prevent cifs_writepages() from skipping unwritten pages
[linux-2.6.git] / 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 write_begin */
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->need_reconnect) {
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         rc = generic_write_checks(file, poffset, &write_size, 0);
837         if (rc)
838                 return rc;
839
840         xid = GetXid();
841
842         if (*poffset > file->f_path.dentry->d_inode->i_size)
843                 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
844         else
845                 long_op = CIFS_LONG_OP;
846
847         for (total_written = 0; write_size > total_written;
848              total_written += bytes_written) {
849                 rc = -EAGAIN;
850                 while (rc == -EAGAIN) {
851                         if (file->private_data == NULL) {
852                                 /* file has been closed on us */
853                                 FreeXid(xid);
854                         /* if we have gotten here we have written some data
855                            and blocked, and the file has been freed on us while
856                            we blocked so return what we managed to write */
857                                 return total_written;
858                         }
859                         if (open_file->closePend) {
860                                 FreeXid(xid);
861                                 if (total_written)
862                                         return total_written;
863                                 else
864                                         return -EBADF;
865                         }
866                         if (open_file->invalidHandle) {
867                                 /* we could deadlock if we called
868                                    filemap_fdatawait from here so tell
869                                    reopen_file not to flush data to server
870                                    now */
871                                 rc = cifs_reopen_file(file, false);
872                                 if (rc != 0)
873                                         break;
874                         }
875
876                         rc = CIFSSMBWrite(xid, pTcon,
877                                 open_file->netfid,
878                                 min_t(const int, cifs_sb->wsize,
879                                       write_size - total_written),
880                                 *poffset, &bytes_written,
881                                 NULL, write_data + total_written, long_op);
882                 }
883                 if (rc || (bytes_written == 0)) {
884                         if (total_written)
885                                 break;
886                         else {
887                                 FreeXid(xid);
888                                 return rc;
889                         }
890                 } else
891                         *poffset += bytes_written;
892                 long_op = CIFS_STD_OP; /* subsequent writes fast -
893                                     15 seconds is plenty */
894         }
895
896         cifs_stats_bytes_written(pTcon, total_written);
897
898         /* since the write may have blocked check these pointers again */
899         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
900                 struct inode *inode = file->f_path.dentry->d_inode;
901 /* Do not update local mtime - server will set its actual value on write
902  *              inode->i_ctime = inode->i_mtime =
903  *                      current_fs_time(inode->i_sb);*/
904                 if (total_written > 0) {
905                         spin_lock(&inode->i_lock);
906                         if (*poffset > file->f_path.dentry->d_inode->i_size)
907                                 i_size_write(file->f_path.dentry->d_inode,
908                                         *poffset);
909                         spin_unlock(&inode->i_lock);
910                 }
911                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
912         }
913         FreeXid(xid);
914         return total_written;
915 }
916
917 static ssize_t cifs_write(struct file *file, const char *write_data,
918                           size_t write_size, loff_t *poffset)
919 {
920         int rc = 0;
921         unsigned int bytes_written = 0;
922         unsigned int total_written;
923         struct cifs_sb_info *cifs_sb;
924         struct cifsTconInfo *pTcon;
925         int xid, long_op;
926         struct cifsFileInfo *open_file;
927
928         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
929
930         pTcon = cifs_sb->tcon;
931
932         cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
933            *poffset, file->f_path.dentry->d_name.name));
934
935         if (file->private_data == NULL)
936                 return -EBADF;
937         open_file = (struct cifsFileInfo *)file->private_data;
938
939         xid = GetXid();
940
941         if (*poffset > file->f_path.dentry->d_inode->i_size)
942                 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
943         else
944                 long_op = CIFS_LONG_OP;
945
946         for (total_written = 0; write_size > total_written;
947              total_written += bytes_written) {
948                 rc = -EAGAIN;
949                 while (rc == -EAGAIN) {
950                         if (file->private_data == NULL) {
951                                 /* file has been closed on us */
952                                 FreeXid(xid);
953                         /* if we have gotten here we have written some data
954                            and blocked, and the file has been freed on us
955                            while we blocked so return what we managed to
956                            write */
957                                 return total_written;
958                         }
959                         if (open_file->closePend) {
960                                 FreeXid(xid);
961                                 if (total_written)
962                                         return total_written;
963                                 else
964                                         return -EBADF;
965                         }
966                         if (open_file->invalidHandle) {
967                                 /* we could deadlock if we called
968                                    filemap_fdatawait from here so tell
969                                    reopen_file not to flush data to
970                                    server now */
971                                 rc = cifs_reopen_file(file, false);
972                                 if (rc != 0)
973                                         break;
974                         }
975                         if (experimEnabled || (pTcon->ses->server &&
976                                 ((pTcon->ses->server->secMode &
977                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
978                                 == 0))) {
979                                 struct kvec iov[2];
980                                 unsigned int len;
981
982                                 len = min((size_t)cifs_sb->wsize,
983                                           write_size - total_written);
984                                 /* iov[0] is reserved for smb header */
985                                 iov[1].iov_base = (char *)write_data +
986                                                   total_written;
987                                 iov[1].iov_len = len;
