346         array = nfs_readdir_get_array(desc->page);
347         if (IS_ERR(array)) {
348                 status = PTR_ERR(array);
349                 goto out;
350         }
351 
352         if (*desc->dir_cookie == 0)
353                 status = nfs_readdir_search_for_pos(array, desc);
354         else
355                 status = nfs_readdir_search_for_cookie(array, desc);
356 
357         if (status == -EAGAIN) {
358                 desc->last_cookie = array->last_cookie;
359                 desc->current_index += array->size;
360                 desc->page_index++;
361         }
362         nfs_readdir_release_array(desc->page);
363 out:
364         return status;
365 }
366 
367 /* Fill a page with xdr information before transferring to the cache page */
368 static
369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
370                         struct nfs_entry *entry, struct file *file, struct inode *inode)
371 {
372         struct nfs_open_dir_context *ctx = file->private_data;
373         struct rpc_cred *cred = ctx->cred;
374         unsigned long   timestamp, gencount;
375         int             error;
376 
377  again:
378         timestamp = jiffies;
379         gencount = nfs_inc_attr_generation_counter();
380         error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
381                                           NFS_SERVER(inode)->dtsize, desc->plus);
382         if (error < 0) {
383                 /* We requested READDIRPLUS, but the server doesn't grok it */
384                 if (error == -ENOTSUPP && desc->plus) {
385                         NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
387                         desc->plus = 0;
388                         goto again;
389                 }
390                 goto error;
391         }
392         desc->timestamp = timestamp;
393         desc->gencount = gencount;
394 error:
395         return error;
396 }
397 
398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399                       struct nfs_entry *entry, struct xdr_stream *xdr)
400 {
401         int error;
402 
403         error = desc->decode(xdr, entry, desc->plus);
404         if (error)
405                 return error;
406         entry->fattr->time_start = desc->timestamp;
407         entry->fattr->gencount = desc->gencount;
408         return 0;
409 }
410 
411 /* Match file and dirent using either filehandle or fileid
412  * Note: caller is responsible for checking the fsid
413  */
414 static
415 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
416 {
417         struct nfs_inode *nfsi;
418 
419         if (d_really_is_negative(dentry))
420                 return 0;
421 
422         nfsi = NFS_I(d_inode(dentry));
423         if (entry->fattr->fileid == nfsi->fileid)
424                 return 1;
425         if (nfs_compare_fh(entry->fh, &nfsi->fh) == 0)
426                 return 1;
427         return 0;
428 }
429 
430 static
431 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
432 {
433         if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
434                 return false;
435         if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
436                 return true;
437         if (ctx->pos == 0)
438                 return true;
439         return false;
440 }
441 
442 /*
443  * This function is called by the lookup code to request the use of
444  * readdirplus to accelerate any future lookups in the same
445  * directory.
446  */
447 static
448 void nfs_advise_use_readdirplus(struct inode *dir)
449 {
450         set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
451 }
452 
453 /*
454  * This function is mainly for use by nfs_getattr().
455  *
456  * If this is an 'ls -l', we want to force use of readdirplus.
457  * Do this by checking if there is an active file descriptor
458  * and calling nfs_advise_use_readdirplus, then forcing a
459  * cache flush.
460  */
461 void nfs_force_use_readdirplus(struct inode *dir)
462 {
463         if (!list_empty(&NFS_I(dir)->open_files)) {
464                 nfs_advise_use_readdirplus(dir);
465                 nfs_zap_mapping(dir, dir->i_mapping);
466         }
467 }
468 
469 static
470 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
471 {
472         struct qstr filename = QSTR_INIT(entry->name, entry->len);
473         struct dentry *dentry;
474         struct dentry *alias;
475         struct inode *dir = d_inode(parent);
476         struct inode *inode;
477         int status;
478 
479         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
480                 return;
481         if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
482                 return;
483         if (filename.name[0] == '.') {
484                 if (filename.len == 1)
485                         return;
486                 if (filename.len == 2 && filename.name[1] == '.')
487                         return;
488         }
489         filename.hash = full_name_hash(filename.name, filename.len);
490 
491         dentry = d_lookup(parent, &filename);
492         if (dentry != NULL) {
493                 /* Is there a mountpoint here? If so, just exit */
494                 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
495                                         &entry->fattr->fsid))
496                         goto out;
497                 if (nfs_same_file(dentry, entry)) {
498                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
499                         status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
500                         if (!status)
501                                 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
502                         goto out;
503                 } else {
504                         d_invalidate(dentry);
505                         dput(dentry);
506                 }
507         }
508 
509         dentry = d_alloc(parent, &filename);
510         if (dentry == NULL)
511                 return;
512 
513         inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
514         if (IS_ERR(inode))
515                 goto out;
516 
517         alias = d_splice_alias(inode, dentry);
518         if (IS_ERR(alias))
519                 goto out;
520         else if (alias) {
521                 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
522                 dput(alias);
523         } else
524                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
525 
526 out:
527         dput(dentry);
528 }
529 
530 /* Perform conversion from xdr to cache array */
531 static
532 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
533                                 struct page **xdr_pages, struct page *page, unsigned int buflen)
534 {
535         struct xdr_stream stream;
536         struct xdr_buf buf;
537         struct page *scratch;
538         struct nfs_cache_array *array;
539         unsigned int count = 0;
540         int status;
541 
542         scratch = alloc_page(GFP_KERNEL);
543         if (scratch == NULL)
544                 return -ENOMEM;
545 
546         if (buflen == 0)
547                 goto out_nopages;
548 
549         xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
550         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
551 
552         do {
553                 status = xdr_decode(desc, entry, &stream);
554                 if (status != 0) {
555                         if (status == -EAGAIN)
556                                 status = 0;
557                         break;
558                 }
559 
560                 count++;
561 
562                 if (desc->plus != 0)
563                         nfs_prime_dcache(file_dentry(desc->file), entry);
564 
565                 status = nfs_readdir_add_to_array(entry, page);
566                 if (status != 0)
567                         break;
568         } while (!entry->eof);
569 
570 out_nopages:
571         if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
572                 array = nfs_readdir_get_array(page);
573                 if (!IS_ERR(array)) {
574                         array->eof_index = array->size;
575                         status = 0;
576                         nfs_readdir_release_array(page);
577                 } else
578                         status = PTR_ERR(array);
579         }
580 
581         put_page(scratch);
582         return status;
583 }
584 
585 static
586 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
587 {
588         unsigned int i;
589         for (i = 0; i < npages; i++)
590                 put_page(pages[i]);
591 }
592 
593 /*
594  * nfs_readdir_large_page will allocate pages that must be freed with a call
595  * to nfs_readdir_free_pagearray
596  */
597 static
598 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
599 {
600         unsigned int i;
601 
602         for (i = 0; i < npages; i++) {
603                 struct page *page = alloc_page(GFP_KERNEL);
604                 if (page == NULL)
605                         goto out_freepages;
606                 pages[i] = page;
607         }
608         return 0;
609 
610 out_freepages:
611         nfs_readdir_free_pages(pages, i);
612         return -ENOMEM;
613 }
614 
615 static
616 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
617 {
618         struct page *pages[NFS_MAX_READDIR_PAGES];
619         struct nfs_entry entry;
620         struct file     *file = desc->file;
621         struct nfs_cache_array *array;
622         int status = -ENOMEM;
623         unsigned int array_size = ARRAY_SIZE(pages);
624 
625         entry.prev_cookie = 0;
626         entry.cookie = desc->last_cookie;
627         entry.eof = 0;
628         entry.fh = nfs_alloc_fhandle();
629         entry.fattr = nfs_alloc_fattr();
630         entry.server = NFS_SERVER(inode);
631         if (entry.fh == NULL || entry.fattr == NULL)
632                 goto out;
633 
634         entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
635         if (IS_ERR(entry.label)) {
636                 status = PTR_ERR(entry.label);
637                 goto out;
638         }
639 
640         array = nfs_readdir_get_array(page);
641         if (IS_ERR(array)) {
642                 status = PTR_ERR(array);
643                 goto out_label_free;
644         }
645         memset(array, 0, sizeof(struct nfs_cache_array));
646         array->eof_index = -1;
647 
648         status = nfs_readdir_alloc_pages(pages, array_size);
649         if (status < 0)
650                 goto out_release_array;
651         do {
652                 unsigned int pglen;
653                 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
654 
655                 if (status < 0)
656                         break;
657                 pglen = status;
658                 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
659                 if (status < 0) {
660                         if (status == -ENOSPC)
661                                 status = 0;
662                         break;
663                 }
664         } while (array->eof_index < 0);
665 
666         nfs_readdir_free_pages(pages, array_size);
667 out_release_array:
668         nfs_readdir_release_array(page);
669 out_label_free:
670         nfs4_label_free(entry.label);
671 out:
672         nfs_free_fattr(entry.fattr);
673         nfs_free_fhandle(entry.fh);
674         return status;
675 }
676 
677 /*
678  * Now we cache directories properly, by converting xdr information
679  * to an array that can be used for lookups later.  This results in
680  * fewer cache pages, since we can store more information on each page.
