]> err.no Git - linux-2.6/blob - fs/cifs/file.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / fs / cifs / file.c
1 /*
2  *   fs/cifs/file.c
3  *
4  *   vfs operations that deal with files
5  *
6  *   Copyright (C) International Business Machines  Corp., 2002,2007
7  *   Author(s): Steve French (sfrench@us.ibm.com)
8  *              Jeremy Allison (jra@samba.org)
9  *
10  *   This library is free software; you can redistribute it and/or modify
11  *   it under the terms of the GNU Lesser General Public License as published
12  *   by the Free Software Foundation; either version 2.1 of the License, or
13  *   (at your option) any later version.
14  *
15  *   This library is distributed in the hope that it will be useful,
16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
18  *   the GNU Lesser General Public License for more details.
19  *
20  *   You should have received a copy of the GNU Lesser General Public License
21  *   along with this library; if not, write to the Free Software
22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23  */
24 #include <linux/fs.h>
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <asm/div64.h>
34 #include "cifsfs.h"
35 #include "cifspdu.h"
36 #include "cifsglob.h"
37 #include "cifsproto.h"
38 #include "cifs_unicode.h"
39 #include "cifs_debug.h"
40 #include "cifs_fs_sb.h"
41
42 static inline struct cifsFileInfo *cifs_init_private(
43         struct cifsFileInfo *private_data, struct inode *inode,
44         struct file *file, __u16 netfid)
45 {
46         memset(private_data, 0, sizeof(struct cifsFileInfo));
47         private_data->netfid = netfid;
48         private_data->pid = current->tgid;
49         init_MUTEX(&private_data->fh_sem);
50         mutex_init(&private_data->lock_mutex);
51         INIT_LIST_HEAD(&private_data->llist);
52         private_data->pfile = file; /* needed for writepage */
53         private_data->pInode = inode;
54         private_data->invalidHandle = false;
55         private_data->closePend = false;
56         /* we have to track num writers to the inode, since writepages
57         does not tell us which handle the write is for so there can
58         be a close (overlapping with write) of the filehandle that
59         cifs_writepages chose to use */
60         atomic_set(&private_data->wrtPending, 0);
61
62         return private_data;
63 }
64
65 static inline int cifs_convert_flags(unsigned int flags)
66 {
67         if ((flags & O_ACCMODE) == O_RDONLY)
68                 return GENERIC_READ;
69         else if ((flags & O_ACCMODE) == O_WRONLY)
70                 return GENERIC_WRITE;
71         else if ((flags & O_ACCMODE) == O_RDWR) {
72                 /* GENERIC_ALL is too much permission to request
73                    can cause unnecessary access denied on create */
74                 /* return GENERIC_ALL; */
75                 return (GENERIC_READ | GENERIC_WRITE);
76         }
77
78         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
79                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
80                 FILE_READ_DATA);
81
82
83 }
84
85 static inline int cifs_get_disposition(unsigned int flags)
86 {
87         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
88                 return FILE_CREATE;
89         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
90                 return FILE_OVERWRITE_IF;
91         else if ((flags & O_CREAT) == O_CREAT)
92                 return FILE_OPEN_IF;
93         else if ((flags & O_TRUNC) == O_TRUNC)
94                 return FILE_OVERWRITE;
95         else
96                 return FILE_OPEN;
97 }
98
99 /* all arguments to this function must be checked for validity in caller */
100 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
101         struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
102         struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
103         char *full_path, int xid)
104 {
105         struct timespec temp;
106         int rc;
107
108         /* want handles we can use to read with first
109            in the list so we do not have to walk the
110            list to search for one in prepare_write */
111         if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
112                 list_add_tail(&pCifsFile->flist,
113                               &pCifsInode->openFileList);
114         } else {
115                 list_add(&pCifsFile->flist,
116                          &pCifsInode->openFileList);
117         }
118         write_unlock(&GlobalSMBSeslock);
119         if (pCifsInode->clientCanCacheRead) {
120                 /* we have the inode open somewhere else
121                    no need to discard cache data */
122                 goto client_can_cache;
123         }
124
125         /* BB need same check in cifs_create too? */
126         /* if not oplocked, invalidate inode pages if mtime or file
127            size changed */
128         temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
129         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
130                            (file->f_path.dentry->d_inode->i_size ==
131                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
132                 cFYI(1, ("inode unchanged on server"));
133         } else {
134                 if (file->f_path.dentry->d_inode->i_mapping) {
135                 /* BB no need to lock inode until after invalidate
136                    since namei code should already have it locked? */
137                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
138                         if (rc != 0)
139                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
140                 }
141                 cFYI(1, ("invalidating remote inode since open detected it "
142                          "changed"));
143                 invalidate_remote_inode(file->f_path.dentry->d_inode);
144         }
145
146 client_can_cache:
147         if (pTcon->unix_ext)
148                 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
149                         full_path, inode->i_sb, xid);
150         else
151                 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
152                         full_path, buf, inode->i_sb, xid, NULL);
153
154         if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
155                 pCifsInode->clientCanCacheAll = true;
156                 pCifsInode->clientCanCacheRead = true;
157                 cFYI(1, ("Exclusive Oplock granted on inode %p",
158                          file->f_path.dentry->d_inode));
159         } else if ((*oplock & 0xF) == OPLOCK_READ)
160                 pCifsInode->clientCanCacheRead = true;
161
162         return rc;
163 }
164
165 int cifs_open(struct inode *inode, struct file *file)
166 {
167         int rc = -EACCES;
168         int xid, oplock;
169         struct cifs_sb_info *cifs_sb;
170         struct cifsTconInfo *pTcon;
171         struct cifsFileInfo *pCifsFile;
172         struct cifsInodeInfo *pCifsInode;
173         struct list_head *tmp;
174         char *full_path = NULL;
175         int desiredAccess;
176         int disposition;
177         __u16 netfid;
178         FILE_ALL_INFO *buf = NULL;
179
180         xid = GetXid();
181
182         cifs_sb = CIFS_SB(inode->i_sb);
183         pTcon = cifs_sb->tcon;
184
185         if (file->f_flags & O_CREAT) {
186                 /* search inode for this file and fill in file->private_data */
187                 pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
188                 read_lock(&GlobalSMBSeslock);
189                 list_for_each(tmp, &pCifsInode->openFileList) {
190                         pCifsFile = list_entry(tmp, struct cifsFileInfo,
191                                                flist);
192                         if ((pCifsFile->pfile == NULL) &&
193                             (pCifsFile->pid == current->tgid)) {
194                                 /* mode set in cifs_create */
195
196                                 /* needed for writepage */
197                                 pCifsFile->pfile = file;
198
199                                 file->private_data = pCifsFile;
200                                 break;
201                         }
202                 }
203                 read_unlock(&GlobalSMBSeslock);
204                 if (file->private_data != NULL) {
205                         rc = 0;
206                         FreeXid(xid);
207                         return rc;
208                 } else {
209                         if (file->f_flags & O_EXCL)
210                                 cERROR(1, ("could not find file instance for "
211                                            "new file %p", file));
212                 }
213         }
214
215         full_path = build_path_from_dentry(file->f_path.dentry);
216         if (full_path == NULL) {
217                 FreeXid(xid);
218                 return -ENOMEM;
219         }
220
221         cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
222                  inode, file->f_flags, full_path));
223         desiredAccess = cifs_convert_flags(file->f_flags);
224
225 /*********************************************************************
226  *  open flag mapping table:
227  *
228  *      POSIX Flag            CIFS Disposition
229  *      ----------            ----------------
230  *      O_CREAT               FILE_OPEN_IF
231  *      O_CREAT | O_EXCL      FILE_CREATE
232  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
233  *      O_TRUNC               FILE_OVERWRITE
234  *      none of the above     FILE_OPEN
235  *
236  *      Note that there is not a direct match between disposition
237  *      FILE_SUPERSEDE (ie create whether or not file exists although
238  *      O_CREAT | O_TRUNC is similar but truncates the existing
239  *      file rather than creating a new file as FILE_SUPERSEDE does
240  *      (which uses the attributes / metadata passed in on open call)
241  *?