988                                 rc = CIFSSMBWrite2(xid, pTcon,
989                                                 open_file->netfid, len,
990                                                 *poffset, &bytes_written,
991                                                 iov, 1, long_op);
992                         } else
993                                 rc = CIFSSMBWrite(xid, pTcon,
994                                          open_file->netfid,
995                                          min_t(const int, cifs_sb->wsize,
996                                                write_size - total_written),
997                                          *poffset, &bytes_written,
998                                          write_data + total_written,
999                                          NULL, long_op);
1000                 }
1001                 if (rc || (bytes_written == 0)) {
1002                         if (total_written)
1003                                 break;
1004                         else {
1005                                 FreeXid(xid);
1006                                 return rc;
1007                         }
1008                 } else
1009                         *poffset += bytes_written;
1010                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1011                                     15 seconds is plenty */
1012         }
1013
1014         cifs_stats_bytes_written(pTcon, total_written);
1015
1016         /* since the write may have blocked check these pointers again */
1017         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1018 /*BB We could make this contingent on superblock ATIME flag too */
1019 /*              file->f_path.dentry->d_inode->i_ctime =
1020                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1021                 if (total_written > 0) {
1022                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1023                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1024                                 i_size_write(file->f_path.dentry->d_inode,
1025                                              *poffset);
1026                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1027                 }
1028                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1029         }
1030         FreeXid(xid);
1031         return total_written;
1032 }
1033
1034 #ifdef CONFIG_CIFS_EXPERIMENTAL
1035 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1036 {
1037         struct cifsFileInfo *open_file = NULL;
1038
1039         read_lock(&GlobalSMBSeslock);
1040         /* we could simply get the first_list_entry since write-only entries
1041            are always at the end of the list but since the first entry might
1042            have a close pending, we go through the whole list */
1043         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1044                 if (open_file->closePend)
1045                         continue;
1046                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1047                     (open_file->pfile->f_flags & O_RDONLY))) {
1048                         if (!open_file->invalidHandle) {
1049                                 /* found a good file */
1050                                 /* lock it so it will not be closed on us */
1051                                 atomic_inc(&open_file->wrtPending);
1052                                 read_unlock(&GlobalSMBSeslock);
1053                                 return open_file;
1054                         } /* else might as well continue, and look for
1055                              another, or simply have the caller reopen it
1056                              again rather than trying to fix this handle */
1057                 } else /* write only file */
1058                         break; /* write only files are last so must be done */
1059         }
1060         read_unlock(&GlobalSMBSeslock);
1061         return NULL;
1062 }
1063 #endif
1064
1065 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1066 {
1067         struct cifsFileInfo *open_file;
1068         bool any_available = false;
1069         int rc;
1070
1071         /* Having a null inode here (because mapping->host was set to zero by
1072         the VFS or MM) should not happen but we had reports of on oops (due to
1073         it being zero) during stress testcases so we need to check for it */
1074
1075         if (cifs_inode == NULL) {
1076                 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1077                 dump_stack();
1078                 return NULL;
1079         }
1080
1081         read_lock(&GlobalSMBSeslock);
1082 refind_writable:
1083         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1084                 if (open_file->closePend ||
1085                     (!any_available && open_file->pid != current->tgid))
1086                         continue;
1087
1088                 if (open_file->pfile &&
1089                     ((open_file->pfile->f_flags & O_RDWR) ||
1090                      (open_file->pfile->f_flags & O_WRONLY))) {
1091                         atomic_inc(&open_file->wrtPending);
1092
1093                         if (!open_file->invalidHandle) {
1094                                 /* found a good writable file */
1095                                 read_unlock(&GlobalSMBSeslock);
1096                                 return open_file;
1097                         }
1098
1099                         read_unlock(&GlobalSMBSeslock);
1100                         /* Had to unlock since following call can block */
1101                         rc = cifs_reopen_file(open_file->pfile, false);
1102                         if (!rc) {
1103                                 if (!open_file->closePend)
1104                                         return open_file;
1105                                 else { /* start over in case this was deleted */
1106                                        /* since the list could be modified */
1107                                         read_lock(&GlobalSMBSeslock);
1108                                         atomic_dec(&open_file->wrtPending);
1109                                         goto refind_writable;
1110                                 }
1111                         }
1112
1113                         /* if it fails, try another handle if possible -
1114                         (we can not do this if closePending since
1115                         loop could be modified - in which case we
1116                         have to start at the beginning of the list
1117                         again. Note that it would be bad
1118                         to hold up writepages here (rather than
1119                         in caller) with continuous retries */
1120                         cFYI(1, ("wp failed on reopen file"));
1121                         read_lock(&GlobalSMBSeslock);
1122                         /* can not use this handle, no write
1123                            pending on this one after all */
1124                         atomic_dec(&open_file->wrtPending);
1125
1126                         if (open_file->closePend) /* list could have changed */
1127                                 goto refind_writable;
1128                         /* else we simply continue to the next entry. Thus
1129                            we do not loop on reopen errors.  If we
1130                            can not reopen the file, for example if we
1131                            reconnected to a server with another client
1132                            racing to delete or lock the file we would not
1133                            make progress if we restarted before the beginning