681  * We only need to convert from xdr once so future lookups are much simpler
682  */
683 static
684 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
685 {
686         struct inode    *inode = file_inode(desc->file);
687         int ret;
688 
689         ret = nfs_readdir_xdr_to_array(desc, page, inode);
690         if (ret < 0)
691                 goto error;
692         SetPageUptodate(page);
693 
694         if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
695                 /* Should never happen */
696                 nfs_zap_mapping(inode, inode->i_mapping);
697         }
698         unlock_page(page);
699         return 0;
700  error:
701         unlock_page(page);
702         return ret;
703 }
704 
705 static
706 void cache_page_release(nfs_readdir_descriptor_t *desc)
707 {
708         if (!desc->page->mapping)
709                 nfs_readdir_clear_array(desc->page);
710         put_page(desc->page);
711         desc->page = NULL;
712 }
713 
714 static
715 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
716 {
717         return read_cache_page(file_inode(desc->file)->i_mapping,
718                         desc->page_index, (filler_t *)nfs_readdir_filler, desc);
719 }
720 
721 /*
722  * Returns 0 if desc->dir_cookie was found on page desc->page_index
723  */
724 static
725 int find_cache_page(nfs_readdir_descriptor_t *desc)
726 {
727         int res;
728 
729         desc->page = get_cache_page(desc);
730         if (IS_ERR(desc->page))
731                 return PTR_ERR(desc->page);
732 
733         res = nfs_readdir_search_array(desc);
734         if (res != 0)
735                 cache_page_release(desc);
736         return res;
737 }
738 
739 /* Search for desc->dir_cookie from the beginning of the page cache */
740 static inline
741 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
742 {
743         int res;
744 
745         if (desc->page_index == 0) {
746                 desc->current_index = 0;
747                 desc->last_cookie = 0;
748         }
749         do {
750                 res = find_cache_page(desc);
751         } while (res == -EAGAIN);
752         return res;
753 }
754 
755 /*
756  * Once we've found the start of the dirent within a page: fill 'er up...
757  */
758 static 
759 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
760 {
761         struct file     *file = desc->file;
762         int i = 0;
763         int res = 0;
764         struct nfs_cache_array *array = NULL;
765         struct nfs_open_dir_context *ctx = file->private_data;
766 
767         array = nfs_readdir_get_array(desc->page);
768         if (IS_ERR(array)) {
769                 res = PTR_ERR(array);
770                 goto out;
771         }
772 
773         for (i = desc->cache_entry_index; i < array->size; i++) {
774                 struct nfs_cache_array_entry *ent;
775 
776                 ent = &array->array[i];
777                 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
778                     nfs_compat_user_ino64(ent->ino), ent->d_type)) {
779                         desc->eof = 1;
780                         break;
781                 }
782                 desc->ctx->pos++;
783                 if (i < (array->size-1))
784                         *desc->dir_cookie = array->array[i+1].cookie;
785                 else
786                         *desc->dir_cookie = array->last_cookie;
787                 if (ctx->duped != 0)
788                         ctx->duped = 1;
789         }
790         if (array->eof_index >= 0)
791                 desc->eof = 1;
792 
793         nfs_readdir_release_array(desc->page);
794 out:
795         cache_page_release(desc);
796         dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
797                         (unsigned long long)*desc->dir_cookie, res);
798         return res;
799 }
800 
801 /*
802  * If we cannot find a cookie in our cache, we suspect that this is
803  * because it points to a deleted file, so we ask the server to return
804  * whatever it thinks is the next entry. We then feed this to filldir.
805  * If all goes well, we should then be able to find our way round the
806  * cache on the next call to readdir_search_pagecache();
807  *
808  * NOTE: we cannot add the anonymous page to the pagecache because
809  *       the data it contains might not be page aligned. Besides,
810  *       we should already have a complete representation of the
811  *       directory in the page cache by the time we get here.
812  */
813 static inline
814 int uncached_readdir(nfs_readdir_descriptor_t *desc)
815 {
816         struct page     *page = NULL;
817         int             status;
818         struct inode *inode = file_inode(desc->file);
819         struct nfs_open_dir_context *ctx = desc->file->private_data;
820 
821         dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
822                         (unsigned long long)*desc->dir_cookie);
823 
824         page = alloc_page(GFP_HIGHUSER);
825         if (!page) {
826                 status = -ENOMEM;
827                 goto out;
828         }
829 
830         desc->page_index = 0;
831         desc->last_cookie = *desc->dir_cookie;
832         desc->page = page;
833         ctx->duped = 0;
834 
835         status = nfs_readdir_xdr_to_array(desc, page, inode);
836         if (status < 0)
837                 goto out_release;
838 
839         status = nfs_do_filldir(desc);
840 
841  out:
842         dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
843                         __func__, status);
844         return status;
845  out_release:
846         cache_page_release(desc);
847         goto out;
848 }
849 
850 static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
851 {
852         struct nfs_inode *nfsi = NFS_I(dir);
853 
854         if (nfs_attribute_cache_expired(dir))
855                 return true;
856         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
857                 return true;
858         return false;
859 }
860 
861 /* The file offset position represents the dirent entry number.  A
862    last cookie cache takes care of the common case of reading the
863    whole directory.
864  */
865 static int nfs_readdir(struct file *file, struct dir_context *ctx)
866 {
867         struct dentry   *dentry = file_dentry(file);
868         struct inode    *inode = d_inode(dentry);
869         nfs_readdir_descriptor_t my_desc,
870                         *desc = &my_desc;
871         struct nfs_open_dir_context *dir_ctx = file->private_data;
872         int res = 0;
873 
874         dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
875                         file, (long long)ctx->pos);
876         nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
877 
878         /*
879          * ctx->pos points to the dirent entry number.
880          * *desc->dir_cookie has the cookie for the next entry. We have
881          * to either find the entry with the appropriate number or
882          * revalidate the cookie.
883          */
884         memset(desc, 0, sizeof(*desc));
885 
886         desc->file = file;
887         desc->ctx = ctx;
888         desc->dir_cookie = &dir_ctx->dir_cookie;
889         desc->decode = NFS_PROTO(inode)->decode_dirent;
890         desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
891 
892         nfs_block_sillyrename(dentry);
893         if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
894                 res = nfs_revalidate_mapping(inode, file->f_mapping);
895         if (res < 0)
896                 goto out;
897 
898         do {
899                 res = readdir_search_pagecache(desc);
900 
901                 if (res == -EBADCOOKIE) {
902                         res = 0;
903                         /* This means either end of directory */
904                         if (*desc->dir_cookie && desc->eof == 0) {
905                                 /* Or that the server has 'lost' a cookie */
906                                 res = uncached_readdir(desc);
907                                 if (res == 0)
908                                         continue;
909                         }
910                         break;
911                 }
912                 if (res == -ETOOSMALL && desc->plus) {
913                         clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
914                         nfs_zap_caches(inode);
915                         desc->page_index = 0;
916                         desc->plus = 0;
917                         desc->eof = 0;
918                         continue;
919                 }
920                 if (res < 0)
921                         break;
922 
923                 res = nfs_do_filldir(desc);
924                 if (res < 0)
925                         break;
926         } while (!desc->eof);
927 out:
928         nfs_unblock_sillyrename(dentry);
929         if (res > 0)
930                 res = 0;
931         dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
932         return res;
933 }
934 
935 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
936 {
937         struct inode *inode = file_inode(filp);
938         struct nfs_open_dir_context *dir_ctx = filp->private_data;
939 
940         dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
941                         filp, offset, whence);
942 
943         inode_lock(inode);
944         switch (whence) {
945                 case 1:
946                         offset += filp->f_pos;
947                 case 0:
948                         if (offset >= 0)
949                                 break;
950                 default:
951                         offset = -EINVAL;
952                         goto out;
953         }
954         if (offset != filp->f_pos) {
955                 filp->f_pos = offset;
956                 dir_ctx->dir_cookie = 0;
957                 dir_ctx->duped = 0;
958         }
959 out:
960         inode_unlock(inode);
961         return offset;
962 }
963 
964 /*
965  * All directory operations under NFS are synchronous, so fsync()
966  * is a dummy operation.