242  *?  O_SYNC is a reasonable match to CIFS writethrough flag
243  *?  and the read write flags match reasonably.  O_LARGEFILE
244  *?  is irrelevant because largefile support is always used
245  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
246  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
247  *********************************************************************/
248
249         disposition = cifs_get_disposition(file->f_flags);
250
251         if (oplockEnabled)
252                 oplock = REQ_OPLOCK;
253         else
254                 oplock = 0;
255
256         /* BB pass O_SYNC flag through on file attributes .. BB */
257
258         /* Also refresh inode by passing in file_info buf returned by SMBOpen
259            and calling get_inode_info with returned buf (at least helps
260            non-Unix server case) */
261
262         /* BB we can not do this if this is the second open of a file
263            and the first handle has writebehind data, we might be
264            able to simply do a filemap_fdatawrite/filemap_fdatawait first */
265         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
266         if (!buf) {
267                 rc = -ENOMEM;
268                 goto out;
269         }
270
271         if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
272                 rc = CIFSSMBOpen(xid, pTcon, full_path, disposition,
273                          desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
274                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
275                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
276         else
277                 rc = -EIO; /* no NT SMB support fall into legacy open below */
278
279         if (rc == -EIO) {
280                 /* Old server, try legacy style OpenX */
281                 rc = SMBLegacyOpen(xid, pTcon, full_path, disposition,
282                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
283                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
284                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
285         }
286         if (rc) {
287                 cFYI(1, ("cifs_open returned 0x%x", rc));
288                 goto out;
289         }
290         file->private_data =
291                 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
292         if (file->private_data == NULL) {
293                 rc = -ENOMEM;
294                 goto out;
295         }
296         pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
297         write_lock(&GlobalSMBSeslock);
298         list_add(&pCifsFile->tlist, &pTcon->openFileList);
299
300         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
301         if (pCifsInode) {
302                 rc = cifs_open_inode_helper(inode, file, pCifsInode,
303                                             pCifsFile, pTcon,
304                                             &oplock, buf, full_path, xid);
305         } else {
306                 write_unlock(&GlobalSMBSeslock);
307         }
308
309         if (oplock & CIFS_CREATE_ACTION) {
310                 /* time to set mode which we can not set earlier due to
311                    problems creating new read-only files */
312                 if (pTcon->unix_ext) {
313                         struct cifs_unix_set_info_args args = {
314                                 .mode   = inode->i_mode,
315                                 .uid    = NO_CHANGE_64,
316                                 .gid    = NO_CHANGE_64,
317                                 .ctime  = NO_CHANGE_64,
318                                 .atime  = NO_CHANGE_64,
319                                 .mtime  = NO_CHANGE_64,
320                                 .device = 0,
321                         };
322                         CIFSSMBUnixSetInfo(xid, pTcon, full_path, &args,
323                                             cifs_sb->local_nls,
324                                             cifs_sb->mnt_cifs_flags &
325                                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
326                 }
327         }
328
329 out:
330         kfree(buf);
331         kfree(full_path);
332         FreeXid(xid);
333         return rc;
334 }
335
336 /* Try to reacquire byte range locks that were released when session */
337 /* to server was lost */
338 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
339 {
340         int rc = 0;
341
342 /* BB list all locks open on this file and relock */
343
344         return rc;
345 }
346
347 static int cifs_reopen_file(struct file *file, bool can_flush)
348 {
349         int rc = -EACCES;
350         int xid, oplock;
351         struct cifs_sb_info *cifs_sb;
352         struct cifsTconInfo *pTcon;
353         struct cifsFileInfo *pCifsFile;
354         struct cifsInodeInfo *pCifsInode;
355         struct inode *inode;
356         char *full_path = NULL;
357         int desiredAccess;
358         int disposition = FILE_OPEN;
359         __u16 netfid;
360
361         if (file->private_data)
362                 pCifsFile = (struct cifsFileInfo *)file->private_data;
363         else
364                 return -EBADF;
365
366         xid = GetXid();
367         down(&pCifsFile->fh_sem);
368         if (!pCifsFile->invalidHandle) {
369                 up(&pCifsFile->fh_sem);
370                 FreeXid(xid);
371                 return 0;
372         }
373
374         if (file->f_path.dentry == NULL) {
375                 cERROR(1, ("no valid name if dentry freed"));
376                 dump_stack();
377                 rc = -EBADF;
378                 goto reopen_error_exit;
379         }
380
381         inode = file->f_path.dentry->d_inode;
382         if (inode == NULL) {
383                 cERROR(1, ("inode not valid"));
384                 dump_stack();
385                 rc = -EBADF;
386                 goto reopen_error_exit;
387         }
388
389         cifs_sb = CIFS_SB(inode->i_sb);
390         pTcon = cifs_sb->tcon;
391
392 /* can not grab rename sem here because various ops, including
393    those that already have the rename sem can end up causing writepage
394    to get called and if the server was down that means we end up here,
395    and we can never tell if the caller already has the rename_sem */
396         full_path = build_path_from_dentry(file->f_path.dentry);
397         if (full_path == NULL) {
398                 rc = -ENOMEM;
399 reopen_error_exit:
400                 up(&pCifsFile->fh_sem);
401                 FreeXid(xid);
402                 return rc;
403         }
404
405         cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
406                  inode, file->f_flags, full_path));
407         desiredAccess = cifs_convert_flags(file->f_flags);
408
409         if (oplockEnabled)
410                 oplock = REQ_OPLOCK;
411         else
412                 oplock = 0;
413
414         /* Can not refresh inode by passing in file_info buf to be returned
415            by SMBOpen and then calling get_inode_info with returned buf
416            since file might have write behind data that needs to be flushed
417            and server version of file size can be stale. If we knew for sure
418            that inode was not dirty locally we could do this */
419
420         rc = CIFSSMBOpen(xid, pTcon, full_path, disposition, desiredAccess,
421                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
422                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
423                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
424         if (rc) {
425                 up(&pCifsFile->fh_sem);
426                 cFYI(1, ("cifs_open returned 0x%x", rc));
427                 cFYI(1, ("oplock: %d", oplock));
428         } else {
429                 pCifsFile->netfid = netfid;
430                 pCifsFile->invalidHandle = false;
431                 up(&pCifsFile->fh_sem);
432                 pCifsInode = CIFS_I(inode);
433                 if (pCifsInode) {
434                         if (can_flush) {
435                                 rc = filemap_write_and_wait(inode->i_mapping);
436                                 if (rc != 0)
437                                         CIFS_I(inode)->write_behind_rc = rc;
438                         /* temporarily disable caching while we
439                            go to server to get inode info */
440                                 pCifsInode->clientCanCacheAll = false;
441                                 pCifsInode->clientCanCacheRead = false;
442                                 if (pTcon->unix_ext)
443                                         rc = cifs_get_inode_info_unix(&inode,
444                                                 full_path, inode->i_sb, xid);
445                                 else
446                                         rc = cifs_get_inode_info(&inode,
447                                                 full_path, NULL, inode->i_sb,
448                                                 xid, NULL);
449                         } /* else we are writing out data to server already
450                              and could deadlock if we tried to flush data, and
451                              since we do not know if we have data that would
452                              invalidate the current end of file on the server
453                              we can not go to the server to get the new inod
454                              info */
455                         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
456                                 pCifsInode->clientCanCacheAll = true;
457                                 pCifsInode->clientCanCacheRead = true;
458                                 cFYI(1, ("Exclusive Oplock granted on inode %p",
459                                          file->f_path.dentry->d_inode));
460                         } else if ((oplock & 0xF) == OPLOCK_READ) {
461                                 pCifsInode->clientCanCacheRead = true;
462                                 pCifsInode->clientCanCacheAll = false;
463                         } else {
464                                 pCifsInode->clientCanCacheRead = false;
465                                 pCifsInode->clientCanCacheAll = false;
466                         }
467                         cifs_relock_file(pCifsFile);
468                 }
469         }
470
471         kfree(full_path);
472         FreeXid(xid);
473         return rc;
474 }
475
476 int cifs_close(struct inode *inode, struct file *file)
477 {
478         int rc = 0;
479         int xid, timeout;
480         struct cifs_sb_info *cifs_sb;
481         struct cifsTconInfo *pTcon;
482         struct cifsFileInfo *pSMBFile =
483                 (struct cifsFileInfo *)file->private_data;
484
485         xid = GetXid();
486
487         cifs_sb = CIFS_SB(inode->i_sb);
488         pTcon = cifs_sb->tcon;
489         if (pSMBFile) {
490                 struct cifsLockInfo *li, *tmp;
491
492                 pSMBFile->closePend = true;
493                 if (pTcon) {
494                         /* no sense reconnecting to close a file that is
495                            already closed */
496                         if (pTcon->tidStatus != CifsNeedReconnect) {
497                                 timeout = 2;
498                                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
499                                         && (timeout <= 2048)) {
500                                         /* Give write a better chance to get to
501                                         server ahead of the close.  We do not
502                                         want to add a wait_q here as it would
503                                         increase the memory utilization as
504                                         the struct would be in each open file,
505                                         but this should give enough time to
506                                         clear the socket */
507                                         cFYI(DBG2,
508                                                 ("close delay, write pending"));
509                                         msleep(timeout);
510                                         timeout *= 4;
511                                 }
512                                 if (atomic_read(&pSMBFile->wrtPending))
513                                         cERROR(1,
514                                                 ("close with pending writes"));
515                                 rc = CIFSSMBClose(xid, pTcon,
516                                                   pSMBFile->netfid);
517                         }
518                 }
519
520                 /* Delete any outstanding lock records.
521                    We'll lose them when the file is closed anyway. */
522                 mutex_lock(&pSMBFile->lock_mutex);
523                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
524                         list_del(&li->llist);
525                         kfree(li);
526                 }
527                 mutex_unlock(&pSMBFile->lock_mutex);
528
529                 write_lock(&GlobalSMBSeslock);
530                 list_del(&pSMBFile->flist);
531                 list_del(&pSMBFile->tlist);
532                 write_unlock(&GlobalSMBSeslock);
533                 timeout = 10;
534                 /* We waited above to give the SMBWrite a chance to issue
535                    on the wire (so we do not get SMBWrite returning EBADF
536                    if writepages is racing with close.  Note that writepages
537                    does not specify a file handle, so it is possible for a file
538                    to be opened twice, and the application close the "wrong"
539                    file handle - in these cases we delay long enough to allow
540                    the SMBWrite to get on the wire before the SMB Close.
541                    We allow total wait here over 45 seconds, more than
542                    oplock break time, and more than enough to allow any write
543                    to complete on the server, or to time out on the client */
544                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
545                                 && (timeout <= 50000)) {
546                         cERROR(1, ("writes pending, delay free of handle"));
547                         msleep(timeout);
548                         timeout *= 8;
549                 }
550                 kfree(file->private_data);
551                 file->private_data = NULL;
552         } else
553                 rc = -EBADF;
554
555         read_lock(&GlobalSMBSeslock);
556         if (list_empty(&(CIFS_I(inode)->openFileList))) {
557                 cFYI(1, ("closing last open instance for inode %p", inode));
558                 /* if the file is not open we do not know if we can cache info
559                    on this inode, much less write behind and read ahead */
560                 CIFS_I(inode)->clientCanCacheRead = false;
561                 CIFS_I(inode)->clientCanCacheAll  = false;
562         }
563         read_unlock(&GlobalSMBSeslock);
564         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
565                 rc = CIFS_I(inode)->write_behind_rc;
566         FreeXid(xid);
567         return rc;
568 }
569
570 int cifs_closedir(struct inode *inode, struct file *file)
571 {
572         int rc = 0;
573         int xid;
574         struct cifsFileInfo *pCFileStruct =
575             (struct cifsFileInfo *)file->private_data;
576         char *ptmp;
577
578         cFYI(1, ("Closedir inode = 0x%p", inode));
579
580         xid = GetXid();
581
582         if (pCFileStruct) {
583                 struct cifsTconInfo *pTcon;
584                 struct cifs_sb_info *cifs_sb =
585                         CIFS_SB(file->f_path.dentry->d_sb);
586
587                 pTcon = cifs_sb->tcon;
588
589                 cFYI(1, ("Freeing private data in close dir"));
590                 if (!pCFileStruct->srch_inf.endOfSearch &&
591                     !pCFileStruct->invalidHandle) {
592                         pCFileStruct->invalidHandle = true;
593                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
594                         cFYI(1, ("Closing uncompleted readdir with rc %d",
595                                  rc));
596                         /* not much we can do if it fails anyway, ignore rc */
597                         rc = 0;
598                 }
599                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
600                 if (ptmp) {
601                         cFYI(1, ("closedir free smb buf in srch struct"));
602                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
603                         if (pCFileStruct->srch_inf.smallBuf)
604                                 cifs_small_buf_release(ptmp);
605                         else
606                                 cifs_buf_release(ptmp);
607                 }
608                 kfree(file->private_data);
609                 file->private_data = NULL;
610         }
611         /* BB can we lock the filestruct while this is going on? */
612         FreeXid(xid);
613         return rc;
614 }
615
616 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
617                                 __u64 offset, __u8 lockType)
618 {
619         struct cifsLockInfo *li =
620                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
621         if (li == NULL)
622                 return -ENOMEM;
623         li->offset = offset;
624         li->length = len;
625         li->type = lockType;
626         mutex_lock(&fid->lock_mutex);
627         list_add(&li->llist, &fid->llist);
628         mutex_unlock(&fid->lock_mutex);
629         return 0;
630 }
631
632 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
633 {
634         int rc, xid;
635         __u32 numLock = 0;
636         __u32 numUnlock = 0;
637         __u64 length;
638         bool wait_flag = false;
639         struct cifs_sb_info *cifs_sb;
640         struct cifsTconInfo *pTcon;
641         __u16 netfid;
642         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
643         bool posix_locking;
644
645         length = 1 + pfLock->fl_end - pfLock->fl_start;
646         rc = -EACCES;
647         xid = GetXid();
648
649         cFYI(1, ("Lock parm: 0x%x flockflags: "
650                  "0x%x flocktype: 0x%x start: %lld end: %lld",
651                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
652                 pfLock->fl_end));
653
654         if (pfLock->fl_flags & FL_POSIX)
655                 cFYI(1, ("Posix"));
656         if (pfLock->fl_flags & FL_FLOCK)
657                 cFYI(1, ("Flock"));
658         if (pfLock->fl_flags & FL_SLEEP) {
659                 cFYI(1, ("Blocking lock"));
660                 wait_flag = true;
661         }
662         if (pfLock->fl_flags & FL_ACCESS)
663                 cFYI(1, ("Process suspended by mandatory locking - "
664                          "not implemented yet"));
665         if (pfLock->fl_flags & FL_LEASE)
666                 cFYI(1, ("Lease on file - not implemented yet"));
667         if (pfLock->fl_flags &
668             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
669                 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
670
671         if (pfLock->fl_type == F_WRLCK) {
672                 cFYI(1, ("F_WRLCK "));
673                 numLock = 1;
674         } else if (pfLock->fl_type == F_UNLCK) {
675                 cFYI(1, ("F_UNLCK"));
676                 numUnlock = 1;
677                 /* Check if unlock includes more than
678                 one lock range */
679         } else if (pfLock->fl_type == F_RDLCK) {
680                 cFYI(1, ("F_RDLCK"));
681                 lockType |= LOCKING_ANDX_SHARED_LOCK;
682                 numLock = 1;
683         } else if (pfLock->fl_type == F_EXLCK) {
684                 cFYI(1, ("F_EXLCK"));
685                 numLock = 1;
686         } else if (pfLock->fl_type == F_SHLCK) {
687                 cFYI(1, ("F_SHLCK"));
688                 lockType |= LOCKING_ANDX_SHARED_LOCK;
689                 numLock = 1;
690         } else
691                 cFYI(1, ("Unknown type of lock"));
692
693         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
694         pTcon = cifs_sb->tcon;
695
696         if (file->private_data == NULL) {
697                 FreeXid(xid);
698                 return -EBADF;
699         }
700         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
701
702         posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
703                         (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
704
705         /* BB add code here to normalize offset and length to
706         account for negative length which we can not accept over the
707         wire */
708         if (IS_GETLK(cmd)) {
709                 if (posix_locking) {
710                         int posix_lock_type;
711                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
712                                 posix_lock_type = CIFS_RDLCK;
713                         else
714                                 posix_lock_type = CIFS_WRLCK;
715                         rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
716                                         length, pfLock,
717                                         posix_lock_type, wait_flag);
718                         FreeXid(xid);
719                         return rc;
720                 }
721
722                 /* BB we could chain these into one lock request BB */
723                 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
724                                  0, 1, lockType, 0 /* wait flag */ );
725                 if (rc == 0) {
726                         rc = CIFSSMBLock(xid, pTcon, netfid, length,
727                                          pfLock->fl_start, 1 /* numUnlock */ ,
728                                          0 /* numLock */ , lockType,
729                                          0 /* wait flag */ );
730                         pfLock->fl_type = F_UNLCK;
731                         if (rc != 0)
732                                 cERROR(1, ("Error unlocking previously locked "
733                                            "range %d during test of lock", rc));
734                         rc = 0;
735
736                 } else {
737                         /* if rc == ERR_SHARING_VIOLATION ? */
738                         rc = 0; /* do not change lock type to unlock
739                                    since range in use */
740                 }
741
742                 FreeXid(xid);
743                 return rc;
744         }
745
746         if (!numLock && !numUnlock) {
747                 /* if no lock or unlock then nothing
748                 to do since we do not know what it is */
749                 FreeXid(xid);
750                 return -EOPNOTSUPP;
751         }
752
753         if (posix_locking) {
754                 int posix_lock_type;
755                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
756                         posix_lock_type = CIFS_RDLCK;
757                 else
758                         posix_lock_type = CIFS_WRLCK;
759
760                 if (numUnlock == 1)
761                         posix_lock_type = CIFS_UNLCK;
762
763                 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
764                                       length, pfLock,
765                                       posix_lock_type, wait_flag);
766         } else {
767                 struct cifsFileInfo *fid =
768                         (struct cifsFileInfo *)file->private_data;
769
770                 if (numLock) {
771                         rc = CIFSSMBLock(xid, pTcon, netfid, length,
772                                         pfLock->fl_start,
773                                         0, numLock, lockType, wait_flag);
774
775                         if (rc == 0) {
776                                 /* For Windows locks we must store them. */
777                                 rc = store_file_lock(fid, length,
778                                                 pfLock->fl_start, lockType);
779                         }
780                 } else if (numUnlock) {
781                         /* For each stored lock that this unlock overlaps
782                            completely, unlock it. */
783                         int stored_rc = 0;
784                         struct cifsLockInfo *li, *tmp;
785
786                         rc = 0;
787                         mutex_lock(&fid->lock_mutex);
788                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
789                                 if (pfLock->fl_start <= li->offset &&
790                                                 (pfLock->fl_start + length) >=
791                                                 (li->offset + li->length)) {
792                                         stored_rc = CIFSSMBLock(xid, pTcon,
793                                                         netfid,
794                                                         li->length, li->offset,
795                                                         1, 0, li->type, false);
796                                         if (stored_rc)
797                                                 rc = stored_rc;
798
799                                         list_del(&li->llist);
800                                         kfree(li);
801                                 }
802                         }
803                         mutex_unlock(&fid->lock_mutex);
804                 }
805         }
806
807         if (pfLock->fl_flags & FL_POSIX)
808                 posix_lock_file_wait(file, pfLock);
809         FreeXid(xid);
810         return rc;
811 }
812
813 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
814         size_t write_size, loff_t *poffset)
815 {
816         int rc = 0;
817         unsigned int bytes_written = 0;
818         unsigned int total_written;
819         struct cifs_sb_info *cifs_sb;
820         struct cifsTconInfo *pTcon;
821         int xid, long_op;
822         struct cifsFileInfo *open_file;
823
824         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
825
826         pTcon = cifs_sb->tcon;
827
828         /* cFYI(1,
829            (" write %d bytes to offset %lld of %s", write_size,
830            *poffset, file->f_path.dentry->d_name.name)); */
831
832         if (file->private_data == NULL)
833                 return -EBADF;
834         open_file = (struct cifsFileInfo *) file->private_data;
835
836         xid = GetXid();
837
838         if (*poffset > file->f_path.dentry->d_inode->i_size)
839                 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
840         else
841                 long_op = CIFS_LONG_OP;
842
843         for (total_written = 0; write_size > total_written;
844              total_written += bytes_written) {
845                 rc = -EAGAIN;
846                 while (rc == -EAGAIN) {
847                         if (file->private_data == NULL) {
848                                 /* file has been closed on us */
849                                 FreeXid(xid);
850                         /* if we have gotten here we have written some data
851                            and blocked, and the file has been freed on us while
852                            we blocked so return what we managed to write */
853                                 return total_written;
854                         }
855                         if (open_file->closePend) {
856                                 FreeXid(xid);
857                                 if (total_written)
858                                         return total_written;
859                                 else
860                                         return -EBADF;
861                         }
862                         if (open_file->invalidHandle) {
863                                 /* we could deadlock if we called
864                                    filemap_fdatawait from here so tell
865                                    reopen_file not to flush data to server
866                                    now */
867                                 rc = cifs_reopen_file(file, false);
868                                 if (rc != 0)
869                                         break;
870                         }
871
872                         rc = CIFSSMBWrite(xid, pTcon,
873                                 open_file->netfid,
874                                 min_t(const int, cifs_sb->wsize,
875                                       write_size - total_written),
876                                 *poffset, &bytes_written,
877                                 NULL, write_data + total_written, long_op);
878                 }
879                 if (rc || (bytes_written == 0)) {
880                         if (total_written)
881                                 break;
882                         else {
883                                 FreeXid(xid);
884                                 return rc;
885                         }
886                 } else
887                         *poffset += bytes_written;
888                 long_op = CIFS_STD_OP; /* subsequent writes fast -
889                                     15 seconds is plenty */
890         }
891
892         cifs_stats_bytes_written(pTcon, total_written);
893
894         /* since the write may have blocked check these pointers again */
895         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
896                 struct inode *inode = file->f_path.dentry->d_inode;
897 /* Do not update local mtime - server will set its actual value on write
898  *              inode->i_ctime = inode->i_mtime =
899  *                      current_fs_time(inode->i_sb);*/
900                 if (total_written > 0) {
901                         spin_lock(&inode->i_lock);
902                         if (*poffset > file->f_path.dentry->d_inode->i_size)
903                                 i_size_write(file->f_path.dentry->d_inode,
904                                         *poffset);
905                         spin_unlock(&inode->i_lock);
906                 }
907                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
908         }
909         FreeXid(xid);
910         return total_written;
911 }
912
913 static ssize_t cifs_write(struct file *file, const char *write_data,
914         size_t write_size, loff_t *poffset)
915 {
916         int rc = 0;
917         unsigned int bytes_written = 0;
918         unsigned int total_written;
919         struct cifs_sb_info *cifs_sb;
920         struct cifsTconInfo *pTcon;
921         int xid, long_op;
922         struct cifsFileInfo *open_file;
923
924         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
925
926         pTcon = cifs_sb->tcon;
927
928         cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
929            *poffset, file->f_path.dentry->d_name.name));
930
931         if (file->private_data == NULL)
932                 return -EBADF;
933         open_file = (struct cifsFileInfo *)file->private_data;
934
935         xid = GetXid();
936
937         if (*poffset > file->f_path.dentry->d_inode->i_size)
938                 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
939         else
940                 long_op = CIFS_LONG_OP;
941
942         for (total_written = 0; write_size > total_written;
943              total_written += bytes_written) {
944                 rc = -EAGAIN;
945                 while (rc == -EAGAIN) {
946                         if (file->private_data == NULL) {
947                                 /* file has been closed on us */
948                                 FreeXid(xid);
949                         /* if we have gotten here we have written some data
950                            and blocked, and the file has been freed on us
951                            while we blocked so return what we managed to
952                            write */
953                                 return total_written;
954                         }
955                         if (open_file->closePend) {
956                                 FreeXid(xid);
957                                 if (total_written)
958                                         return total_written;
959                                 else
960                                         return -EBADF;
961                         }
962                         if (open_file->invalidHandle) {
963                                 /* we could deadlock if we called
964                                    filemap_fdatawait from here so tell
965                                    reopen_file not to flush data to
966                                    server now */
967                                 rc = cifs_reopen_file(file, false);
968                                 if (rc != 0)
969                                         break;
970                         }
971                         if (experimEnabled || (pTcon->ses->server &&
972                                 ((pTcon->ses->server->secMode &
973                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
974                                 == 0))) {
975                                 struct kvec iov[2];
976                                 unsigned int len;
977
978                                 len = min((size_t)cifs_sb->wsize,
979                                           write_size - total_written);
980                                 /* iov[0] is reserved for smb header */
981                                 iov[1].iov_base = (char *)write_data +
982                                                   total_written;
983                                 iov[1].iov_len = len;
984                                 rc = CIFSSMBWrite2(xid, pTcon,
985                                                 open_file->netfid, len,
986                                                 *poffset, &bytes_written,
987                                                 iov, 1, long_op);
988                         } else
989                                 rc = CIFSSMBWrite(xid, pTcon,
990                                          open_file->netfid,
991                                          min_t(const int, cifs_sb->wsize,
992                                                write_size - total_written),
993                                          *poffset, &bytes_written,
994                                          write_data + total_written,
995                                          NULL, long_op);
996                 }
997                 if (rc || (bytes_written == 0)) {
998                         if (total_written)
999                                 break;
1000                         else {
1001                                 FreeXid(xid);
1002                                 return rc;
1003                         }
1004                 } else
1005                         *poffset += bytes_written;
1006                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1007                                     15 seconds is plenty */
1008         }
1009
1010         cifs_stats_bytes_written(pTcon, total_written);
1011
1012         /* since the write may have blocked check these pointers again */
1013         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1014 /*BB We could make this contingent on superblock ATIME flag too */
1015 /*              file->f_path.dentry->d_inode->i_ctime =
1016                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1017                 if (total_written > 0) {
1018                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1019                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1020                                 i_size_write(file->f_path.dentry->d_inode,
1021                                              *poffset);
1022                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1023                 }
1024                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1025         }
1026         FreeXid(xid);
1027         return total_written;
1028 }
1029
1030 #ifdef CONFIG_CIFS_EXPERIMENTAL
1031 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1032 {
1033         struct cifsFileInfo *open_file = NULL;
1034
1035         read_lock(&GlobalSMBSeslock);
1036         /* we could simply get the first_list_entry since write-only entries
1037            are always at the end of the list but since the first entry might
1038            have a close pending, we go through the whole list */
1039         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1040                 if (open_file->closePend)
1041                         continue;
1042                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1043                     (open_file->pfile->f_flags & O_RDONLY))) {
1044                         if (!open_file->invalidHandle) {
1045                                 /* found a good file */
1046                                 /* lock it so it will not be closed on us */
1047                                 atomic_inc(&open_file->wrtPending);
1048                                 read_unlock(&GlobalSMBSeslock);
1049                                 return open_file;
1050                         } /* else might as well continue, and look for
1051                              another, or simply have the caller reopen it
1052                              again rather than trying to fix this handle */
1053                 } else /* write only file */
1054                         break; /* write only files are last so must be done */
1055         }
1056         read_unlock(&GlobalSMBSeslock);
1057         return NULL;
1058 }
1059 #endif
1060
1061 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1062 {
1063         struct cifsFileInfo *open_file;
1064         int rc;
1065
1066         /* Having a null inode here (because mapping->host was set to zero by
1067         the VFS or MM) should not happen but we had reports of on oops (due to
1068         it being zero) during stress testcases so we need to check for it */
1069
1070         if (cifs_inode == NULL) {
1071                 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1072                 dump_stack();
1073                 return NULL;
1074         }
1075
1076         read_lock(&GlobalSMBSeslock);
1077 refind_writable:
1078         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1079                 if (open_file->closePend)
1080                         continue;
1081                 if (open_file->pfile &&
1082                     ((open_file->pfile->f_flags & O_RDWR) ||
1083                      (open_file->pfile->f_flags & O_WRONLY))) {
1084                         atomic_inc(&open_file->wrtPending);
1085
1086                         if (!open_file->invalidHandle) {
1087                                 /* found a good writable file */
1088                                 read_unlock(&GlobalSMBSeslock);
1089                                 return open_file;
1090                         }
1091
1092                         read_unlock(&GlobalSMBSeslock);
1093                         /* Had to unlock since following call can block */
1094                         rc = cifs_reopen_file(open_file->pfile, false);
1095                         if (!rc) {
1096                                 if (!open_file->closePend)
1097                                         return open_file;
1098                                 else { /* start over in case this was deleted */
1099                                        /* since the list could be modified */
1100                                         read_lock(&GlobalSMBSeslock);
1101                                         atomic_dec(&open_file->wrtPending);
1102                                         goto refind_writable;
1103                                 }
1104                         }
1105
1106                         /* if it fails, try another handle if possible -
1107                         (we can not do this if closePending since
1108                         loop could be modified - in which case we
1109                         have to start at the beginning of the list
1110                         again. Note that it would be bad
1111                         to hold up writepages here (rather than
1112                         in caller) with continuous retries */
1113                         cFYI(1, ("wp failed on reopen file"));
1114                         read_lock(&GlobalSMBSeslock);
1115                         /* can not use this handle, no write
1116                            pending on this one after all */
1117                         atomic_dec(&open_file->wrtPending);
1118
1119                         if (open_file->closePend) /* list could have changed */
1120                                 goto refind_writable;
1121                         /* else we simply continue to the next entry. Thus
1122                            we do not loop on reopen errors.  If we
1123                            can not reopen the file, for example if we
1124                            reconnected to a server with another client
1125                            racing to delete or lock the file we would not
1126                            make progress if we restarted before the beginning
1127                            of the loop here. */
1128                 }
1129         }
1130         read_unlock(&GlobalSMBSeslock);
1131         return NULL;
1132 }
1133
1134 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1135 {
1136         struct address_space *mapping = page->mapping;
1137         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1138         char *write_data;
1139         int rc = -EFAULT;
1140         int bytes_written = 0;
1141         struct cifs_sb_info *cifs_sb;
1142         struct cifsTconInfo *pTcon;
1143         struct inode *inode;
1144         struct cifsFileInfo *open_file;
1145
1146         if (!mapping || !mapping->host)
1147                 return -EFAULT;
1148
1149         inode = page->mapping->host;
1150         cifs_sb = CIFS_SB(inode->i_sb);
1151         pTcon = cifs_sb->tcon;
1152
1153         offset += (loff_t)from;
1154         write_data = kmap(page);
1155         write_data += from;
1156
1157         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1158                 kunmap(page);
1159                 return -EIO;
1160         }
1161
1162         /* racing with truncate? */
1163         if (offset > mapping->host->i_size) {
1164                 kunmap(page);
1165                 return 0; /* don't care */
1166         }
1167
1168         /* check to make sure that we are not extending the file */
1169         if (mapping->host->i_size - offset < (loff_t)to)
1170                 to = (unsigned)(mapping->host->i_size - offset);
1171
1172         open_file = find_writable_file(CIFS_I(mapping->host));
1173         if (open_file) {
1174                 bytes_written = cifs_write(open_file->pfile, write_data,
1175                                            to-from, &offset);
1176                 atomic_dec(&open_file->wrtPending);
1177                 /* Does mm or vfs already set times? */
1178                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1179                 if ((bytes_written > 0) && (offset))
1180                         rc = 0;
1181                 else if (bytes_written < 0)
1182                         rc = bytes_written;
1183         } else {
1184                 cFYI(1, ("No writeable filehandles for inode"));
1185                 rc = -EIO;
1186         }
1187
1188         kunmap(page);
1189         return rc;
1190 }
1191
1192 static int cifs_writepages(struct address_space *mapping,
1193                            struct writeback_control *wbc)
1194 {
1195         struct backing_dev_info *bdi = mapping->backing_dev_info;
1196         unsigned int bytes_to_write;
1197         unsigned int bytes_written;
1198         struct cifs_sb_info *cifs_sb;
1199         int done = 0;
1200         pgoff_t end;
1201         pgoff_t index;
1202         int range_whole = 0;
1203         struct kvec *iov;
1204         int len;
1205         int n_iov = 0;
1206         pgoff_t next;
1207         int nr_pages;
1208         __u64 offset = 0;
1209         struct cifsFileInfo *open_file;
1210         struct page *page;
1211         struct pagevec pvec;
1212         int rc = 0;
1213         int scanned = 0;
1214         int xid;
1215
1216         cifs_sb = CIFS_SB(mapping->host->i_sb);
1217
1218         /*
1219          * If wsize is smaller that the page cache size, default to writing
1220          * one page at a time via cifs_writepage
1221          */
1222         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1223                 return generic_writepages(mapping, wbc);
1224
1225         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1226                 if (cifs_sb->tcon->ses->server->secMode &
1227                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1228                         if (!experimEnabled)
1229                                 return generic_writepages(mapping, wbc);
1230
1231         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1232         if (iov == NULL)
1233                 return generic_writepages(mapping, wbc);
1234
1235
1236         /*
1237          * BB: Is this meaningful for a non-block-device file system?
1238          * If it is, we should test it again after we do I/O
1239          */
1240         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1241                 wbc->encountered_congestion = 1;
1242                 kfree(iov);
1243                 return 0;
1244         }
1245
1246         xid = GetXid();
1247
1248         pagevec_init(&pvec, 0);
1249         if (wbc->range_cyclic) {
1250                 index = mapping->writeback_index; /* Start from prev offset */
1251                 end = -1;
1252         } else {
1253                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1254                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1255                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1256                         range_whole = 1;
1257                 scanned = 1;
1258         }
1259 retry:
1260         while (!done && (index <= end) &&
1261                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1262                         PAGECACHE_TAG_DIRTY,
1263                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1264                 int first;
1265                 unsigned int i;
1266
1267                 first = -1;
1268                 next = 0;
1269                 n_iov = 0;
1270                 bytes_to_write = 0;
1271
1272                 for (i = 0; i < nr_pages; i++) {
1273                         page = pvec.pages[i];
1274                         /*
1275                          * At this point we hold neither mapping->tree_lock nor
1276                          * lock on the page itself: the page may be truncated or
1277                          * invalidated (changing page->mapping to NULL), or even
1278                          * swizzled back from swapper_space to tmpfs file
1279                          * mapping
1280                          */
1281
1282                         if (first < 0)
1283                                 lock_page(page);
1284                         else if (!trylock_page(page))
1285                                 break;
1286
1287                         if (unlikely(page->mapping != mapping)) {
1288                                 unlock_page(page);
1289                                 break;
1290                         }
1291
1292                         if (!wbc->range_cyclic && page->index > end) {
1293                                 done = 1;
1294                                 unlock_page(page);
1295                                 break;
1296                         }
1297
1298                         if (next && (page->index != next)) {
1299                                 /* Not next consecutive page */
1300                                 unlock_page(page);
1301                                 break;
1302                         }
1303
1304                         if (wbc->sync_mode != WB_SYNC_NONE)
1305                                 wait_on_page_writeback(page);
1306
1307                         if (PageWriteback(page) ||
1308                                         !clear_page_dirty_for_io(page)) {
1309                                 unlock_page(page);
1310                                 break;
1311                         }
1312
1313                         /*
1314                          * This actually clears the dirty bit in the radix tree.
1315                          * See cifs_writepage() for more commentary.
1316                          */
1317                         set_page_writeback(page);
1318
1319                         if (page_offset(page) >= mapping->host->i_size) {
1320                                 done = 1;
1321                                 unlock_page(page);
1322                                 end_page_writeback(page);
1323                                 break;
1324                         }
1325
1326                         /*
1327                          * BB can we get rid of this?  pages are held by pvec
1328                          */
1329                         page_cache_get(page);
1330
1331                         len = min(mapping->host->i_size - page_offset(page),
1332                                   (loff_t)PAGE_CACHE_SIZE);
1333
1334                         /* reserve iov[0] for the smb header */
1335                         n_iov++;
1336                         iov[n_iov].iov_base = kmap(page);
1337                         iov[n_iov].iov_len = len;
1338                         bytes_to_write += len;
1339
1340                         if (first < 0) {
1341                                 first = i;
1342                                 offset = page_offset(page);
1343                         }
1344                         next = page->index + 1;
1345                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1346                                 break;
1347                 }
1348                 if (n_iov) {
1349                         /* Search for a writable handle every time we call
1350                          * CIFSSMBWrite2.  We can't rely on the last handle
1351                          * we used to still be valid
1352                          */
1353                         open_file = find_writable_file(CIFS_I(mapping->host));
1354                         if (!open_file) {
1355                                 cERROR(1, ("No writable handles for inode"));
1356                                 rc = -EBADF;
1357                         } else {
1358                                 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1359                                                    open_file->netfid,
1360                                                    bytes_to_write, offset,
1361                                                    &bytes_written, iov, n_iov,
1362                                                    CIFS_LONG_OP);
1363                                 atomic_dec(&open_file->wrtPending);
1364                                 if (rc || bytes_written < bytes_to_write) {
1365                                         cERROR(1, ("Write2 ret %d, wrote %d",
1366                                                   rc, bytes_written));
1367                                         /* BB what if continued retry is
1368                                            requested via mount flags? */
1369                                         if (rc == -ENOSPC)
1370                                                 set_bit(AS_ENOSPC, &mapping->flags);
1371                                         else
1372                                                 set_bit(AS_EIO, &mapping->flags);
1373                                 } else {
1374                                         cifs_stats_bytes_written(cifs_sb->tcon,
1375                                                                  bytes_written);
1376                                 }
1377                         }
1378                         for (i = 0; i < n_iov; i++) {
1379                                 page = pvec.pages[first + i];
1380                                 /* Should we also set page error on
1381                                 success rc but too little data written? */
1382                                 /* BB investigate retry logic on temporary
1383                                 server crash cases and how recovery works
1384                                 when page marked as error */
1385                                 if (rc)
1386                                         SetPageError(page);
1387                                 kunmap(page);
1388                                 unlock_page(page);
1389                                 end_page_writeback(page);
1390                                 page_cache_release(page);
1391                         }
1392                         if ((wbc->nr_to_write -= n_iov) <= 0)
1393                                 done = 1;
1394                         index = next;
1395                 }
1396                 pagevec_release(&pvec);
1397         }
1398         if (!scanned && !done) {
1399                 /*
1400                  * We hit the last page and there is more work to be done: wrap
1401                  * back to the start of the file
1402                  */
1403                 scanned = 1;
1404                 index = 0;
1405                 goto retry;
1406         }
1407         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1408                 mapping->writeback_index = index;
1409
1410         FreeXid(xid);
1411         kfree(iov);
1412         return rc;
1413 }
1414
1415 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1416 {
1417         int rc = -EFAULT;
1418         int xid;
1419
1420         xid = GetXid();
1421 /* BB add check for wbc flags */
1422         page_cache_get(page);
1423         if (!PageUptodate(page))
1424                 cFYI(1, ("ppw - page not up to date"));
1425
1426         /*
1427          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1428          *
1429          * A writepage() implementation always needs to do either this,
1430          * or re-dirty the page with "redirty_page_for_writepage()" in
1431          * the case of a failure.
1432          *
1433          * Just unlocking the page will cause the radix tree tag-bits
1434          * to fail to update with the state of the page correctly.
1435          */
1436         set_page_writeback(page);
1437         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1438         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1439         unlock_page(page);
1440         end_page_writeback(page);
1441         page_cache_release(page);
1442         FreeXid(xid);
1443         return rc;
1444 }
1445
1446 static int cifs_commit_write(struct file *file, struct page *page,
1447         unsigned offset, unsigned to)
1448 {
1449         int xid;
1450         int rc = 0;
1451         struct inode *inode = page->mapping->host;
1452         loff_t position = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1453         char *page_data;
1454
1455         xid = GetXid();
1456         cFYI(1, ("commit write for page %p up to position %lld for %d",
1457                  page, position, to));
1458         spin_lock(&inode->i_lock);
1459         if (position > inode->i_size)
1460                 i_size_write(inode, position);
1461
1462         spin_unlock(&inode->i_lock);
1463         if (!PageUptodate(page)) {
1464                 position =  ((loff_t)page->index << PAGE_CACHE_SHIFT) + offset;
1465                 /* can not rely on (or let) writepage write this data */
1466                 if (to < offset) {
1467                         cFYI(1, ("Illegal offsets, can not copy from %d to %d",
1468                                 offset, to));
1469                         FreeXid(xid);
1470                         return rc;
1471                 }
1472                 /* this is probably better than directly calling
1473                    partialpage_write since in this function the file handle is
1474                    known which we might as well leverage */
1475                 /* BB check if anything else missing out of ppw
1476                    such as updating last write time */
1477                 page_data = kmap(page);
1478                 rc = cifs_write(file, page_data + offset, to-offset,
1479                                 &position);
1480                 if (rc > 0)
1481                         rc = 0;
1482                 /* else if (rc < 0) should we set writebehind rc? */
1483                 kunmap(page);
1484         } else {
1485                 set_page_dirty(page);
1486         }
1487
1488         FreeXid(xid);
1489         return rc;
1490 }
1491
1492 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1493 {
1494         int xid;
1495         int rc = 0;
1496         struct inode *inode = file->f_path.dentry->d_inode;
1497
1498         xid = GetXid();
1499
1500         cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1501                 dentry->d_name.name, datasync));
1502
1503         rc = filemap_write_and_wait(inode->i_mapping);
1504         if (rc == 0) {
1505                 rc = CIFS_I(inode)->write_behind_rc;
1506                 CIFS_I(inode)->write_behind_rc = 0;
1507         }
1508         FreeXid(xid);
1509         return rc;
1510 }
1511
1512 /* static void cifs_sync_page(struct page *page)
1513 {
1514         struct address_space *mapping;
1515         struct inode *inode;
1516         unsigned long index = page->index;
1517         unsigned int rpages = 0;
1518         int rc = 0;
1519
1520         cFYI(1, ("sync page %p",page));
1521         mapping = page->mapping;
1522         if (!mapping)
1523                 return 0;
1524         inode = mapping->host;
1525         if (!inode)
1526                 return; */
1527
1528 /*      fill in rpages then
1529         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1530
1531 /*      cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1532
1533 #if 0
1534         if (rc < 0)
1535                 return rc;
1536         return 0;
1537 #endif
1538 } */
1539
1540 /*
1541  * As file closes, flush all cached write data for this inode checking
1542  * for write behind errors.
1543  */
1544 int cifs_flush(struct file *file, fl_owner_t id)
1545 {
1546         struct inode *inode = file->f_path.dentry->d_inode;
1547         int rc = 0;
1548
1549         /* Rather than do the steps manually:
1550            lock the inode for writing
1551            loop through pages looking for write behind data (dirty pages)
1552            coalesce into contiguous 16K (or smaller) chunks to write to server
1553            send to server (prefer in parallel)
1554            deal with writebehind errors
1555            unlock inode for writing
1556            filemapfdatawrite appears easier for the time being */
1557
1558         rc = filemap_fdatawrite(inode->i_mapping);
1559         /* reset wb rc if we were able to write out dirty pages */
1560         if (!rc) {
1561                 rc = CIFS_I(inode)->write_behind_rc;
1562                 CIFS_I(inode)->write_behind_rc = 0;
1563         }
1564
1565         cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1566
1567         return rc;
1568 }
1569
1570 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1571         size_t read_size, loff_t *poffset)
1572 {
1573         int rc = -EACCES;
1574         unsigned int bytes_read = 0;
1575         unsigned int total_read = 0;
1576         unsigned int current_read_size;
1577         struct cifs_sb_info *cifs_sb;
1578         struct cifsTconInfo *pTcon;
1579         int xid;
1580         struct cifsFileInfo *open_file;
1581         char *smb_read_data;
1582         char __user *current_offset;
1583         struct smb_com_read_rsp *pSMBr;
1584
1585         xid = GetXid();
1586         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1587         pTcon = cifs_sb->tcon;
1588
1589         if (file->private_data == NULL) {
1590                 FreeXid(xid);
1591                 return -EBADF;
1592         }
1593         open_file = (struct cifsFileInfo *)file->private_data;
1594
1595         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1596                 cFYI(1, ("attempting read on write only file instance"));
1597
1598         for (total_read = 0, current_offset = read_data;
1599              read_size > total_read;
1600              total_read += bytes_read, current_offset += bytes_read) {
1601                 current_read_size = min_t(const int, read_size - total_read,
1602                                           cifs_sb->rsize);
1603                 rc = -EAGAIN;
1604                 smb_read_data = NULL;
1605                 while (rc == -EAGAIN) {
1606                         int buf_type = CIFS_NO_BUFFER;
1607                         if ((open_file->invalidHandle) &&
1608                             (!open_file->closePend)) {
1609                                 rc = cifs_reopen_file(file, true);
1610                                 if (rc != 0)
1611                                         break;
1612                         }
1613                         rc = CIFSSMBRead(xid, pTcon,
1614                                          open_file->netfid,
1615                                          current_read_size, *poffset,
1616                                          &bytes_read, &smb_read_data,
1617                                          &buf_type);
1618                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1619                         if (smb_read_data) {
1620                                 if (copy_to_user(current_offset,
1621                                                 smb_read_data +
1622                                                 4 /* RFC1001 length field */ +
1623                                                 le16_to_cpu(pSMBr->DataOffset),
1624                                                 bytes_read))
1625                                         rc = -EFAULT;
1626
1627                                 if (buf_type == CIFS_SMALL_BUFFER)
1628                                         cifs_small_buf_release(smb_read_data);
1629                                 else if (buf_type == CIFS_LARGE_BUFFER)
1630                                         cifs_buf_release(smb_read_data);
1631                                 smb_read_data = NULL;
1632                         }
1633                 }
1634                 if (rc || (bytes_read == 0)) {
1635                         if (total_read) {
1636                                 break;
1637                         } else {
1638                                 FreeXid(xid);
1639                                 return rc;
1640                         }
1641                 } else {
1642                         cifs_stats_bytes_read(pTcon, bytes_read);
1643                         *poffset += bytes_read;
1644                 }
1645         }
1646         FreeXid(xid);
1647         return total_read;
1648 }
1649
1650
1651 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1652         loff_t *poffset)
1653 {
1654         int rc = -EACCES;
1655         unsigned int bytes_read = 0;
1656         unsigned int total_read;
1657         unsigned int current_read_size;
1658         struct cifs_sb_info *cifs_sb;
1659         struct cifsTconInfo *pTcon;
1660         int xid;
1661         char *current_offset;
1662         struct cifsFileInfo *open_file;
1663         int buf_type = CIFS_NO_BUFFER;
1664
1665         xid = GetXid();
1666         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1667         pTcon = cifs_sb->tcon;
1668
1669         if (file->private_data == NULL) {
1670                 FreeXid(xid);
1671                 return -EBADF;
1672         }
1673         open_file = (struct cifsFileInfo *)file->private_data;
1674
1675         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1676                 cFYI(1, ("attempting read on write only file instance"));
1677
1678         for (total_read = 0, current_offset = read_data;
1679              read_size > total_read;
1680              total_read += bytes_read, current_offset += bytes_read) {
1681                 current_read_size = min_t(const int, read_size - total_read,
1682                                           cifs_sb->rsize);
1683                 /* For windows me and 9x we do not want to request more
1684                 than it negotiated since it will refuse the read then */
1685                 if ((pTcon->ses) &&
1686                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1687                         current_read_size = min_t(const int, current_read_size,
1688                                         pTcon->ses->server->maxBuf - 128);
1689                 }
1690                 rc = -EAGAIN;
1691                 while (rc == -EAGAIN) {
1692                         if ((open_file->invalidHandle) &&
1693                             (!open_file->closePend)) {
1694                                 rc = cifs_reopen_file(file, true);
1695                                 if (rc != 0)
1696                                         break;
1697                         }
1698                         rc = CIFSSMBRead(xid, pTcon,
1699                                          open_file->netfid,
1700                                          current_read_size, *poffset,
1701                                          &bytes_read, &current_offset,
1702                                          &buf_type);
1703                 }
1704                 if (rc || (bytes_read == 0)) {
1705                         if (total_read) {
1706                                 break;
1707                         } else {
1708                                 FreeXid(xid);
1709                                 return rc;
1710                         }
1711                 } else {
1712                         cifs_stats_bytes_read(pTcon, total_read);
1713                         *poffset += bytes_read;
1714                 }
1715         }
1716         FreeXid(xid);
1717         return total_read;
1718 }
1719
1720 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1721 {
1722         struct dentry *dentry = file->f_path.dentry;
1723         int rc, xid;
1724
1725         xid = GetXid();
1726         rc = cifs_revalidate(dentry);
1727         if (rc) {
1728                 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1729                 FreeXid(xid);
1730                 return rc;
1731         }
1732         rc = generic_file_mmap(file, vma);
1733         FreeXid(xid);
1734         return rc;
1735 }
1736
1737
1738 static void cifs_copy_cache_pages(struct address_space *mapping,
1739         struct list_head *pages, int bytes_read, char *data,
1740         struct pagevec *plru_pvec)
1741 {
1742         struct page *page;
1743         char *target;
1744
1745         while (bytes_read > 0) {
1746                 if (list_empty(pages))
1747                         break;
1748
1749                 page = list_entry(pages->prev, struct page, lru);
1750                 list_del(&page->lru);
1751
1752                 if (add_to_page_cache(page, mapping, page->index,
1753                                       GFP_KERNEL)) {
1754                         page_cache_release(page);
1755                         cFYI(1, ("Add page cache failed"));
1756                         data += PAGE_CACHE_SIZE;
1757                         bytes_read -= PAGE_CACHE_SIZE;
1758                         continue;
1759                 }
1760
1761                 target = kmap_atomic(page, KM_USER0);
1762
1763                 if (PAGE_CACHE_SIZE > bytes_read) {
1764                         memcpy(target, data, bytes_read);
1765                         /* zero the tail end of this partial page */
1766                         memset(target + bytes_read, 0,
1767                                PAGE_CACHE_SIZE - bytes_read);
1768                         bytes_read = 0;
1769                 } else {
1770                         memcpy(target, data, PAGE_CACHE_SIZE);
1771                         bytes_read -= PAGE_CACHE_SIZE;
1772                 }
1773                 kunmap_atomic(target, KM_USER0);
1774
1775                 flush_dcache_page(page);
1776                 SetPageUptodate(page);
1777                 unlock_page(page);
1778                 if (!pagevec_add(plru_pvec, page))
1779                         __pagevec_lru_add(plru_pvec);
1780                 data += PAGE_CACHE_SIZE;
1781         }
1782         return;
1783 }
1784
1785 static int cifs_readpages(struct file *file, struct address_space *mapping,
1786         struct list_head *page_list, unsigned num_pages)
1787 {
1788         int rc = -EACCES;
1789         int xid;
1790         loff_t offset;
1791         struct page *page;
1792         struct cifs_sb_info *cifs_sb;
1793         struct cifsTconInfo *pTcon;
1794         unsigned int bytes_read = 0;
1795         unsigned int read_size, i;
1796         char *smb_read_data = NULL;
1797         struct smb_com_read_rsp *pSMBr;
1798         struct pagevec lru_pvec;
1799         struct cifsFileInfo *open_file;
1800         int buf_type = CIFS_NO_BUFFER;
1801
1802         xid = GetXid();
1803         if (file->private_data == NULL) {
1804                 FreeXid(xid);
1805                 return -EBADF;
1806         }
1807         open_file = (struct cifsFileInfo *)file->private_data;
1808         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1809         pTcon = cifs_sb->tcon;
1810
1811         pagevec_init(&lru_pvec, 0);
1812                 cFYI(DBG2, ("rpages: num pages %d", num_pages));
1813         for (i = 0; i < num_pages; ) {
1814                 unsigned contig_pages;
1815                 struct page *tmp_page;
1816                 unsigned long expected_index;
1817
1818                 if (list_empty(page_list))
1819                         break;
1820
1821                 page = list_entry(page_list->prev, struct page, lru);
1822                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1823
1824                 /* count adjacent pages that we will read into */
1825                 contig_pages = 0;
1826                 expected_index =
1827                         list_entry(page_list->prev, struct page, lru)->index;
1828                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1829                         if (tmp_page->index == expected_index) {
1830                                 contig_pages++;
1831                                 expected_index++;
1832                         } else
1833                                 break;
1834                 }
1835                 if (contig_pages + i >  num_pages)
1836                         contig_pages = num_pages - i;
1837
1838                 /* for reads over a certain size could initiate async
1839                    read ahead */
1840
1841                 read_size = contig_pages * PAGE_CACHE_SIZE;
1842                 /* Read size needs to be in multiples of one page */
1843                 read_size = min_t(const unsigned int, read_size,
1844                                   cifs_sb->rsize & PAGE_CACHE_MASK);
1845                 cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
1846                                 read_size, contig_pages));
1847                 rc = -EAGAIN;
1848                 while (rc == -EAGAIN) {
1849                         if ((open_file->invalidHandle) &&
1850                             (!open_file->closePend)) {
1851                                 rc = cifs_reopen_file(file, true);
1852                                 if (rc != 0)
1853                                         break;
1854                         }
1855
1856                         rc = CIFSSMBRead(xid, pTcon,
1857                                          open_file->netfid,
1858                                          read_size, offset,
1859                                          &bytes_read, &smb_read_data,
1860                                          &buf_type);
1861                         /* BB more RC checks ? */
1862                         if (rc == -EAGAIN) {
1863                                 if (smb_read_data) {
1864                                         if (buf_type == CIFS_SMALL_BUFFER)
1865                                                 cifs_small_buf_release(smb_read_data);
1866                                         else if (buf_type == CIFS_LARGE_BUFFER)
1867                                                 cifs_buf_release(smb_read_data);
1868                                         smb_read_data = NULL;
1869                                 }
1870                         }
1871                 }
1872                 if ((rc < 0) || (smb_read_data == NULL)) {
1873                         cFYI(1, ("Read error in readpages: %d", rc));
1874                         break;
1875                 } else if (bytes_read > 0) {
1876                         task_io_account_read(bytes_read);
1877                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1878                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
1879                                 smb_read_data + 4 /* RFC1001 hdr */ +
1880                                 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1881
1882                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
1883                         cifs_stats_bytes_read(pTcon, bytes_read);
1884                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1885                                 i++; /* account for partial page */
1886
1887                                 /* server copy of file can have smaller size
1888                                    than client */
1889                                 /* BB do we need to verify this common case ?
1890                                    this case is ok - if we are at server EOF
1891                                    we will hit it on next read */
1892
1893                                 /* break; */
1894                         }
1895                 } else {
1896                         cFYI(1, ("No bytes read (%d) at offset %lld . "
1897                                  "Cleaning remaining pages from readahead list",
1898                                  bytes_read, offset));
1899                         /* BB turn off caching and do new lookup on
1900                            file size at server? */
1901                         break;
1902                 }
1903                 if (smb_read_data) {
1904                         if (buf_type == CIFS_SMALL_BUFFER)
1905                                 cifs_small_buf_release(smb_read_data);
1906                         else if (buf_type == CIFS_LARGE_BUFFER)
1907                                 cifs_buf_release(smb_read_data);
1908                         smb_read_data = NULL;
1909                 }
1910                 bytes_read = 0;
1911         }
1912
1913         pagevec_lru_add(&lru_pvec);
1914
1915 /* need to free smb_read_data buf before exit */
1916         if (smb_read_data) {
1917                 if (buf_type == CIFS_SMALL_BUFFER)
1918                         cifs_small_buf_release(smb_read_data);
1919                 else if (buf_type == CIFS_LARGE_BUFFER)
1920                         cifs_buf_release(smb_read_data);
1921                 smb_read_data = NULL;
1922         }
1923
1924         FreeXid(xid);
1925         return rc;
1926 }
1927
1928 static int cifs_readpage_worker(struct file *file, struct page *page,
1929         loff_t *poffset)
1930 {
1931         char *read_data;
1932         int rc;
1933
1934         page_cache_get(page);
1935         read_data = kmap(page);
1936         /* for reads over a certain size could initiate async read ahead */
1937
1938         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1939
1940         if (rc < 0)
1941                 goto io_error;
1942         else
1943                 cFYI(1, ("Bytes read %d", rc));
1944
1945         file->f_path.dentry->d_inode->i_atime =
1946                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1947
1948         if (PAGE_CACHE_SIZE > rc)
1949                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1950
1951         flush_dcache_page(page);
1952         SetPageUptodate(page);
1953         rc = 0;
1954
1955 io_error:
1956         kunmap(page);
1957         page_cache_release(page);
1958         return rc;
1959 }
1960
1961 static int cifs_readpage(struct file *file, struct page *page)
1962 {
1963         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1964         int rc = -EACCES;
1965         int xid;
1966
1967         xid = GetXid();
1968
1969         if (file->private_data == NULL) {
1970                 FreeXid(xid);
1971                 return -EBADF;
1972         }
1973
1974         cFYI(1, ("readpage %p at offset %d 0x%x\n",
1975                  page, (int)offset, (int)offset));
1976
1977         rc = cifs_readpage_worker(file, page, &offset);
1978
1979         unlock_page(page);
1980
1981         FreeXid(xid);
1982         return rc;
1983 }
1984
1985 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
1986 {
1987         struct cifsFileInfo *open_file;
1988
1989         read_lock(&GlobalSMBSeslock);
1990         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1991                 if (open_file->closePend)
1992                         continue;
1993                 if (open_file->pfile &&
1994                     ((open_file->pfile->f_flags & O_RDWR) ||
1995                      (open_file->pfile->f_flags & O_WRONLY))) {
1996                         read_unlock(&GlobalSMBSeslock);
1997                         return 1;
1998                 }
1999         }
2000         read_unlock(&GlobalSMBSeslock);
2001         return 0;
2002 }
2003
2004 /* We do not want to update the file size from server for inodes
2005    open for write - to avoid races with writepage extending
2006    the file - in the future we could consider allowing
2007    refreshing the inode only on increases in the file size
2008    but this is tricky to do without racing with writebehind
2009    page caching in the current Linux kernel design */
2010 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2011 {
2012         if (!cifsInode)
2013                 return true;
2014
2015         if (is_inode_writable(cifsInode)) {
2016                 /* This inode is open for write at least once */
2017                 struct cifs_sb_info *cifs_sb;
2018
2019                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2020                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2021                         /* since no page cache to corrupt on directio
2022                         we can change size safely */
2023                         return true;
2024                 }
2025
2026                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2027                         return true;
2028
2029                 return false;
2030         } else
2031                 return true;
2032 }
2033
2034 static int cifs_prepare_write(struct file *file, struct page *page,
2035         unsigned from, unsigned to)
2036 {
2037         int rc = 0;
2038         loff_t i_size;
2039         loff_t offset;
2040
2041         cFYI(1, ("prepare write for page %p from %d to %d", page, from, to));
2042         if (PageUptodate(page))
2043                 return 0;
2044
2045         /* If we are writing a full page it will be up to date,
2046            no need to read from the server */
2047         if ((to == PAGE_CACHE_SIZE) && (from == 0)) {
2048                 SetPageUptodate(page);
2049                 return 0;
2050         }
2051
2052         offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2053         i_size = i_size_read(page->mapping->host);
2054
2055         if ((offset >= i_size) ||
2056             ((from == 0) && (offset + to) >= i_size)) {
2057                 /*
2058                  * We don't need to read data beyond the end of the file.
2059                  * zero it, and set the page uptodate
2060                  */
2061                 simple_prepare_write(file, page, from, to);
2062                 SetPageUptodate(page);
2063         } else if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2064                 /* might as well read a page, it is fast enough */
2065                 rc = cifs_readpage_worker(file, page, &offset);
2066         } else {
2067                 /* we could try using another file handle if there is one -
2068                    but how would we lock it to prevent close of that handle
2069                    racing with this read? In any case
2070                    this will be written out by commit_write so is fine */
2071         }
2072
2073         /* we do not need to pass errors back
2074            e.g. if we do not have read access to the file
2075            because cifs_commit_write will do the right thing.  -- shaggy */
2076
2077         return 0;
2078 }
2079
2080 const struct address_space_operations cifs_addr_ops = {
2081         .readpage = cifs_readpage,
2082         .readpages = cifs_readpages,
2083         .writepage = cifs_writepage,
2084         .writepages = cifs_writepages,
2085         .prepare_write = cifs_prepare_write,
2086         .commit_write = cifs_commit_write,
2087         .set_page_dirty = __set_page_dirty_nobuffers,
2088         /* .sync_page = cifs_sync_page, */
2089         /* .direct_IO = */
2090 };
2091
2092 /*
2093  * cifs_readpages requires the server to support a buffer large enough to
2094  * contain the header plus one complete page of data.  Otherwise, we need
2095  * to leave cifs_readpages out of the address space operations.
2096  */
2097 const struct address_space_operations cifs_addr_ops_smallbuf = {
2098         .readpage = cifs_readpage,
2099         .writepage = cifs_writepage,
2100         .writepages = cifs_writepages,
2101         .prepare_write = cifs_prepare_write,
2102         .commit_write = cifs_commit_write,
2103         .set_page_dirty = __set_page_dirty_nobuffers,
2104         /* .sync_page = cifs_sync_page, */
2105         /* .direct_IO = */
2106 };