1134                            of the loop here. */
1135                 }
1136         }
1137         /* couldn't find useable FH with same pid, try any available */
1138         if (!any_available) {
1139                 any_available = true;
1140                 goto refind_writable;
1141         }
1142         read_unlock(&GlobalSMBSeslock);
1143         return NULL;
1144 }
1145
1146 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1147 {
1148         struct address_space *mapping = page->mapping;
1149         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1150         char *write_data;
1151         int rc = -EFAULT;
1152         int bytes_written = 0;
1153         struct cifs_sb_info *cifs_sb;
1154         struct cifsTconInfo *pTcon;
1155         struct inode *inode;
1156         struct cifsFileInfo *open_file;
1157
1158         if (!mapping || !mapping->host)
1159                 return -EFAULT;
1160
1161         inode = page->mapping->host;
1162         cifs_sb = CIFS_SB(inode->i_sb);
1163         pTcon = cifs_sb->tcon;
1164
1165         offset += (loff_t)from;
1166         write_data = kmap(page);
1167         write_data += from;
1168
1169         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1170                 kunmap(page);
1171                 return -EIO;
1172         }
1173
1174         /* racing with truncate? */
1175         if (offset > mapping->host->i_size) {
1176                 kunmap(page);
1177                 return 0; /* don't care */
1178         }
1179
1180         /* check to make sure that we are not extending the file */
1181         if (mapping->host->i_size - offset < (loff_t)to)
1182                 to = (unsigned)(mapping->host->i_size - offset);
1183
1184         open_file = find_writable_file(CIFS_I(mapping->host));
1185         if (open_file) {
1186                 bytes_written = cifs_write(open_file->pfile, write_data,
1187                                            to-from, &offset);
1188                 atomic_dec(&open_file->wrtPending);
1189                 /* Does mm or vfs already set times? */
1190                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1191                 if ((bytes_written > 0) && (offset))
1192                         rc = 0;
1193                 else if (bytes_written < 0)
1194                         rc = bytes_written;
1195         } else {
1196                 cFYI(1, ("No writeable filehandles for inode"));
1197                 rc = -EIO;
1198         }
1199
1200         kunmap(page);
1201         return rc;
1202 }
1203
1204 static int cifs_writepages(struct address_space *mapping,
1205                            struct writeback_control *wbc)
1206 {
1207         struct backing_dev_info *bdi = mapping->backing_dev_info;
1208         unsigned int bytes_to_write;
1209         unsigned int bytes_written;
1210         struct cifs_sb_info *cifs_sb;
1211         int done = 0;
1212         pgoff_t end;
1213         pgoff_t index;
1214         int range_whole = 0;
1215         struct kvec *iov;
1216         int len;
1217         int n_iov = 0;
1218         pgoff_t next;
1219         int nr_pages;
1220         __u64 offset = 0;
1221         struct cifsFileInfo *open_file;
1222         struct page *page;
1223         struct pagevec pvec;
1224         int rc = 0;
1225         int scanned = 0;
1226         int xid;
1227
1228         cifs_sb = CIFS_SB(mapping->host->i_sb);
1229
1230         /*
1231          * If wsize is smaller that the page cache size, default to writing
1232          * one page at a time via cifs_writepage
1233          */
1234         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1235                 return generic_writepages(mapping, wbc);
1236
1237         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1238                 if (cifs_sb->tcon->ses->server->secMode &
1239                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1240                         if (!experimEnabled)
1241                                 return generic_writepages(mapping, wbc);
1242
1243         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1244         if (iov == NULL)
1245                 return generic_writepages(mapping, wbc);
1246
1247
1248         /*
1249          * BB: Is this meaningful for a non-block-device file system?
1250          * If it is, we should test it again after we do I/O
1251          */
1252         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1253                 wbc->encountered_congestion = 1;
1254                 kfree(iov);
1255                 return 0;
1256         }
1257
1258         xid = GetXid();
1259
1260         pagevec_init(&pvec, 0);
1261         if (wbc->range_cyclic) {
1262                 index = mapping->writeback_index; /* Start from prev offset */
1263                 end = -1;
1264         } else {
1265                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1266                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1267                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1268                         range_whole = 1;
1269                 scanned = 1;
1270         }
1271 retry:
1272         while (!done && (index <= end) &&
1273                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1274                         PAGECACHE_TAG_DIRTY,
1275                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1276                 int first;
1277                 unsigned int i;
1278
1279                 first = -1;
1280                 next = 0;
1281                 n_iov = 0;
1282                 bytes_to_write = 0;
1283
1284                 for (i = 0; i < nr_pages; i++) {
1285                         page = pvec.pages[i];
1286                         /*
1287                          * At this point we hold neither mapping->tree_lock nor
1288                          * lock on the page itself: the page may be truncated or
1289                          * invalidated (changing page->mapping to NULL), or even
1290                          * swizzled back from swapper_space to tmpfs file
1291                          * mapping
1292                          */
1293
1294                         if (first < 0)
1295                                 lock_page(page);
1296                         else if (!trylock_page(page))
1297                                 break;
1298
1299                         if (unlikely(page->mapping != mapping)) {
1300                                 unlock_page(page);
1301                                 break;
1302                         }
1303
1304                         if (!wbc->range_cyclic && page->index > end) {
1305                                 done = 1;
1306                                 unlock_page(page);
1307                                 break;
1308                         }
1309
1310                         if (next && (page->index != next)) {
1311                                 /* Not next consecutive page */
1312                                 unlock_page(page);
1313                                 break;
1314                         }
1315
1316                         if (wbc->sync_mode != WB_SYNC_NONE)
1317                                 wait_on_page_writeback(page);
1318
1319                         if (PageWriteback(page) ||
1320                                         !clear_page_dirty_for_io(page)) {
1321                                 unlock_page(page);
1322                                 break;
1323                         }
1324
1325                         /*
1326                          * This actually clears the dirty bit in the radix tree.