967  */
968 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
969                          int datasync)
970 {
971         struct inode *inode = file_inode(filp);
972 
973         dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
974 
975         inode_lock(inode);
976         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
977         inode_unlock(inode);
978         return 0;
979 }
980 
981 /**
982  * nfs_force_lookup_revalidate - Mark the directory as having changed
983  * @dir - pointer to directory inode
984  *
985  * This forces the revalidation code in nfs_lookup_revalidate() to do a
986  * full lookup on all child dentries of 'dir' whenever a change occurs
987  * on the server that might have invalidated our dcache.
988  *
989  * The caller should be holding dir->i_lock
990  */
991 void nfs_force_lookup_revalidate(struct inode *dir)
992 {
993         NFS_I(dir)->cache_change_attribute++;
994 }
995 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
996 
997 /*
998  * A check for whether or not the parent directory has changed.
999  * In the case it has, we assume that the dentries are untrustworthy
1000  * and may need to be looked up again.
1001  * If rcu_walk prevents us from performing a full check, return 0.
1002  */
1003 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1004                               int rcu_walk)
1005 {
1006         int ret;
1007 
1008         if (IS_ROOT(dentry))
1009                 return 1;
1010         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1011                 return 0;
1012         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1013                 return 0;
1014         /* Revalidate nfsi->cache_change_attribute before we declare a match */
1015         if (rcu_walk)
1016                 ret = nfs_revalidate_inode_rcu(NFS_SERVER(dir), dir);
1017         else
1018                 ret = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1019         if (ret < 0)
1020                 return 0;
1021         if (!nfs_verify_change_attribute(dir, dentry->d_time))
1022                 return 0;
1023         return 1;
1024 }
1025 
1026 /*
1027  * Use intent information to check whether or not we're going to do
1028  * an O_EXCL create using this path component.
1029  */
1030 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1031 {
1032         if (NFS_PROTO(dir)->version == 2)
1033                 return 0;
1034         return flags & LOOKUP_EXCL;
1035 }
1036 
1037 /*
1038  * Inode and filehandle revalidation for lookups.
1039  *
1040  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1041  * or if the intent information indicates that we're about to open this
1042  * particular file and the "nocto" mount flag is not set.
1043  *
1044  */
1045 static
1046 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1047 {
1048         struct nfs_server *server = NFS_SERVER(inode);
1049         int ret;
1050 
1051         if (IS_AUTOMOUNT(inode))
1052                 return 0;
1053         /* VFS wants an on-the-wire revalidation */
1054         if (flags & LOOKUP_REVAL)
1055                 goto out_force;
1056         /* This is an open(2) */
1057         if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1058             (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1059                 goto out_force;
1060 out:
1061         return (inode->i_nlink == 0) ? -ENOENT : 0;
1062 out_force:
1063         if (flags & LOOKUP_RCU)
1064                 return -ECHILD;
1065         ret = __nfs_revalidate_inode(server, inode);
1066         if (ret != 0)
1067                 return ret;
1068         goto out;
1069 }
1070 
1071 /*
1072  * We judge how long we want to trust negative
1073  * dentries by looking at the parent inode mtime.
1074  *
1075  * If parent mtime has changed, we revalidate, else we wait for a
1076  * period corresponding to the parent's attribute cache timeout value.
1077  *
1078  * If LOOKUP_RCU prevents us from performing a full check, return 1
1079  * suggesting a reval is needed.
1080  */
1081 static inline
1082 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1083                        unsigned int flags)
1084 {
1085         /* Don't revalidate a negative dentry if we're creating a new file */
1086         if (flags & LOOKUP_CREATE)
1087                 return 0;
1088         if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1089                 return 1;
1090         return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1091 }
1092 
1093 /*
1094  * This is called every time the dcache has a lookup hit,
1095  * and we should check whether we can really trust that
1096  * lookup.
1097  *
1098  * NOTE! The hit can be a negative hit too, don't assume
1099  * we have an inode!
1100  *
1101  * If the parent directory is seen to have changed, we throw out the
1102  * cached dentry and do a new lookup.
1103  */
1104 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1105 {
1106         struct inode *dir;
1107         struct inode *inode;
1108         struct dentry *parent;
1109         struct nfs_fh *fhandle = NULL;
1110         struct nfs_fattr *fattr = NULL;
1111         struct nfs4_label *label = NULL;
1112         int error;
1113 
1114         if (flags & LOOKUP_RCU) {
1115                 parent = ACCESS_ONCE(dentry->d_parent);
1116                 dir = d_inode_rcu(parent);
1117                 if (!dir)
1118                         return -ECHILD;
1119         } else {
1120                 parent = dget_parent(dentry);
1121                 dir = d_inode(parent);
1122         }
1123         nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1124         inode = d_inode(dentry);
1125 
1126         if (!inode) {
1127                 if (nfs_neg_need_reval(dir, dentry, flags)) {
1128                         if (flags & LOOKUP_RCU)
1129                                 return -ECHILD;
1130                         goto out_bad;
1131                 }
1132                 goto out_valid_noent;
1133         }
1134 
1135         if (is_bad_inode(inode)) {
1136                 if (flags & LOOKUP_RCU)
1137                         return -ECHILD;
1138                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1139                                 __func__, dentry);
1140                 goto out_bad;
1141         }
1142 
1143         if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1144                 goto out_set_verifier;
1145 
1146         /* Force a full look up iff the parent directory has changed */
1147         if (!nfs_is_exclusive_create(dir, flags) &&
1148             nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1149 
1150                 if (nfs_lookup_verify_inode(inode, flags)) {
1151                         if (flags & LOOKUP_RCU)
1152                                 return -ECHILD;
1153                         goto out_zap_parent;
1154                 }
1155                 goto out_valid;
1156         }
1157 
1158         if (flags & LOOKUP_RCU)
1159                 return -ECHILD;
1160 
1161         if (NFS_STALE(inode))
1162                 goto out_bad;
1163 
1164         error = -ENOMEM;
1165         fhandle = nfs_alloc_fhandle();
1166         fattr = nfs_alloc_fattr();
1167         if (fhandle == NULL || fattr == NULL)
1168                 goto out_error;
1169 
1170         label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1171         if (IS_ERR(label))
1172                 goto out_error;
1173 
1174         trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1175         error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1176         trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1177         if (error)
1178                 goto out_bad;
1179         if (nfs_compare_fh(NFS_FH(inode), fhandle))
1180                 goto out_bad;
1181         if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1182                 goto out_bad;
1183 
1184         nfs_setsecurity(inode, fattr, label);
1185 
1186         nfs_free_fattr(fattr);
1187         nfs_free_fhandle(fhandle);
1188         nfs4_label_free(label);
1189 
1190 out_set_verifier:
1191         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1192  out_valid:
1193         /* Success: notify readdir to use READDIRPLUS */
1194         nfs_advise_use_readdirplus(dir);
1195  out_valid_noent:
1196         if (flags & LOOKUP_RCU) {
1197                 if (parent != ACCESS_ONCE(dentry->d_parent))
1198                         return -ECHILD;
1199         } else
1200                 dput(parent);
1201         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1202                         __func__, dentry);
1203         return 1;
1204 out_zap_parent:
1205         nfs_zap_caches(dir);
1206  out_bad:
1207         WARN_ON(flags & LOOKUP_RCU);
1208         nfs_free_fattr(fattr);
1209         nfs_free_fhandle(fhandle);
1210         nfs4_label_free(label);
1211         nfs_mark_for_revalidate(dir);
1212         if (inode && S_ISDIR(inode->i_mode)) {
1213                 /* Purge readdir caches. */
1214                 nfs_zap_caches(inode);
1215                 /*
1216                  * We can't d_drop the root of a disconnected tree:
1217                  * its d_hash is on the s_anon list and d_drop() would hide
1218                  * it from shrink_dcache_for_unmount(), leading to busy
1219                  * inodes on unmount and further oopses.
1220                  */
1221                 if (IS_ROOT(dentry))
1222                         goto out_valid;
1223         }
1224         dput(parent);
1225         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1226                         __func__, dentry);
1227         return 0;
1228 out_error:
1229         WARN_ON(flags & LOOKUP_RCU);
1230         nfs_free_fattr(fattr);
1231         nfs_free_fhandle(fhandle);
1232         nfs4_label_free(label);
1233         dput(parent);
1234         dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1235                         __func__, dentry, error);
1236         return error;
1237 }
1238 
1239 /*
1240  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1241  * when we don't really care about the dentry name. This is called when a
1242  * pathwalk ends on a dentry that was not found via a normal lookup in the
1243  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1244  *
1245  * In this situation, we just want to verify that the inode itself is OK
1246  * since the dentry might have changed on the server.
1247  */
1248 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1249 {
1250         int error;
1251         struct inode *inode = d_inode(dentry);
1252 
1253         /*
1254          * I believe we can only get a negative dentry here in the case of a
1255          * procfs-style symlink. Just assume it's correct for now, but we may
1256          * eventually need to do something more here.
1257          */
1258         if (!inode) {
1259                 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1260                                 __func__, dentry);
1261                 return 1;
1262         }
1263 
1264         if (is_bad_inode(inode)) {
1265                 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1266                                 __func__, dentry);
1267                 return 0;
1268         }
1269 
1270         error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1271         dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1272                         __func__, inode->i_ino, error ? "invalid" : "valid");
1273         return !error;
1274 }
1275 
1276 /*
1277  * This is called from dput() when d_count is going to 0.
1278  */
1279 static int nfs_dentry_delete(const struct dentry *dentry)
1280 {
1281         dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1282                 dentry, dentry->d_flags);
1283 
1284         /* Unhash any dentry with a stale inode */
1285         if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1286                 return 1;
1287 
1288         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1289                 /* Unhash it, so that ->d_iput() would be called */
1290                 return 1;
1291         }
1292         if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1293                 /* Unhash it, so that ancestors of killed async unlink
1294                  * files will be cleaned up during umount */
1295                 return 1;
1296         }
1297         return 0;
1298 
1299 }
1300 
1301 /* Ensure that we revalidate inode->i_nlink */
1302 static void nfs_drop_nlink(struct inode *inode)
1303 {
1304         spin_lock(&inode->i_lock);
1305         /* drop the inode if we're reasonably sure this is the last link */
1306         if (inode->i_nlink == 1)
1307                 clear_nlink(inode);
1308         NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1309         spin_unlock(&inode->i_lock);
1310 }
1311 
1312 /*
1313  * Called when the dentry loses inode.
1314  * We use it to clean up silly-renamed files.
1315  */
1316 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1317 {
1318         if (S_ISDIR(inode->i_mode))
1319                 /* drop any readdir cache as it could easily be old */
1320                 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1321 
1322         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1323                 nfs_complete_unlink(dentry, inode);
1324                 nfs_drop_nlink(inode);
1325         }
1326         iput(inode);
1327 }
1328 
1329 static void nfs_d_release(struct dentry *dentry)
1330 {
1331         /* free cached devname value, if it survived that far */
1332         if (unlikely(dentry->d_fsdata)) {
1333                 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1334                         WARN_ON(1);
1335                 else
1336                         kfree(dentry->d_fsdata);
1337         }
1338 }
1339 
1340 const struct dentry_operations nfs_dentry_operations = {
1341         .d_revalidate   = nfs_lookup_revalidate,
1342         .d_weak_revalidate      = nfs_weak_revalidate,
1343         .d_delete       = nfs_dentry_delete,
1344         .d_iput         = nfs_dentry_iput,
1345         .d_automount    = nfs_d_automount,
1346         .d_release      = nfs_d_release,
1347 };
1348 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1349 
1350 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1351 {
1352         struct dentry *res;
1353         struct dentry *parent;
1354         struct inode *inode = NULL;
1355         struct nfs_fh *fhandle = NULL;
1356         struct nfs_fattr *fattr = NULL;
1357         struct nfs4_label *label = NULL;
1358         int error;
1359 
1360         dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1361         nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1362 
1363         if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1364                 return ERR_PTR(-ENAMETOOLONG);
1365 
1366         /*
1367          * If we're doing an exclusive create, optimize away the lookup
1368          * but don't hash the dentry.
1369          */
1370         if (nfs_is_exclusive_create(dir, flags))
1371                 return NULL;
1372 
1373         res = ERR_PTR(-ENOMEM);
1374         fhandle = nfs_alloc_fhandle();
1375         fattr = nfs_alloc_fattr();
1376         if (fhandle == NULL || fattr == NULL)
1377                 goto out;
1378 
1379         label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1380         if (IS_ERR(label))
1381                 goto out;
1382 
1383         parent = dentry->d_parent;
1384         /* Protect against concurrent sillydeletes */
1385         trace_nfs_lookup_enter(dir, dentry, flags);
1386         nfs_block_sillyrename(parent);
1387         error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1388         if (error == -ENOENT)
1389                 goto no_entry;
1390         if (error < 0) {
1391                 res = ERR_PTR(error);
1392                 goto out_unblock_sillyrename;
1393         }
1394         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1395         res = ERR_CAST(inode);
1396         if (IS_ERR(res))
1397                 goto out_unblock_sillyrename;
1398 
1399         /* Success: notify readdir to use READDIRPLUS */
1400         nfs_advise_use_readdirplus(dir);
1401 
1402 no_entry:
1403         res = d_splice_alias(inode, dentry);
1404         if (res != NULL) {
1405                 if (IS_ERR(res))
1406                         goto out_unblock_sillyrename;
1407                 dentry = res;
1408         }
1409         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1410 out_unblock_sillyrename:
1411         nfs_unblock_sillyrename(parent);
1412         trace_nfs_lookup_exit(dir, dentry, flags, error);
1413         nfs4_label_free(label);
1414 out:
1415         nfs_free_fattr(fattr);
1416         nfs_free_fhandle(fhandle);
1417         return res;
1418 }
1419 EXPORT_SYMBOL_GPL(nfs_lookup);
1420 
1421 #if IS_ENABLED(CONFIG_NFS_V4)
1422 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1423 
1424 const struct dentry_operations nfs4_dentry_operations = {
1425         .d_revalidate   = nfs4_lookup_revalidate,
1426         .d_delete       = nfs_dentry_delete,
1427         .d_iput         = nfs_dentry_iput,
1428         .d_automount    = nfs_d_automount,
1429         .d_release      = nfs_d_release,
1430 };
1431 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1432 
1433 static fmode_t flags_to_mode(int flags)
1434 {
1435         fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1436         if ((flags & O_ACCMODE) != O_WRONLY)
1437                 res |= FMODE_READ;
1438         if ((flags & O_ACCMODE) != O_RDONLY)
1439                 res |= FMODE_WRITE;
1440         return res;
1441 }
1442 
1443 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1444 {
1445         return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1446 }
1447 
1448 static int do_open(struct inode *inode, struct file *filp)
1449 {
1450         nfs_fscache_open_file(inode, filp);
1451         return 0;
1452 }
1453 
1454 static int nfs_finish_open(struct nfs_open_context *ctx,
1455                            struct dentry *dentry,
1456                            struct file *file, unsigned open_flags,
1457                            int *opened)
1458 {
1459         int err;
1460 
1461         err = finish_open(file, dentry, do_open, opened);
1462         if (err)
1463                 goto out;
1464         nfs_file_set_open_context(file, ctx);
1465 
1466 out:
1467         return err;
1468 }
1469 
1470 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1471                     struct file *file, unsigned open_flags,
1472                     umode_t mode, int *opened)