1327                          * See cifs_writepage() for more commentary.
1328                          */
1329                         set_page_writeback(page);
1330
1331                         if (page_offset(page) >= mapping->host->i_size) {
1332                                 done = 1;
1333                                 unlock_page(page);
1334                                 end_page_writeback(page);
1335                                 break;
1336                         }
1337
1338                         /*
1339                          * BB can we get rid of this?  pages are held by pvec
1340                          */
1341                         page_cache_get(page);
1342
1343                         len = min(mapping->host->i_size - page_offset(page),
1344                                   (loff_t)PAGE_CACHE_SIZE);
1345
1346                         /* reserve iov[0] for the smb header */
1347                         n_iov++;
1348                         iov[n_iov].iov_base = kmap(page);
1349                         iov[n_iov].iov_len = len;
1350                         bytes_to_write += len;
1351
1352                         if (first < 0) {
1353                                 first = i;
1354                                 offset = page_offset(page);
1355                         }
1356                         next = page->index + 1;
1357                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1358                                 break;
1359                 }
1360                 if (n_iov) {
1361                         /* Search for a writable handle every time we call
1362                          * CIFSSMBWrite2.  We can't rely on the last handle
1363                          * we used to still be valid
1364                          */
1365                         open_file = find_writable_file(CIFS_I(mapping->host));
1366                         if (!open_file) {
1367                                 cERROR(1, ("No writable handles for inode"));
1368                                 rc = -EBADF;
1369                         } else {
1370                                 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1371                                                    open_file->netfid,
1372                                                    bytes_to_write, offset,
1373                                                    &bytes_written, iov, n_iov,
1374                                                    CIFS_LONG_OP);
1375                                 atomic_dec(&open_file->wrtPending);
1376                                 if (rc || bytes_written < bytes_to_write) {
1377                                         cERROR(1, ("Write2 ret %d, wrote %d",
1378                                                   rc, bytes_written));
1379                                         /* BB what if continued retry is
1380                                            requested via mount flags? */
1381                                         if (rc == -ENOSPC)
1382                                                 set_bit(AS_ENOSPC, &mapping->flags);
1383                                         else
1384                                                 set_bit(AS_EIO, &mapping->flags);
1385                                 } else {
1386                                         cifs_stats_bytes_written(cifs_sb->tcon,
1387                                                                  bytes_written);
1388                                 }
1389                         }
1390                         for (i = 0; i < n_iov; i++) {
1391                                 page = pvec.pages[first + i];
1392                                 /* Should we also set page error on
1393                                 success rc but too little data written? */
1394                                 /* BB investigate retry logic on temporary
1395                                 server crash cases and how recovery works
1396                                 when page marked as error */
1397                                 if (rc)
1398                                         SetPageError(page);
1399                                 kunmap(page);
1400                                 unlock_page(page);
1401                                 end_page_writeback(page);
1402                                 page_cache_release(page);
1403                         }
1404                         if ((wbc->nr_to_write -= n_iov) <= 0)
1405                                 done = 1;
1406                         index = next;
1407                 } else
1408                         /* Need to re-find the pages we skipped */
1409                         index = pvec.pages[0]->index + 1;
1410
1411                 pagevec_release(&pvec);
1412         }
1413         if (!scanned && !done) {
1414                 /*
1415                  * We hit the last page and there is more work to be done: wrap
1416                  * back to the start of the file
1417                  */
1418                 scanned = 1;
1419                 index = 0;
1420                 goto retry;
1421         }
1422         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1423                 mapping->writeback_index = index;
1424
1425         FreeXid(xid);
1426         kfree(iov);
1427         return rc;
1428 }
1429
1430 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1431 {
1432         int rc = -EFAULT;
1433         int xid;
1434
1435         xid = GetXid();
1436 /* BB add check for wbc flags */
1437         page_cache_get(page);
1438         if (!PageUptodate(page))
1439                 cFYI(1, ("ppw - page not up to date"));
1440
1441         /*
1442          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1443          *
1444          * A writepage() implementation always needs to do either this,
1445          * or re-dirty the page with "redirty_page_for_writepage()" in
1446          * the case of a failure.
1447          *
1448          * Just unlocking the page will cause the radix tree tag-bits
1449          * to fail to update with the state of the page correctly.