1473 {
1474         struct nfs_open_context *ctx;
1475         struct dentry *res;
1476         struct iattr attr = { .ia_valid = ATTR_OPEN };
1477         struct inode *inode;
1478         unsigned int lookup_flags = 0;
1479         int err;
1480 
1481         /* Expect a negative dentry */
1482         BUG_ON(d_inode(dentry));
1483 
1484         dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1485                         dir->i_sb->s_id, dir->i_ino, dentry);
1486 
1487         err = nfs_check_flags(open_flags);
1488         if (err)
1489                 return err;
1490 
1491         /* NFS only supports OPEN on regular files */
1492         if ((open_flags & O_DIRECTORY)) {
1493                 if (!d_unhashed(dentry)) {
1494                         /*
1495                          * Hashed negative dentry with O_DIRECTORY: dentry was
1496                          * revalidated and is fine, no need to perform lookup
1497                          * again
1498                          */
1499                         return -ENOENT;
1500                 }
1501                 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1502                 goto no_open;
1503         }
1504 
1505         if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1506                 return -ENAMETOOLONG;
1507 
1508         if (open_flags & O_CREAT) {
1509                 attr.ia_valid |= ATTR_MODE;
1510                 attr.ia_mode = mode & ~current_umask();
1511         }
1512         if (open_flags & O_TRUNC) {
1513                 attr.ia_valid |= ATTR_SIZE;
1514                 attr.ia_size = 0;
1515         }
1516 
1517         ctx = create_nfs_open_context(dentry, open_flags);
1518         err = PTR_ERR(ctx);
1519         if (IS_ERR(ctx))
1520                 goto out;
1521 
1522         trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1523         nfs_block_sillyrename(dentry->d_parent);
1524         inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1525         nfs_unblock_sillyrename(dentry->d_parent);
1526         if (IS_ERR(inode)) {
1527                 err = PTR_ERR(inode);
1528                 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1529                 put_nfs_open_context(ctx);
1530                 d_drop(dentry);
1531                 switch (err) {
1532                 case -ENOENT:
1533                         d_add(dentry, NULL);
1534                         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1535                         break;
1536                 case -EISDIR:
1537                 case -ENOTDIR:
1538                         goto no_open;
1539                 case -ELOOP:
1540                         if (!(open_flags & O_NOFOLLOW))
1541                                 goto no_open;
1542                         break;
1543                         /* case -EINVAL: */
1544                 default:
1545                         break;
1546                 }
1547                 goto out;
1548         }
1549 
1550         err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1551         trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1552         put_nfs_open_context(ctx);
1553 out:
1554         return err;
1555 
1556 no_open:
1557         res = nfs_lookup(dir, dentry, lookup_flags);
1558         err = PTR_ERR(res);
1559         if (IS_ERR(res))
1560                 goto out;
1561 
1562         return finish_no_open(file, res);
1563 }
1564 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1565 
1566 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1567 {
1568         struct inode *inode;
1569         int ret = 0;
1570 
1571         if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1572                 goto no_open;
1573         if (d_mountpoint(dentry))
1574                 goto no_open;
1575         if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1576                 goto no_open;
1577 
1578         inode = d_inode(dentry);
1579 
1580         /* We can't create new files in nfs_open_revalidate(), so we
1581          * optimize away revalidation of negative dentries.
1582          */
1583         if (inode == NULL) {
1584                 struct dentry *parent;
1585                 struct inode *dir;
1586 
1587                 if (flags & LOOKUP_RCU) {
1588                         parent = ACCESS_ONCE(dentry->d_parent);
1589                         dir = d_inode_rcu(parent);
1590                         if (!dir)
1591                                 return -ECHILD;
1592                 } else {
1593                         parent = dget_parent(dentry);
1594                         dir = d_inode(parent);
1595                 }
1596                 if (!nfs_neg_need_reval(dir, dentry, flags))
1597                         ret = 1;
1598                 else if (flags & LOOKUP_RCU)
1599                         ret = -ECHILD;
1600                 if (!(flags & LOOKUP_RCU))
1601                         dput(parent);
1602                 else if (parent != ACCESS_ONCE(dentry->d_parent))
1603                         return -ECHILD;
1604                 goto out;
1605         }
1606 
1607         /* NFS only supports OPEN on regular files */
1608         if (!S_ISREG(inode->i_mode))
1609                 goto no_open;
1610         /* We cannot do exclusive creation on a positive dentry */
1611         if (flags & LOOKUP_EXCL)
1612                 goto no_open;
1613 
1614         /* Let f_op->open() actually open (and revalidate) the file */
1615         ret = 1;
1616 
1617 out:
1618         return ret;
1619 
1620 no_open:
1621         return nfs_lookup_revalidate(dentry, flags);
1622 }
1623 
1624 #endif /* CONFIG_NFSV4 */
1625 
1626 /*
1627  * Code common to create, mkdir, and mknod.
1628  */
1629 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1630                                 struct nfs_fattr *fattr,
1631                                 struct nfs4_label *label)
1632 {
1633         struct dentry *parent = dget_parent(dentry);
1634         struct inode *dir = d_inode(parent);
1635         struct inode *inode;
1636         int error = -EACCES;
1637 
1638         d_drop(dentry);
1639 
1640         /* We may have been initialized further down */
1641         if (d_really_is_positive(dentry))
1642                 goto out;
1643         if (fhandle->size == 0) {
1644                 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1645                 if (error)
1646                         goto out_error;
1647         }
1648         nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1649         if (!(fattr->valid & NFS_ATTR_FATTR)) {
1650                 struct nfs_server *server = NFS_SB(dentry->d_sb);
1651                 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1652                 if (error < 0)
1653                         goto out_error;
1654         }
1655         inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1656         error = PTR_ERR(inode);
1657         if (IS_ERR(inode))
1658                 goto out_error;
1659         d_add(dentry, inode);
1660 out:
1661         dput(parent);
1662         return 0;
1663 out_error:
1664         nfs_mark_for_revalidate(dir);
1665         dput(parent);
1666         return error;
1667 }
1668 EXPORT_SYMBOL_GPL(nfs_instantiate);
1669 
1670 /*
1671  * Following a failed create operation, we drop the dentry rather
1672  * than retain a negative dentry. This avoids a problem in the event
1673  * that the operation succeeded on the server, but an error in the
1674  * reply path made it appear to have failed.
1675  */
1676 int nfs_create(struct inode *dir, struct dentry *dentry,
1677                 umode_t mode, bool excl)
1678 {
1679         struct iattr attr;
1680         int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1681         int error;
1682 
1683         dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1684                         dir->i_sb->s_id, dir->i_ino, dentry);
1685 
1686         attr.ia_mode = mode;
1687         attr.ia_valid = ATTR_MODE;
1688 
1689         trace_nfs_create_enter(dir, dentry, open_flags);
1690         error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1691         trace_nfs_create_exit(dir, dentry, open_flags, error);
1692         if (error != 0)
1693                 goto out_err;
1694         return 0;
1695 out_err:
1696         d_drop(dentry);
1697         return error;
1698 }
1699 EXPORT_SYMBOL_GPL(nfs_create);
1700 
1701 /*
1702  * See comments for nfs_proc_create regarding failed operations.
1703  */
1704 int
1705 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1706 {
1707         struct iattr attr;
1708         int status;
1709 
1710         dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1711                         dir->i_sb->s_id, dir->i_ino, dentry);
1712 
1713         attr.ia_mode = mode;
1714         attr.ia_valid = ATTR_MODE;
1715 
1716         trace_nfs_mknod_enter(dir, dentry);
1717         status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1718         trace_nfs_mknod_exit(dir, dentry, status);
1719         if (status != 0)
1720                 goto out_err;
1721         return 0;
1722 out_err:
1723         d_drop(dentry);
1724         return status;
1725 }
1726 EXPORT_SYMBOL_GPL(nfs_mknod);
1727 
1728 /*
1729  * See comments for nfs_proc_create regarding failed operations.