1450          */
1451         set_page_writeback(page);
1452         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1453         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1454         unlock_page(page);
1455         end_page_writeback(page);
1456         page_cache_release(page);
1457         FreeXid(xid);
1458         return rc;
1459 }
1460
1461 static int cifs_write_end(struct file *file, struct address_space *mapping,
1462                         loff_t pos, unsigned len, unsigned copied,
1463                         struct page *page, void *fsdata)
1464 {
1465         int rc;
1466         struct inode *inode = mapping->host;
1467
1468         cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1469                  page, pos, copied));
1470
1471         if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1472                 SetPageUptodate(page);
1473
1474         if (!PageUptodate(page)) {
1475                 char *page_data;
1476                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1477                 int xid;
1478
1479                 xid = GetXid();
1480                 /* this is probably better than directly calling
1481                    partialpage_write since in this function the file handle is
1482                    known which we might as well leverage */
1483                 /* BB check if anything else missing out of ppw
1484                    such as updating last write time */
1485                 page_data = kmap(page);
1486                 rc = cifs_write(file, page_data + offset, copied, &pos);
1487                 /* if (rc < 0) should we set writebehind rc? */
1488                 kunmap(page);
1489
1490                 FreeXid(xid);
1491         } else {
1492                 rc = copied;
1493                 pos += copied;
1494                 set_page_dirty(page);
1495         }
1496
1497         if (rc > 0) {
1498                 spin_lock(&inode->i_lock);
1499                 if (pos > inode->i_size)
1500                         i_size_write(inode, pos);
1501                 spin_unlock(&inode->i_lock);
1502         }
1503
1504         unlock_page(page);
1505         page_cache_release(page);
1506
1507         return rc;
1508 }
1509
1510 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1511 {
1512         int xid;
1513         int rc = 0;
1514         struct inode *inode = file->f_path.dentry->d_inode;
1515
1516         xid = GetXid();
1517
1518         cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1519                 dentry->d_name.name, datasync));
1520
1521         rc = filemap_write_and_wait(inode->i_mapping);
1522         if (rc == 0) {
1523                 rc = CIFS_I(inode)->write_behind_rc;
1524                 CIFS_I(inode)->write_behind_rc = 0;
1525         }
1526         FreeXid(xid);
1527         return rc;
1528 }
1529
1530 /* static void cifs_sync_page(struct page *page)
1531 {
1532         struct address_space *mapping;
1533         struct inode *inode;
1534         unsigned long index = page->index;
1535         unsigned int rpages = 0;
1536         int rc = 0;
1537
1538         cFYI(1, ("sync page %p",page));
1539         mapping = page->mapping;
1540         if (!mapping)
1541                 return 0;
1542         inode = mapping->host;
1543         if (!inode)
1544                 return; */
1545
1546 /*      fill in rpages then
1547         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1548
1549 /*      cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1550
1551 #if 0
1552         if (rc < 0)
1553                 return rc;
1554         return 0;
1555 #endif
1556 } */
1557
1558 /*
1559  * As file closes, flush all cached write data for this inode checking
1560  * for write behind errors.
1561  */
1562 int cifs_flush(struct file *file, fl_owner_t id)
1563 {
1564         struct inode *inode = file->f_path.dentry->d_inode;
1565         int rc = 0;
1566
1567         /* Rather than do the steps manually:
1568            lock the inode for writing
1569            loop through pages looking for write behind data (dirty pages)
1570            coalesce into contiguous 16K (or smaller) chunks to write to server
1571            send to server (prefer in parallel)
1572            deal with writebehind errors
1573            unlock inode for writing
1574            filemapfdatawrite appears easier for the time being */
1575
1576         rc = filemap_fdatawrite(inode->i_mapping);
1577         /* reset wb rc if we were able to write out dirty pages */
1578         if (!rc) {
1579                 rc = CIFS_I(inode)->write_behind_rc;
1580                 CIFS_I(inode)->write_behind_rc = 0;
1581         }
1582
1583         cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1584
1585         return rc;
1586 }
1587
1588 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1589         size_t read_size, loff_t *poffset)
1590 {
1591         int rc = -EACCES;
1592         unsigned int bytes_read = 0;
1593         unsigned int total_read = 0;
1594         unsigned int current_read_size;
1595         struct cifs_sb_info *cifs_sb;
1596         struct cifsTconInfo *pTcon;
1597         int xid;
1598         struct cifsFileInfo *open_file;
1599         char *smb_read_data;
1600         char __user *current_offset;
1601         struct smb_com_read_rsp *pSMBr;
1602
1603         xid = GetXid();
1604         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1605         pTcon = cifs_sb->tcon;
1606
1607         if (file->private_data == NULL) {
1608                 FreeXid(xid);
1609                 return -EBADF;
1610         }
1611         open_file = (struct cifsFileInfo *)file->private_data;
1612
1613         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1614                 cFYI(1, ("attempting read on write only file instance"));
1615
1616         for (total_read = 0, current_offset = read_data;
1617              read_size > total_read;
1618              total_read += bytes_read, current_offset += bytes_read) {
1619                 current_read_size = min_t(const int, read_size - total_read,
1620                                           cifs_sb->rsize);
1621                 rc = -EAGAIN;
1622                 smb_read_data = NULL;
1623                 while (rc == -EAGAIN) {
1624                         int buf_type = CIFS_NO_BUFFER;
1625                         if ((open_file->invalidHandle) &&
1626                             (!open_file->closePend)) {