1730  */
1731 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1732 {
1733         struct iattr attr;
1734         int error;
1735 
1736         dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1737                         dir->i_sb->s_id, dir->i_ino, dentry);
1738 
1739         attr.ia_valid = ATTR_MODE;
1740         attr.ia_mode = mode | S_IFDIR;
1741 
1742         trace_nfs_mkdir_enter(dir, dentry);
1743         error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1744         trace_nfs_mkdir_exit(dir, dentry, error);
1745         if (error != 0)
1746                 goto out_err;
1747         return 0;
1748 out_err:
1749         d_drop(dentry);
1750         return error;
1751 }
1752 EXPORT_SYMBOL_GPL(nfs_mkdir);
1753 
1754 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1755 {
1756         if (simple_positive(dentry))
1757                 d_delete(dentry);
1758 }
1759 
1760 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1761 {
1762         int error;
1763 
1764         dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1765                         dir->i_sb->s_id, dir->i_ino, dentry);
1766 
1767         trace_nfs_rmdir_enter(dir, dentry);
1768         if (d_really_is_positive(dentry)) {
1769                 nfs_wait_on_sillyrename(dentry);
1770                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1771                 /* Ensure the VFS deletes this inode */
1772                 switch (error) {
1773                 case 0:
1774                         clear_nlink(d_inode(dentry));
1775                         break;
1776                 case -ENOENT:
1777                         nfs_dentry_handle_enoent(dentry);
1778                 }
1779         } else
1780                 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1781         trace_nfs_rmdir_exit(dir, dentry, error);
1782 
1783         return error;
1784 }
1785 EXPORT_SYMBOL_GPL(nfs_rmdir);
1786 
1787 /*
1788  * Remove a file after making sure there are no pending writes,
1789  * and after checking that the file has only one user. 
1790  *
1791  * We invalidate the attribute cache and free the inode prior to the operation
1792  * to avoid possible races if the server reuses the inode.
1793  */
1794 static int nfs_safe_remove(struct dentry *dentry)
1795 {
1796         struct inode *dir = d_inode(dentry->d_parent);
1797         struct inode *inode = d_inode(dentry);
1798         int error = -EBUSY;
1799                 
1800         dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1801 
1802         /* If the dentry was sillyrenamed, we simply call d_delete() */
1803         if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1804                 error = 0;
1805                 goto out;
1806         }
1807 
1808         trace_nfs_remove_enter(dir, dentry);
1809         if (inode != NULL) {
1810                 NFS_PROTO(inode)->return_delegation(inode);
1811                 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1812                 if (error == 0)
1813                         nfs_drop_nlink(inode);
1814         } else
1815                 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1816         if (error == -ENOENT)
1817                 nfs_dentry_handle_enoent(dentry);
1818         trace_nfs_remove_exit(dir, dentry, error);
1819 out:
1820         return error;
1821 }
1822 
1823 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1824  *  belongs to an active ".nfs..." file and we return -EBUSY.
1825  *
1826  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1827  */
1828 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1829 {
1830         int error;
1831         int need_rehash = 0;
1832 
1833         dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1834                 dir->i_ino, dentry);
1835 
1836         trace_nfs_unlink_enter(dir, dentry);
1837         spin_lock(&dentry->d_lock);
1838         if (d_count(dentry) > 1) {
1839                 spin_unlock(&dentry->d_lock);
1840                 /* Start asynchronous writeout of the inode */
1841                 write_inode_now(d_inode(dentry), 0);
1842                 error = nfs_sillyrename(dir, dentry);
1843                 goto out;
1844         }
1845         if (!d_unhashed(dentry)) {
1846                 __d_drop(dentry);
1847                 need_rehash = 1;
1848         }
1849         spin_unlock(&dentry->d_lock);
1850         error = nfs_safe_remove(dentry);
1851         if (!error || error == -ENOENT) {
1852                 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1853         } else if (need_rehash)
1854                 d_rehash(dentry);
1855 out:
1856         trace_nfs_unlink_exit(dir, dentry, error);
1857         return error;
1858 }
1859 EXPORT_SYMBOL_GPL(nfs_unlink);
1860 
1861 /*
1862  * To create a symbolic link, most file systems instantiate a new inode,
1863  * add a page to it containing the path, then write it out to the disk
1864  * using prepare_write/commit_write.
1865  *
1866  * Unfortunately the NFS client can't create the in-core inode first
1867  * because it needs a file handle to create an in-core inode (see
1868  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1869  * symlink request has completed on the server.
1870  *
1871  * So instead we allocate a raw page, copy the symname into it, then do
1872  * the SYMLINK request with the page as the buffer.  If it succeeds, we
1873  * now have a new file handle and can instantiate an in-core NFS inode
1874  * and move the raw page into its mapping.
1875  */
1876 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1877 {
1878         struct page *page;
1879         char *kaddr;
1880         struct iattr attr;
1881         unsigned int pathlen = strlen(symname);
1882         int error;
1883 
1884         dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1885                 dir->i_ino, dentry, symname);
1886 
1887         if (pathlen > PAGE_SIZE)
1888                 return -ENAMETOOLONG;
1889 
1890         attr.ia_mode = S_IFLNK | S_IRWXUGO;
1891         attr.ia_valid = ATTR_MODE;
1892 
1893         page = alloc_page(GFP_USER);
1894         if (!page)
1895                 return -ENOMEM;
1896 
1897         kaddr = page_address(page);
1898         memcpy(kaddr, symname, pathlen);
1899         if (pathlen < PAGE_SIZE)
1900                 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1901 
1902         trace_nfs_symlink_enter(dir, dentry);
1903         error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1904         trace_nfs_symlink_exit(dir, dentry, error);
1905         if (error != 0) {
1906                 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1907                         dir->i_sb->s_id, dir->i_ino,
1908                         dentry, symname, error);
1909                 d_drop(dentry);
1910                 __free_page(page);
1911                 return error;
1912         }
1913 
1914         /*
1915          * No big deal if we can't add this page to the page cache here.
1916          * READLINK will get the missing page from the server if needed.
1917          */
1918         if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1919                                                         GFP_KERNEL)) {
1920                 SetPageUptodate(page);
1921                 unlock_page(page);
1922                 /*
1923                  * add_to_page_cache_lru() grabs an extra page refcount.
1924                  * Drop it here to avoid leaking this page later.
1925                  */
1926                 put_page(page);
1927         } else
1928                 __free_page(page);
1929 
1930         return 0;
1931 }
1932 EXPORT_SYMBOL_GPL(nfs_symlink);
1933 
1934 int
1935 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1936 {
1937         struct inode *inode = d_inode(old_dentry);
1938         int error;
1939 
1940         dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1941                 old_dentry, dentry);
1942 
1943         trace_nfs_link_enter(inode, dir, dentry);
1944         NFS_PROTO(inode)->return_delegation(inode);
1945 
1946         d_drop(dentry);
1947         error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1948         if (error == 0) {
1949                 ihold(inode);
1950                 d_add(dentry, inode);
1951         }
1952         trace_nfs_link_exit(inode, dir, dentry, error);
1953         return error;
1954 }
1955 EXPORT_SYMBOL_GPL(nfs_link);
1956 
1957 /*
1958  * RENAME
1959  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1960  * different file handle for the same inode after a rename (e.g. when
1961  * moving to a different directory). A fail-safe method to do so would
1962  * be to look up old_dir/old_name, create a link to new_dir/new_name and
1963  * rename the old file using the sillyrename stuff. This way, the original
1964  * file in old_dir will go away when the last process iput()s the inode.
1965  *
1966  * FIXED.
1967  * 
1968  * It actually works quite well. One needs to have the possibility for
1969  * at least one ".nfs..." file in each directory the file ever gets
1970  * moved or linked to which happens automagically with the new
1971  * implementation that only depends on the dcache stuff instead of
1972  * using the inode layer
1973  *
1974  * Unfortunately, things are a little more complicated than indicated
1975  * above. For a cross-directory move, we want to make sure we can get
1976  * rid of the old inode after the operation.  This means there must be
1977  * no pending writes (if it's a file), and the use count must be 1.
1978  * If these conditions are met, we can drop the dentries before doing
1979  * the rename.
1980  */
1981 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1982                       struct inode *new_dir, struct dentry *new_dentry)
1983 {
1984         struct inode *old_inode = d_inode(old_dentry);
1985         struct inode *new_inode = d_inode(new_dentry);
1986         struct dentry *dentry = NULL, *rehash = NULL;
1987         struct rpc_task *task;
1988         int error = -EBUSY;
1989 
1990         dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1991                  old_dentry, new_dentry,
1992                  d_count(new_dentry));
1993 
1994         trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1995         /*
1996          * For non-directories, check whether the target is busy and if so,
1997          * make a copy of the dentry and then do a silly-rename. If the
1998          * silly-rename succeeds, the copied dentry is hashed and becomes
1999          * the new target.
2000          */
2001         if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2002                 /*
2003                  * To prevent any new references to the target during the
2004                  * rename, we unhash the dentry in advance.
2005                  */
2006                 if (!d_unhashed(new_dentry)) {
2007                         d_drop(new_dentry);
2008                         rehash = new_dentry;
2009                 }
2010 
2011                 if (d_count(new_dentry) > 2) {
2012                         int err;
2013 
2014                         /* copy the target dentry's name */
2015                         dentry = d_alloc(new_dentry->d_parent,
2016                                          &new_dentry->d_name);
2017                         if (!dentry)
2018                                 goto out;
2019 
2020                         /* silly-rename the existing target ... */
2021                         err = nfs_sillyrename(new_dir, new_dentry);
2022                         if (err)
2023                                 goto out;
2024 
2025                         new_dentry = dentry;
2026                         rehash = NULL;
2027                         new_inode = NULL;
2028                 }
2029         }
2030 
2031         NFS_PROTO(old_inode)->return_delegation(old_inode);
2032         if (new_inode != NULL)
2033                 NFS_PROTO(new_inode)->return_delegation(new_inode);
2034 
2035         task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2036         if (IS_ERR(task)) {
2037                 error = PTR_ERR(task);
2038                 goto out;
2039         }
2040 
2041         error = rpc_wait_for_completion_task(task);
2042         if (error == 0)
2043                 error = task->tk_status;
2044         rpc_put_task(task);
2045         nfs_mark_for_revalidate(old_inode);
2046 out:
2047         if (rehash)
2048                 d_rehash(rehash);
2049         trace_nfs_rename_exit(old_dir, old_dentry,
2050                         new_dir, new_dentry, error);
2051         if (!error) {
2052                 if (new_inode != NULL)
2053                         nfs_drop_nlink(new_inode);
2054                 d_move(old_dentry, new_dentry);
2055                 nfs_set_verifier(new_dentry,
2056                                         nfs_save_change_attribute(new_dir));
2057         } else if (error == -ENOENT)
2058                 nfs_dentry_handle_enoent(old_dentry);
2059 
2060         /* new dentry created? */
2061         if (dentry)
2062                 dput(dentry);
2063         return error;
2064 }
2065 EXPORT_SYMBOL_GPL(nfs_rename);
2066 
2067 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2068 static LIST_HEAD(nfs_access_lru_list);
2069 static atomic_long_t nfs_access_nr_entries;
2070 
2071 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2072 module_param(nfs_access_max_cachesize, ulong, 0644);
2073 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2074 
2075 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2076 {
2077         put_rpccred(entry->cred);
2078         kfree_rcu(entry, rcu_head);
2079         smp_mb__before_atomic();
2080         atomic_long_dec(&nfs_access_nr_entries);
2081         smp_mb__after_atomic();
2082 }
2083 
2084 static void nfs_access_free_list(struct list_head *head)
2085 {
2086         struct nfs_access_entry *cache;
2087 
2088         while (!list_empty(head)) {
2089                 cache = list_entry(head->next, struct nfs_access_entry, lru);
2090                 list_del(&cache->lru);
2091                 nfs_access_free_entry(cache);
2092         }
2093 }
2094 
2095 static unsigned long
2096 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2097 {
2098         LIST_HEAD(head);
2099         struct nfs_inode *nfsi, *next;
2100         struct nfs_access_entry *cache;
2101         long freed = 0;
2102 
2103         spin_lock(&nfs_access_lru_lock);
2104         list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2105                 struct inode *inode;
2106 
2107                 if (nr_to_scan-- == 0)
2108                         break;
2109                 inode = &nfsi->vfs_inode;
2110                 spin_lock(&inode->i_lock);
2111                 if (list_empty(&nfsi->access_cache_entry_lru))
2112                         goto remove_lru_entry;
2113                 cache = list_entry(nfsi->access_cache_entry_lru.next,
2114                                 struct nfs_access_entry, lru);
2115                 list_move(&cache->lru, &head);
2116                 rb_erase(&cache->rb_node, &nfsi->access_cache);
2117                 freed++;
2118                 if (!list_empty(&nfsi->access_cache_entry_lru))
2119                         list_move_tail(&nfsi->access_cache_inode_lru,
2120                                         &nfs_access_lru_list);
2121                 else {
2122 remove_lru_entry:
2123                         list_del_init(&nfsi->access_cache_inode_lru);
2124                         smp_mb__before_atomic();
2125                         clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2126                         smp_mb__after_atomic();
2127                 }
2128                 spin_unlock(&inode->i_lock);
2129         }
2130         spin_unlock(&nfs_access_lru_lock);
2131         nfs_access_free_list(&head);
2132         return freed;
2133 }
2134 
2135 unsigned long
2136 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2137 {
2138         int nr_to_scan = sc->nr_to_scan;
2139         gfp_t gfp_mask = sc->gfp_mask;
2140 
2141         if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2142                 return SHRINK_STOP;
2143         return nfs_do_access_cache_scan(nr_to_scan);
2144 }
2145 
2146 
2147 unsigned long
2148 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2149 {
2150         return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2151 }
2152 
2153 static void
2154 nfs_access_cache_enforce_limit(void)
2155 {
2156         long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2157         unsigned long diff;
2158         unsigned int nr_to_scan;
2159 
2160         if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2161                 return;
2162         nr_to_scan = 100;
2163         diff = nr_entries - nfs_access_max_cachesize;
2164         if (diff < nr_to_scan)
2165                 nr_to_scan = diff;
2166         nfs_do_access_cache_scan(nr_to_scan);
2167 }
2168 
2169 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2170 {
2171         struct rb_root *root_node = &nfsi->access_cache;
2172         struct rb_node *n;
2173         struct nfs_access_entry *entry;
2174 
2175         /* Unhook entries from the cache */
2176         while ((n = rb_first(root_node)) != NULL) {
2177                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2178                 rb_erase(n, root_node);
2179                 list_move(&entry->lru, head);
2180         }
2181         nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2182 }
2183 
2184 void nfs_access_zap_cache(struct inode *inode)
2185 {
2186         LIST_HEAD(head);
2187 
2188         if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2189                 return;
2190         /* Remove from global LRU init */
2191         spin_lock(&nfs_access_lru_lock);
2192         if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2193                 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2194 
2195         spin_lock(&inode->i_lock);
2196         __nfs_access_zap_cache(NFS_I(inode), &head);
2197         spin_unlock(&inode->i_lock);
2198         spin_unlock(&nfs_access_lru_lock);
2199         nfs_access_free_list(&head);
2200 }
2201 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2202 
2203 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2204 {
2205         struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2206         struct nfs_access_entry *entry;
2207 
2208         while (n != NULL) {
2209                 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2210 
2211                 if (cred < entry->cred)
2212                         n = n->rb_left;
2213                 else if (cred > entry->cred)
2214                         n = n->rb_right;
2215                 else
2216                         return entry;
2217         }
2218         return NULL;
2219 }
2220 
2221 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2222 {
2223         struct nfs_inode *nfsi = NFS_I(inode);
2224         struct nfs_access_entry *cache;
2225         int err = -ENOENT;
2226 
2227         spin_lock(&inode->i_lock);
2228         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2229                 goto out_zap;
2230         cache = nfs_access_search_rbtree(inode, cred);
2231         if (cache == NULL)
2232                 goto out;
2233         if (!nfs_have_delegated_attributes(inode) &&
2234             !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2235                 goto out_stale;
2236         res->jiffies = cache->jiffies;
2237         res->cred = cache->cred;
2238         res->mask = cache->mask;
2239         list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2240         err = 0;
2241 out:
2242         spin_unlock(&inode->i_lock);
2243         return err;
2244 out_stale:
2245         rb_erase(&cache->rb_node, &nfsi->access_cache);
2246         list_del(&cache->lru);
2247         spin_unlock(&inode->i_lock);
2248         nfs_access_free_entry(cache);
2249         return -ENOENT;
2250 out_zap:
2251         spin_unlock(&inode->i_lock);
2252         nfs_access_zap_cache(inode);
2253         return -ENOENT;
2254 }
2255 
2256 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2257 {
2258         /* Only check the most recently returned cache entry,
2259          * but do it without locking.