1627                                 rc = cifs_reopen_file(file, true);
1628                                 if (rc != 0)
1629                                         break;
1630                         }
1631                         rc = CIFSSMBRead(xid, pTcon,
1632                                          open_file->netfid,
1633                                          current_read_size, *poffset,
1634                                          &bytes_read, &smb_read_data,
1635                                          &buf_type);
1636                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1637                         if (smb_read_data) {
1638                                 if (copy_to_user(current_offset,
1639                                                 smb_read_data +
1640                                                 4 /* RFC1001 length field */ +
1641                                                 le16_to_cpu(pSMBr->DataOffset),
1642                                                 bytes_read))
1643                                         rc = -EFAULT;
1644
1645                                 if (buf_type == CIFS_SMALL_BUFFER)
1646                                         cifs_small_buf_release(smb_read_data);
1647                                 else if (buf_type == CIFS_LARGE_BUFFER)
1648                                         cifs_buf_release(smb_read_data);
1649                                 smb_read_data = NULL;
1650                         }
1651                 }
1652                 if (rc || (bytes_read == 0)) {
1653                         if (total_read) {
1654                                 break;
1655                         } else {
1656                                 FreeXid(xid);
1657                                 return rc;
1658                         }
1659                 } else {
1660                         cifs_stats_bytes_read(pTcon, bytes_read);
1661                         *poffset += bytes_read;
1662                 }
1663         }
1664         FreeXid(xid);
1665         return total_read;
1666 }
1667
1668
1669 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1670         loff_t *poffset)
1671 {
1672         int rc = -EACCES;
1673         unsigned int bytes_read = 0;
1674         unsigned int total_read;
1675         unsigned int current_read_size;
1676         struct cifs_sb_info *cifs_sb;
1677         struct cifsTconInfo *pTcon;
1678         int xid;
1679         char *current_offset;
1680         struct cifsFileInfo *open_file;
1681         int buf_type = CIFS_NO_BUFFER;
1682
1683         xid = GetXid();
1684         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1685         pTcon = cifs_sb->tcon;
1686
1687         if (file->private_data == NULL) {
1688                 FreeXid(xid);
1689                 return -EBADF;
1690         }
1691         open_file = (struct cifsFileInfo *)file->private_data;
1692
1693         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1694                 cFYI(1, ("attempting read on write only file instance"));
1695
1696         for (total_read = 0, current_offset = read_data;
1697              read_size > total_read;
1698              total_read += bytes_read, current_offset += bytes_read) {
1699                 current_read_size = min_t(const int, read_size - total_read,
1700                                           cifs_sb->rsize);
1701                 /* For windows me and 9x we do not want to request more
1702                 than it negotiated since it will refuse the read then */
1703                 if ((pTcon->ses) &&
1704                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1705                         current_read_size = min_t(const int, current_read_size,
1706                                         pTcon->ses->server->maxBuf - 128);
1707                 }
1708                 rc = -EAGAIN;
1709                 while (rc == -EAGAIN) {
1710                         if ((open_file->invalidHandle) &&
1711                             (!open_file->closePend)) {
1712                                 rc = cifs_reopen_file(file, true);
1713                                 if (rc != 0)
1714                                         break;
1715                         }
1716                         rc = CIFSSMBRead(xid, pTcon,
1717                                          open_file->netfid,
1718                                          current_read_size, *poffset,
1719                                          &bytes_read, &current_offset,
1720                                          &buf_type);
1721                 }
1722                 if (rc || (bytes_read == 0)) {
1723                         if (total_read) {
1724                                 break;
1725                         } else {
1726                                 FreeXid(xid);
1727                                 return rc;
1728                         }
1729                 } else {
1730                         cifs_stats_bytes_read(pTcon, total_read);
1731                         *poffset += bytes_read;
1732                 }
1733         }
1734         FreeXid(xid);
1735         return total_read;
1736 }
1737
1738 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1739 {
1740         struct dentry *dentry = file->f_path.dentry;
1741         int rc, xid;
1742
1743         xid = GetXid();
1744         rc = cifs_revalidate(dentry);
1745         if (rc) {
1746                 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1747                 FreeXid(xid);
1748                 return rc;
1749         }
1750         rc = generic_file_mmap(file, vma);
1751         FreeXid(xid);
1752         return rc;
1753 }
1754
1755
1756 static void cifs_copy_cache_pages(struct address_space *mapping,
1757         struct list_head *pages, int bytes_read, char *data,
1758         struct pagevec *plru_pvec)
1759 {
1760         struct page *page;
1761         char *target;
1762
1763         while (bytes_read > 0) {
1764                 if (list_empty(pages))
1765                         break;
1766
1767                 page = list_entry(pages->prev, struct page, lru);
1768                 list_del(&page->lru);
1769
1770                 if (add_to_page_cache(page, mapping, page->index,
1771                                       GFP_KERNEL)) {
1772                         page_cache_release(page);
1773                         cFYI(1, ("Add page cache failed"));
1774                         data += PAGE_CACHE_SIZE;
1775                         bytes_read -= PAGE_CACHE_SIZE;
1776                         continue;
1777                 }
1778
1779                 target = kmap_atomic(page, KM_USER0);
1780
1781                 if (PAGE_CACHE_SIZE > bytes_read) {
1782                         memcpy(target, data, bytes_read);
1783                         /* zero the tail end of this partial page */
1784                         memset(target + bytes_read, 0,
1785                                PAGE_CACHE_SIZE - bytes_read);
1786                         bytes_read = 0;
1787                 } else {
1788                         memcpy(target, data, PAGE_CACHE_SIZE);
1789                         bytes_read -= PAGE_CACHE_SIZE;
1790                 }
1791                 kunmap_atomic(target, KM_USER0);
1792
1793                 flush_dcache_page(page);
1794                 SetPageUptodate(page);
1795                 unlock_page(page);
1796                 if (!pagevec_add(plru_pvec, page))
1797                         __pagevec_lru_add_file(plru_pvec);
1798                 data += PAGE_CACHE_SIZE;
1799         }
1800         return;
1801 }
1802
1803 static int cifs_readpages(struct file *file, struct address_space *mapping,
1804         struct list_head *page_list, unsigned num_pages)
1805 {
1806         int rc = -EACCES;
1807         int xid;
1808         loff_t offset;
1809         struct page *page;
1810         struct cifs_sb_info *cifs_sb;
1811         struct cifsTconInfo *pTcon;
1812         unsigned int bytes_read = 0;
1813         unsigned int read_size, i;
1814         char *smb_read_data = NULL;
1815         struct smb_com_read_rsp *pSMBr;
1816         struct pagevec lru_pvec;
1817         struct cifsFileInfo *open_file;
1818         int buf_type = CIFS_NO_BUFFER;
1819
1820         xid = GetXid();
1821         if (file->private_data == NULL) {
1822                 FreeXid(xid);
1823                 return -EBADF;
1824         }
1825         open_file = (struct cifsFileInfo *)file->private_data;
1826         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1827         pTcon = cifs_sb->tcon;
1828
1829         pagevec_init(&lru_pvec, 0);
1830         cFYI(DBG2, ("rpages: num pages %d", num_pages));
1831         for (i = 0; i < num_pages; ) {
1832                 unsigned contig_pages;
1833                 struct page *tmp_page;
1834                 unsigned long expected_index;
1835
1836                 if (list_empty(page_list))
1837                         break;
1838
1839                 page = list_entry(page_list->prev, struct page, lru);
1840                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1841
1842                 /* count adjacent pages that we will read into */
1843                 contig_pages = 0;
1844                 expected_index =
1845                         list_entry(page_list->prev, struct page, lru)->index;
1846                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1847                         if (tmp_page->index == expected_index) {
1848                                 contig_pages++;
1849                                 expected_index++;
1850                         } else
1851                                 break;
1852                 }
1853                 if (contig_pages + i >  num_pages)
1854                         contig_pages = num_pages - i;
1855
1856                 /* for reads over a certain size could initiate async
1857                    read ahead */
1858
1859                 read_size = contig_pages * PAGE_CACHE_SIZE;
1860                 /* Read size needs to be in multiples of one page */
1861                 read_size = min_t(const unsigned int, read_size,
1862                                   cifs_sb->rsize & PAGE_CACHE_MASK);
1863                 cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
1864                                 read_size, contig_pages));
1865                 rc = -EAGAIN;
1866                 while (rc == -EAGAIN) {
1867                         if ((open_file->invalidHandle) &&
1868                             (!open_file->closePend)) {
1869                                 rc = cifs_reopen_file(file, true);
1870                                 if (rc != 0)
1871                                         break;
1872                         }
1873
1874                         rc = CIFSSMBRead(xid, pTcon,
1875                                          open_file->netfid,
1876                                          read_size, offset,
1877                                          &bytes_read, &smb_read_data,
1878                                          &buf_type);
1879                         /* BB more RC checks ? */
1880                         if (rc == -EAGAIN) {
1881                                 if (smb_read_data) {
1882                                         if (buf_type == CIFS_SMALL_BUFFER)
1883                                                 cifs_small_buf_release(smb_read_data);
1884                                         else if (buf_type == CIFS_LARGE_BUFFER)
1885                                                 cifs_buf_release(smb_read_data);
1886                                         smb_read_data = NULL;
1887                                 }
1888                         }
1889                 }
1890                 if ((rc < 0) || (smb_read_data == NULL)) {
1891                         cFYI(1, ("Read error in readpages: %d", rc));
1892                         break;
1893                 } else if (bytes_read > 0) {
1894                         task_io_account_read(bytes_read);
1895                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1896                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
1897                                 smb_read_data + 4 /* RFC1001 hdr */ +
1898                                 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1899
1900                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
1901                         cifs_stats_bytes_read(pTcon, bytes_read);
1902                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1903                                 i++; /* account for partial page */
1904
1905                                 /* server copy of file can have smaller size
1906                                    than client */
1907                                 /* BB do we need to verify this common case ?
1908                                    this case is ok - if we are at server EOF
1909                                    we will hit it on next read */
1910
1911                                 /* break; */
1912                         }
1913                 } else {
1914                         cFYI(1, ("No bytes read (%d) at offset %lld . "
1915                                  "Cleaning remaining pages from readahead list",
1916                                  bytes_read, offset));
1917                         /* BB turn off caching and do new lookup on
1918                            file size at server? */
1919                         break;
1920                 }
1921                 if (smb_read_data) {
1922                         if (buf_type == CIFS_SMALL_BUFFER)
1923                                 cifs_small_buf_release(smb_read_data);
1924                         else if (buf_type == CIFS_LARGE_BUFFER)
1925                                 cifs_buf_release(smb_read_data);
1926                         smb_read_data = NULL;
1927                 }
1928                 bytes_read = 0;
1929         }
1930
1931         pagevec_lru_add_file(&lru_pvec);
1932
1933 /* need to free smb_read_data buf before exit */
1934         if (smb_read_data) {
1935                 if (buf_type == CIFS_SMALL_BUFFER)
1936                         cifs_small_buf_release(smb_read_data);
1937                 else if (buf_type == CIFS_LARGE_BUFFER)
1938                         cifs_buf_release(smb_read_data);
1939                 smb_read_data = NULL;
1940         }
1941
1942         FreeXid(xid);
1943         return rc;
1944 }
1945
1946 static int cifs_readpage_worker(struct file *file, struct page *page,
1947         loff_t *poffset)
1948 {
1949         char *read_data;
1950         int rc;
1951
1952         page_cache_get(page);
1953         read_data = kmap(page);
1954         /* for reads over a certain size could initiate async read ahead */
1955
1956         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1957
1958         if (rc < 0)
1959                 goto io_error;
1960         else
1961                 cFYI(1, ("Bytes read %d", rc));
1962
1963         file->f_path.dentry->d_inode->i_atime =
1964                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1965
1966         if (PAGE_CACHE_SIZE > rc)
1967                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1968
1969         flush_dcache_page(page);
1970         SetPageUptodate(page);
1971         rc = 0;
1972
1973 io_error:
1974         kunmap(page);
1975         page_cache_release(page);
1976         return rc;
1977 }
1978
1979 static int cifs_readpage(struct file *file, struct page *page)
1980 {
1981         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1982         int rc = -EACCES;
1983         int xid;
1984
1985         xid = GetXid();
1986
1987         if (file->private_data == NULL) {
1988                 FreeXid(xid);
1989                 return -EBADF;
1990         }
1991
1992         cFYI(1, ("readpage %p at offset %d 0x%x\n",
1993                  page, (int)offset, (int)offset));
1994
1995         rc = cifs_readpage_worker(file, page, &offset);
1996
1997         unlock_page(page);
1998
1999         FreeXid(xid);
2000         return rc;
2001 }
2002
2003 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2004 {
2005         struct cifsFileInfo *open_file;
2006
2007         read_lock(&GlobalSMBSeslock);
2008         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2009                 if (open_file->closePend)
2010                         continue;
2011                 if (open_file->pfile &&
2012                     ((open_file->pfile->f_flags & O_RDWR) ||
2013                      (open_file->pfile->f_flags & O_WRONLY))) {
2014                         read_unlock(&GlobalSMBSeslock);
2015                         return 1;
2016                 }
2017         }
2018         read_unlock(&GlobalSMBSeslock);
2019         return 0;
2020 }
2021
2022 /* We do not want to update the file size from server for inodes
2023    open for write - to avoid races with writepage extending
2024    the file - in the future we could consider allowing
2025    refreshing the inode only on increases in the file size
2026    but this is tricky to do without racing with writebehind
2027    page caching in the current Linux kernel design */
2028 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2029 {
2030         if (!cifsInode)
2031                 return true;
2032
2033         if (is_inode_writable(cifsInode)) {
2034                 /* This inode is open for write at least once */
2035                 struct cifs_sb_info *cifs_sb;
2036
2037                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2038                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2039                         /* since no page cache to corrupt on directio
2040                         we can change size safely */
2041                         return true;
2042                 }
2043
2044                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2045                         return true;
2046
2047                 return false;
2048         } else
2049                 return true;
2050 }
2051
2052 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2053                         loff_t pos, unsigned len, unsigned flags,
2054                         struct page **pagep, void **fsdata)
2055 {
2056         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2057         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2058
2059         cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2060
2061         *pagep = __grab_cache_page(mapping, index);
2062         if (!*pagep)
2063                 return -ENOMEM;
2064
2065         if (PageUptodate(*pagep))
2066                 return 0;
2067
2068         /* If we are writing a full page it will be up to date,
2069            no need to read from the server */
2070         if (len == PAGE_CACHE_SIZE && flags & AOP_FLAG_UNINTERRUPTIBLE)
2071                 return 0;
2072
2073         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2074                 int rc;
2075
2076                 /* might as well read a page, it is fast enough */
2077                 rc = cifs_readpage_worker(file, *pagep, &offset);
2078
2079                 /* we do not need to pass errors back
2080                    e.g. if we do not have read access to the file
2081                    because cifs_write_end will attempt synchronous writes
2082                    -- shaggy */
2083         } else {
2084                 /* we could try using another file handle if there is one -
2085                    but how would we lock it to prevent close of that handle
2086                    racing with this read? In any case
2087                    this will be written out by write_end so is fine */
2088         }
2089
2090         return 0;
2091 }
2092
2093 const struct address_space_operations cifs_addr_ops = {
2094         .readpage = cifs_readpage,
2095         .readpages = cifs_readpages,
2096         .writepage = cifs_writepage,
2097         .writepages = cifs_writepages,
2098         .write_begin = cifs_write_begin,
2099         .write_end = cifs_write_end,
2100         .set_page_dirty = __set_page_dirty_nobuffers,
2101         /* .sync_page = cifs_sync_page, */
2102         /* .direct_IO = */
2103 };
2104
2105 /*
2106  * cifs_readpages requires the server to support a buffer large enough to
2107  * contain the header plus one complete page of data.  Otherwise, we need
2108  * to leave cifs_readpages out of the address space operations.
2109  */
2110 const struct address_space_operations cifs_addr_ops_smallbuf = {
2111         .readpage = cifs_readpage,
2112         .writepage = cifs_writepage,
2113         .writepages = cifs_writepages,
2114         .write_begin = cifs_write_begin,
2115         .write_end = cifs_write_end,
2116         .set_page_dirty = __set_page_dirty_nobuffers,
2117         /* .sync_page = cifs_sync_page, */
2118         /* .direct_IO = */
2119 };