2260          */
2261         struct nfs_inode *nfsi = NFS_I(inode);
2262         struct nfs_access_entry *cache;
2263         int err = -ECHILD;
2264         struct list_head *lh;
2265 
2266         rcu_read_lock();
2267         if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2268                 goto out;
2269         lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2270         cache = list_entry(lh, struct nfs_access_entry, lru);
2271         if (lh == &nfsi->access_cache_entry_lru ||
2272             cred != cache->cred)
2273                 cache = NULL;
2274         if (cache == NULL)
2275                 goto out;
2276         if (!nfs_have_delegated_attributes(inode) &&
2277             !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2278                 goto out;
2279         res->jiffies = cache->jiffies;
2280         res->cred = cache->cred;
2281         res->mask = cache->mask;
2282         err = 0;
2283 out:
2284         rcu_read_unlock();
2285         return err;
2286 }
2287 
2288 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2289 {
2290         struct nfs_inode *nfsi = NFS_I(inode);
2291         struct rb_root *root_node = &nfsi->access_cache;
2292         struct rb_node **p = &root_node->rb_node;
2293         struct rb_node *parent = NULL;
2294         struct nfs_access_entry *entry;
2295 
2296         spin_lock(&inode->i_lock);
2297         while (*p != NULL) {
2298                 parent = *p;
2299                 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2300 
2301                 if (set->cred < entry->cred)
2302                         p = &parent->rb_left;
2303                 else if (set->cred > entry->cred)
2304                         p = &parent->rb_right;
2305                 else
2306                         goto found;
2307         }
2308         rb_link_node(&set->rb_node, parent, p);
2309         rb_insert_color(&set->rb_node, root_node);
2310         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2311         spin_unlock(&inode->i_lock);
2312         return;
2313 found:
2314         rb_replace_node(parent, &set->rb_node, root_node);
2315         list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2316         list_del(&entry->lru);
2317         spin_unlock(&inode->i_lock);
2318         nfs_access_free_entry(entry);
2319 }
2320 
2321 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2322 {
2323         struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2324         if (cache == NULL)
2325                 return;
2326         RB_CLEAR_NODE(&cache->rb_node);
2327         cache->jiffies = set->jiffies;
2328         cache->cred = get_rpccred(set->cred);
2329         cache->mask = set->mask;
2330 
2331         /* The above field assignments must be visible
2332          * before this item appears on the lru.  We cannot easily
2333          * use rcu_assign_pointer, so just force the memory barrier.
2334          */
2335         smp_wmb();
2336         nfs_access_add_rbtree(inode, cache);
2337 
2338         /* Update accounting */
2339         smp_mb__before_atomic();
2340         atomic_long_inc(&nfs_access_nr_entries);
2341         smp_mb__after_atomic();
2342 
2343         /* Add inode to global LRU list */
2344         if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2345                 spin_lock(&nfs_access_lru_lock);
2346                 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2347                         list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2348                                         &nfs_access_lru_list);
2349                 spin_unlock(&nfs_access_lru_lock);
2350         }
2351         nfs_access_cache_enforce_limit();
2352 }
2353 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2354 
2355 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2356 {
2357         entry->mask = 0;
2358         if (access_result & NFS4_ACCESS_READ)
2359                 entry->mask |= MAY_READ;
2360         if (access_result &
2361             (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2362                 entry->mask |= MAY_WRITE;
2363         if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2364                 entry->mask |= MAY_EXEC;
2365 }
2366 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2367 
2368 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2369 {
2370         struct nfs_access_entry cache;
2371         int status;
2372 
2373         trace_nfs_access_enter(inode);
2374 
2375         status = nfs_access_get_cached_rcu(inode, cred, &cache);
2376         if (status != 0)
2377                 status = nfs_access_get_cached(inode, cred, &cache);
2378         if (status == 0)
2379                 goto out_cached;
2380 
2381         status = -ECHILD;
2382         if (mask & MAY_NOT_BLOCK)
2383                 goto out;
2384 
2385         /* Be clever: ask server to check for all possible rights */
2386         cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2387         cache.cred = cred;
2388         cache.jiffies = jiffies;
2389         status = NFS_PROTO(inode)->access(inode, &cache);
2390         if (status != 0) {
2391                 if (status == -ESTALE) {
2392                         nfs_zap_caches(inode);
2393                         if (!S_ISDIR(inode->i_mode))
2394                                 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2395                 }
2396                 goto out;
2397         }
2398         nfs_access_add_cache(inode, &cache);
2399 out_cached:
2400         if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2401                 status = -EACCES;
2402 out:
2403         trace_nfs_access_exit(inode, status);
2404         return status;
2405 }
2406 
2407 static int nfs_open_permission_mask(int openflags)
2408 {
2409         int mask = 0;
2410 
2411         if (openflags & __FMODE_EXEC) {
2412                 /* ONLY check exec rights */
2413                 mask = MAY_EXEC;
2414         } else {
2415                 if ((openflags & O_ACCMODE) != O_WRONLY)
2416                         mask |= MAY_READ;
2417                 if ((openflags & O_ACCMODE) != O_RDONLY)
2418                         mask |= MAY_WRITE;
2419         }
2420 
2421         return mask;
2422 }
2423 
2424 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2425 {
2426         return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2427 }
2428 EXPORT_SYMBOL_GPL(nfs_may_open);
2429 
2430 static int nfs_execute_ok(struct inode *inode, int mask)
2431 {
2432         struct nfs_server *server = NFS_SERVER(inode);
2433         int ret;
2434 
2435         if (mask & MAY_NOT_BLOCK)
2436                 ret = nfs_revalidate_inode_rcu(server, inode);
2437         else
2438                 ret = nfs_revalidate_inode(server, inode);
2439         if (ret == 0 && !execute_ok(inode))
2440                 ret = -EACCES;
2441         return ret;
2442 }
2443 
2444 int nfs_permission(struct inode *inode, int mask)
2445 {
2446         struct rpc_cred *cred;
2447         int res = 0;
2448 
2449         nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2450 
2451         if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2452                 goto out;
2453         /* Is this sys_access() ? */
2454         if (mask & (MAY_ACCESS | MAY_CHDIR))
2455                 goto force_lookup;
2456 
2457         switch (inode->i_mode & S_IFMT) {
2458                 case S_IFLNK:
2459                         goto out;
2460                 case S_IFREG:
2461                         if ((mask & MAY_OPEN) &&
2462                            nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2463                                 return 0;
2464                         break;
2465                 case S_IFDIR:
2466                         /*
2467                          * Optimize away all write operations, since the server
2468                          * will check permissions when we perform the op.
2469                          */
2470                         if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2471                                 goto out;
2472         }
2473 
2474 force_lookup:
2475         if (!NFS_PROTO(inode)->access)
2476                 goto out_notsup;
2477 
2478         /* Always try fast lookups first */
2479         rcu_read_lock();
2480         cred = rpc_lookup_cred_nonblock();
2481         if (!IS_ERR(cred))
2482                 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2483         else
2484                 res = PTR_ERR(cred);
2485         rcu_read_unlock();
2486         if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2487                 /* Fast lookup failed, try the slow way */
2488                 cred = rpc_lookup_cred();
2489                 if (!IS_ERR(cred)) {
2490                         res = nfs_do_access(inode, cred, mask);
2491                         put_rpccred(cred);
2492                 } else
2493                         res = PTR_ERR(cred);
2494         }
2495 out:
2496         if (!res && (mask & MAY_EXEC))
2497                 res = nfs_execute_ok(inode, mask);
2498 
2499         dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2500                 inode->i_sb->s_id, inode->i_ino, mask, res);
2501         return res;
2502 out_notsup:
2503         if (mask & MAY_NOT_BLOCK)
2504                 return -ECHILD;
2505 
2506         res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2507         if (res == 0)
2508                 res = generic_permission(inode, mask);
2509         goto out;
2510 }
2511 EXPORT_SYMBOL_GPL(nfs_permission);
2512 
2513 /*
2514  * Local variables:
2515  *  version-control: t
2516  *  kept-new-versions: 5
2517  * End:
2518  */
2519 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp