Merge commit 'v2.6.28-rc9' into x86/apic
[linux-2.6] / fs / cifs / file.c
1 /*
2  *   fs/cifs/file.c
3  *
4  *   vfs operations that deal with files
5  *
6  *   Copyright (C) International Business Machines  Corp., 2002,2007
7  *   Author(s): Steve French (sfrench@us.ibm.com)
8  *              Jeremy Allison (jra@samba.org)
9  *
10  *   This library is free software; you can redistribute it and/or modify
11  *   it under the terms of the GNU Lesser General Public License as published
12  *   by the Free Software Foundation; either version 2.1 of the License, or
13  *   (at your option) any later version.
14  *
15  *   This library is distributed in the hope that it will be useful,
16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
18  *   the GNU Lesser General Public License for more details.
19  *
20  *   You should have received a copy of the GNU Lesser General Public License
21  *   along with this library; if not, write to the Free Software
22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23  */
24 #include <linux/fs.h>
25 #include <linux/backing-dev.h>
26 #include <linux/stat.h>
27 #include <linux/fcntl.h>
28 #include <linux/pagemap.h>
29 #include <linux/pagevec.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/delay.h>
33 #include <asm/div64.h>
34 #include "cifsfs.h"
35 #include "cifspdu.h"
36 #include "cifsglob.h"
37 #include "cifsproto.h"
38 #include "cifs_unicode.h"
39 #include "cifs_debug.h"
40 #include "cifs_fs_sb.h"
41
42 static inline struct cifsFileInfo *cifs_init_private(
43         struct cifsFileInfo *private_data, struct inode *inode,
44         struct file *file, __u16 netfid)
45 {
46         memset(private_data, 0, sizeof(struct cifsFileInfo));
47         private_data->netfid = netfid;
48         private_data->pid = current->tgid;
49         init_MUTEX(&private_data->fh_sem);
50         mutex_init(&private_data->lock_mutex);
51         INIT_LIST_HEAD(&private_data->llist);
52         private_data->pfile = file; /* needed for writepage */
53         private_data->pInode = inode;
54         private_data->invalidHandle = false;
55         private_data->closePend = false;
56         /* we have to track num writers to the inode, since writepages
57         does not tell us which handle the write is for so there can
58         be a close (overlapping with write) of the filehandle that
59         cifs_writepages chose to use */
60         atomic_set(&private_data->wrtPending, 0);
61
62         return private_data;
63 }
64
65 static inline int cifs_convert_flags(unsigned int flags)
66 {
67         if ((flags & O_ACCMODE) == O_RDONLY)
68                 return GENERIC_READ;
69         else if ((flags & O_ACCMODE) == O_WRONLY)
70                 return GENERIC_WRITE;
71         else if ((flags & O_ACCMODE) == O_RDWR) {
72                 /* GENERIC_ALL is too much permission to request
73                    can cause unnecessary access denied on create */
74                 /* return GENERIC_ALL; */
75                 return (GENERIC_READ | GENERIC_WRITE);
76         }
77
78         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
79                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
80                 FILE_READ_DATA);
81
82
83 }
84
85 static inline int cifs_get_disposition(unsigned int flags)
86 {
87         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
88                 return FILE_CREATE;
89         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
90                 return FILE_OVERWRITE_IF;
91         else if ((flags & O_CREAT) == O_CREAT)
92                 return FILE_OPEN_IF;
93         else if ((flags & O_TRUNC) == O_TRUNC)
94                 return FILE_OVERWRITE;
95         else
96                 return FILE_OPEN;
97 }
98
99 /* all arguments to this function must be checked for validity in caller */
100 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
101         struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
102         struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
103         char *full_path, int xid)
104 {
105         struct timespec temp;
106         int rc;
107
108         /* want handles we can use to read with first
109            in the list so we do not have to walk the
110            list to search for one in 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                 write_lock(&GlobalSMBSeslock);
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                                 write_unlock(&GlobalSMBSeslock);
498                                 timeout = 2;
499                                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
500                                         && (timeout <= 2048)) {
501                                         /* Give write a better chance to get to
502                                         server ahead of the close.  We do not
503                                         want to add a wait_q here as it would
504                                         increase the memory utilization as
505                                         the struct would be in each open file,
506                                         but this should give enough time to
507                                         clear the socket */
508                                         cFYI(DBG2,
509                                                 ("close delay, write pending"));
510                                         msleep(timeout);
511                                         timeout *= 4;
512                                 }
513                                 if (atomic_read(&pSMBFile->wrtPending))
514                                         cERROR(1, ("close with pending write"));
515                                 if (!pTcon->need_reconnect &&
516                                     !pSMBFile->invalidHandle)
517                                         rc = CIFSSMBClose(xid, pTcon,
518                                                   pSMBFile->netfid);
519                         } else
520                                 write_unlock(&GlobalSMBSeslock);
521                 } else
522                         write_unlock(&GlobalSMBSeslock);
523
524                 /* Delete any outstanding lock records.
525                    We'll lose them when the file is closed anyway. */
526                 mutex_lock(&pSMBFile->lock_mutex);
527                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
528                         list_del(&li->llist);
529                         kfree(li);
530                 }
531                 mutex_unlock(&pSMBFile->lock_mutex);
532
533                 write_lock(&GlobalSMBSeslock);
534                 list_del(&pSMBFile->flist);
535                 list_del(&pSMBFile->tlist);
536                 write_unlock(&GlobalSMBSeslock);
537                 timeout = 10;
538                 /* We waited above to give the SMBWrite a chance to issue
539                    on the wire (so we do not get SMBWrite returning EBADF
540                    if writepages is racing with close.  Note that writepages
541                    does not specify a file handle, so it is possible for a file
542                    to be opened twice, and the application close the "wrong"
543                    file handle - in these cases we delay long enough to allow
544                    the SMBWrite to get on the wire before the SMB Close.
545                    We allow total wait here over 45 seconds, more than
546                    oplock break time, and more than enough to allow any write
547                    to complete on the server, or to time out on the client */
548                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
549                                 && (timeout <= 50000)) {
550                         cERROR(1, ("writes pending, delay free of handle"));
551                         msleep(timeout);
552                         timeout *= 8;
553                 }
554                 kfree(file->private_data);
555                 file->private_data = NULL;
556         } else
557                 rc = -EBADF;
558
559         read_lock(&GlobalSMBSeslock);
560         if (list_empty(&(CIFS_I(inode)->openFileList))) {
561                 cFYI(1, ("closing last open instance for inode %p", inode));
562                 /* if the file is not open we do not know if we can cache info
563                    on this inode, much less write behind and read ahead */
564                 CIFS_I(inode)->clientCanCacheRead = false;
565                 CIFS_I(inode)->clientCanCacheAll  = false;
566         }
567         read_unlock(&GlobalSMBSeslock);
568         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
569                 rc = CIFS_I(inode)->write_behind_rc;
570         FreeXid(xid);
571         return rc;
572 }
573
574 int cifs_closedir(struct inode *inode, struct file *file)
575 {
576         int rc = 0;
577         int xid;
578         struct cifsFileInfo *pCFileStruct =
579             (struct cifsFileInfo *)file->private_data;
580         char *ptmp;
581
582         cFYI(1, ("Closedir inode = 0x%p", inode));
583
584         xid = GetXid();
585
586         if (pCFileStruct) {
587                 struct cifsTconInfo *pTcon;
588                 struct cifs_sb_info *cifs_sb =
589                         CIFS_SB(file->f_path.dentry->d_sb);
590
591                 pTcon = cifs_sb->tcon;
592
593                 cFYI(1, ("Freeing private data in close dir"));
594                 write_lock(&GlobalSMBSeslock);
595                 if (!pCFileStruct->srch_inf.endOfSearch &&
596                     !pCFileStruct->invalidHandle) {
597                         pCFileStruct->invalidHandle = true;
598                         write_unlock(&GlobalSMBSeslock);
599                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
600                         cFYI(1, ("Closing uncompleted readdir with rc %d",
601                                  rc));
602                         /* not much we can do if it fails anyway, ignore rc */
603                         rc = 0;
604                 } else
605                         write_unlock(&GlobalSMBSeslock);
606                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
607                 if (ptmp) {
608                         cFYI(1, ("closedir free smb buf in srch struct"));
609                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
610                         if (pCFileStruct->srch_inf.smallBuf)
611                                 cifs_small_buf_release(ptmp);
612                         else
613                                 cifs_buf_release(ptmp);
614                 }
615                 kfree(file->private_data);
616                 file->private_data = NULL;
617         }
618         /* BB can we lock the filestruct while this is going on? */
619         FreeXid(xid);
620         return rc;
621 }
622
623 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
624                                 __u64 offset, __u8 lockType)
625 {
626         struct cifsLockInfo *li =
627                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
628         if (li == NULL)
629                 return -ENOMEM;
630         li->offset = offset;
631         li->length = len;
632         li->type = lockType;
633         mutex_lock(&fid->lock_mutex);
634         list_add(&li->llist, &fid->llist);
635         mutex_unlock(&fid->lock_mutex);
636         return 0;
637 }
638
639 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
640 {
641         int rc, xid;
642         __u32 numLock = 0;
643         __u32 numUnlock = 0;
644         __u64 length;
645         bool wait_flag = false;
646         struct cifs_sb_info *cifs_sb;
647         struct cifsTconInfo *pTcon;
648         __u16 netfid;
649         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
650         bool posix_locking;
651
652         length = 1 + pfLock->fl_end - pfLock->fl_start;
653         rc = -EACCES;
654         xid = GetXid();
655
656         cFYI(1, ("Lock parm: 0x%x flockflags: "
657                  "0x%x flocktype: 0x%x start: %lld end: %lld",
658                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
659                 pfLock->fl_end));
660
661         if (pfLock->fl_flags & FL_POSIX)
662                 cFYI(1, ("Posix"));
663         if (pfLock->fl_flags & FL_FLOCK)
664                 cFYI(1, ("Flock"));
665         if (pfLock->fl_flags & FL_SLEEP) {
666                 cFYI(1, ("Blocking lock"));
667                 wait_flag = true;
668         }
669         if (pfLock->fl_flags & FL_ACCESS)
670                 cFYI(1, ("Process suspended by mandatory locking - "
671                          "not implemented yet"));
672         if (pfLock->fl_flags & FL_LEASE)
673                 cFYI(1, ("Lease on file - not implemented yet"));
674         if (pfLock->fl_flags &
675             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
676                 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
677
678         if (pfLock->fl_type == F_WRLCK) {
679                 cFYI(1, ("F_WRLCK "));
680                 numLock = 1;
681         } else if (pfLock->fl_type == F_UNLCK) {
682                 cFYI(1, ("F_UNLCK"));
683                 numUnlock = 1;
684                 /* Check if unlock includes more than
685                 one lock range */
686         } else if (pfLock->fl_type == F_RDLCK) {
687                 cFYI(1, ("F_RDLCK"));
688                 lockType |= LOCKING_ANDX_SHARED_LOCK;
689                 numLock = 1;
690         } else if (pfLock->fl_type == F_EXLCK) {
691                 cFYI(1, ("F_EXLCK"));
692                 numLock = 1;
693         } else if (pfLock->fl_type == F_SHLCK) {
694                 cFYI(1, ("F_SHLCK"));
695                 lockType |= LOCKING_ANDX_SHARED_LOCK;
696                 numLock = 1;
697         } else
698                 cFYI(1, ("Unknown type of lock"));
699
700         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
701         pTcon = cifs_sb->tcon;
702
703         if (file->private_data == NULL) {
704                 FreeXid(xid);
705                 return -EBADF;
706         }
707         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
708
709         posix_locking = (cifs_sb->tcon->ses->capabilities & CAP_UNIX) &&
710                         (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(cifs_sb->tcon->fsUnixInfo.Capability));
711
712         /* BB add code here to normalize offset and length to
713         account for negative length which we can not accept over the
714         wire */
715         if (IS_GETLK(cmd)) {
716                 if (posix_locking) {
717                         int posix_lock_type;
718                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
719                                 posix_lock_type = CIFS_RDLCK;
720                         else
721                                 posix_lock_type = CIFS_WRLCK;
722                         rc = CIFSSMBPosixLock(xid, pTcon, netfid, 1 /* get */,
723                                         length, pfLock,
724                                         posix_lock_type, wait_flag);
725                         FreeXid(xid);
726                         return rc;
727                 }
728
729                 /* BB we could chain these into one lock request BB */
730                 rc = CIFSSMBLock(xid, pTcon, netfid, length, pfLock->fl_start,
731                                  0, 1, lockType, 0 /* wait flag */ );
732                 if (rc == 0) {
733                         rc = CIFSSMBLock(xid, pTcon, netfid, length,
734                                          pfLock->fl_start, 1 /* numUnlock */ ,
735                                          0 /* numLock */ , lockType,
736                                          0 /* wait flag */ );
737                         pfLock->fl_type = F_UNLCK;
738                         if (rc != 0)
739                                 cERROR(1, ("Error unlocking previously locked "
740                                            "range %d during test of lock", rc));
741                         rc = 0;
742
743                 } else {
744                         /* if rc == ERR_SHARING_VIOLATION ? */
745                         rc = 0; /* do not change lock type to unlock
746                                    since range in use */
747                 }
748
749                 FreeXid(xid);
750                 return rc;
751         }
752
753         if (!numLock && !numUnlock) {
754                 /* if no lock or unlock then nothing
755                 to do since we do not know what it is */
756                 FreeXid(xid);
757                 return -EOPNOTSUPP;
758         }
759
760         if (posix_locking) {
761                 int posix_lock_type;
762                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
763                         posix_lock_type = CIFS_RDLCK;
764                 else
765                         posix_lock_type = CIFS_WRLCK;
766
767                 if (numUnlock == 1)
768                         posix_lock_type = CIFS_UNLCK;
769
770                 rc = CIFSSMBPosixLock(xid, pTcon, netfid, 0 /* set */,
771                                       length, pfLock,
772                                       posix_lock_type, wait_flag);
773         } else {
774                 struct cifsFileInfo *fid =
775                         (struct cifsFileInfo *)file->private_data;
776
777                 if (numLock) {
778                         rc = CIFSSMBLock(xid, pTcon, netfid, length,
779                                         pfLock->fl_start,
780                                         0, numLock, lockType, wait_flag);
781
782                         if (rc == 0) {
783                                 /* For Windows locks we must store them. */
784                                 rc = store_file_lock(fid, length,
785                                                 pfLock->fl_start, lockType);
786                         }
787                 } else if (numUnlock) {
788                         /* For each stored lock that this unlock overlaps
789                            completely, unlock it. */
790                         int stored_rc = 0;
791                         struct cifsLockInfo *li, *tmp;
792
793                         rc = 0;
794                         mutex_lock(&fid->lock_mutex);
795                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
796                                 if (pfLock->fl_start <= li->offset &&
797                                                 (pfLock->fl_start + length) >=
798                                                 (li->offset + li->length)) {
799                                         stored_rc = CIFSSMBLock(xid, pTcon,
800                                                         netfid,
801                                                         li->length, li->offset,
802                                                         1, 0, li->type, false);
803                                         if (stored_rc)
804                                                 rc = stored_rc;
805
806                                         list_del(&li->llist);
807                                         kfree(li);
808                                 }
809                         }
810                         mutex_unlock(&fid->lock_mutex);
811                 }
812         }
813
814         if (pfLock->fl_flags & FL_POSIX)
815                 posix_lock_file_wait(file, pfLock);
816         FreeXid(xid);
817         return rc;
818 }
819
820 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
821         size_t write_size, loff_t *poffset)
822 {
823         int rc = 0;
824         unsigned int bytes_written = 0;
825         unsigned int total_written;
826         struct cifs_sb_info *cifs_sb;
827         struct cifsTconInfo *pTcon;
828         int xid, long_op;
829         struct cifsFileInfo *open_file;
830
831         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
832
833         pTcon = cifs_sb->tcon;
834
835         /* cFYI(1,
836            (" write %d bytes to offset %lld of %s", write_size,
837            *poffset, file->f_path.dentry->d_name.name)); */
838
839         if (file->private_data == NULL)
840                 return -EBADF;
841         open_file = (struct cifsFileInfo *) file->private_data;
842
843         rc = generic_write_checks(file, poffset, &write_size, 0);
844         if (rc)
845                 return rc;
846
847         xid = GetXid();
848
849         if (*poffset > file->f_path.dentry->d_inode->i_size)
850                 long_op = CIFS_VLONG_OP; /* writes past EOF take long time */
851         else
852                 long_op = CIFS_LONG_OP;
853
854         for (total_written = 0; write_size > total_written;
855              total_written += bytes_written) {
856                 rc = -EAGAIN;
857                 while (rc == -EAGAIN) {
858                         if (file->private_data == NULL) {
859                                 /* file has been closed on us */
860                                 FreeXid(xid);
861                         /* if we have gotten here we have written some data
862                            and blocked, and the file has been freed on us while
863                            we blocked so return what we managed to write */
864                                 return total_written;
865                         }
866                         if (open_file->closePend) {
867                                 FreeXid(xid);
868                                 if (total_written)
869                                         return total_written;
870                                 else
871                                         return -EBADF;
872                         }
873                         if (open_file->invalidHandle) {
874                                 /* we could deadlock if we called
875                                    filemap_fdatawait from here so tell
876                                    reopen_file not to flush data to server
877                                    now */
878                                 rc = cifs_reopen_file(file, false);
879                                 if (rc != 0)
880                                         break;
881                         }
882
883                         rc = CIFSSMBWrite(xid, pTcon,
884                                 open_file->netfid,
885                                 min_t(const int, cifs_sb->wsize,
886                                       write_size - total_written),
887                                 *poffset, &bytes_written,
888                                 NULL, write_data + total_written, long_op);
889                 }
890                 if (rc || (bytes_written == 0)) {
891                         if (total_written)
892                                 break;
893                         else {
894                                 FreeXid(xid);
895                                 return rc;
896                         }
897                 } else
898                         *poffset += bytes_written;
899                 long_op = CIFS_STD_OP; /* subsequent writes fast -
900                                     15 seconds is plenty */
901         }
902
903         cifs_stats_bytes_written(pTcon, total_written);
904
905         /* since the write may have blocked check these pointers again */
906         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
907                 struct inode *inode = file->f_path.dentry->d_inode;
908 /* Do not update local mtime - server will set its actual value on write
909  *              inode->i_ctime = inode->i_mtime =
910  *                      current_fs_time(inode->i_sb);*/
911                 if (total_written > 0) {
912                         spin_lock(&inode->i_lock);
913                         if (*poffset > file->f_path.dentry->d_inode->i_size)
914                                 i_size_write(file->f_path.dentry->d_inode,
915                                         *poffset);
916                         spin_unlock(&inode->i_lock);
917                 }
918                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
919         }
920         FreeXid(xid);
921         return total_written;
922 }
923
924 static ssize_t cifs_write(struct file *file, const char *write_data,
925                           size_t write_size, loff_t *poffset)
926 {
927         int rc = 0;
928         unsigned int bytes_written = 0;
929         unsigned int total_written;
930         struct cifs_sb_info *cifs_sb;
931         struct cifsTconInfo *pTcon;
932         int xid, long_op;
933         struct cifsFileInfo *open_file;
934
935         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
936
937         pTcon = cifs_sb->tcon;
938
939         cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
940            *poffset, file->f_path.dentry->d_name.name));
941
942         if (file->private_data == NULL)
943                 return -EBADF;
944         open_file = (struct cifsFileInfo *)file->private_data;
945
946         xid = GetXid();
947
948         if (*poffset > file->f_path.dentry->d_inode->i_size)
949                 long_op = CIFS_VLONG_OP; /* writes past EOF can be slow */
950         else
951                 long_op = CIFS_LONG_OP;
952
953         for (total_written = 0; write_size > total_written;
954              total_written += bytes_written) {
955                 rc = -EAGAIN;
956                 while (rc == -EAGAIN) {
957                         if (file->private_data == NULL) {
958                                 /* file has been closed on us */
959                                 FreeXid(xid);
960                         /* if we have gotten here we have written some data
961                            and blocked, and the file has been freed on us
962                            while we blocked so return what we managed to
963                            write */
964                                 return total_written;
965                         }
966                         if (open_file->closePend) {
967                                 FreeXid(xid);
968                                 if (total_written)
969                                         return total_written;
970                                 else
971                                         return -EBADF;
972                         }
973                         if (open_file->invalidHandle) {
974                                 /* we could deadlock if we called
975                                    filemap_fdatawait from here so tell
976                                    reopen_file not to flush data to
977                                    server now */
978                                 rc = cifs_reopen_file(file, false);
979                                 if (rc != 0)
980                                         break;
981                         }
982                         if (experimEnabled || (pTcon->ses->server &&
983                                 ((pTcon->ses->server->secMode &
984                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
985                                 == 0))) {
986                                 struct kvec iov[2];
987                                 unsigned int len;
988
989                                 len = min((size_t)cifs_sb->wsize,
990                                           write_size - total_written);
991                                 /* iov[0] is reserved for smb header */
992                                 iov[1].iov_base = (char *)write_data +
993                                                   total_written;
994                                 iov[1].iov_len = len;
995                                 rc = CIFSSMBWrite2(xid, pTcon,
996                                                 open_file->netfid, len,
997                                                 *poffset, &bytes_written,
998                                                 iov, 1, long_op);
999                         } else
1000                                 rc = CIFSSMBWrite(xid, pTcon,
1001                                          open_file->netfid,
1002                                          min_t(const int, cifs_sb->wsize,
1003                                                write_size - total_written),
1004                                          *poffset, &bytes_written,
1005                                          write_data + total_written,
1006                                          NULL, long_op);
1007                 }
1008                 if (rc || (bytes_written == 0)) {
1009                         if (total_written)
1010                                 break;
1011                         else {
1012                                 FreeXid(xid);
1013                                 return rc;
1014                         }
1015                 } else
1016                         *poffset += bytes_written;
1017                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1018                                     15 seconds is plenty */
1019         }
1020
1021         cifs_stats_bytes_written(pTcon, total_written);
1022
1023         /* since the write may have blocked check these pointers again */
1024         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1025 /*BB We could make this contingent on superblock ATIME flag too */
1026 /*              file->f_path.dentry->d_inode->i_ctime =
1027                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1028                 if (total_written > 0) {
1029                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1030                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1031                                 i_size_write(file->f_path.dentry->d_inode,
1032                                              *poffset);
1033                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1034                 }
1035                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1036         }
1037         FreeXid(xid);
1038         return total_written;
1039 }
1040
1041 #ifdef CONFIG_CIFS_EXPERIMENTAL
1042 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1043 {
1044         struct cifsFileInfo *open_file = NULL;
1045
1046         read_lock(&GlobalSMBSeslock);
1047         /* we could simply get the first_list_entry since write-only entries
1048            are always at the end of the list but since the first entry might
1049            have a close pending, we go through the whole list */
1050         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1051                 if (open_file->closePend)
1052                         continue;
1053                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1054                     (open_file->pfile->f_flags & O_RDONLY))) {
1055                         if (!open_file->invalidHandle) {
1056                                 /* found a good file */
1057                                 /* lock it so it will not be closed on us */
1058                                 atomic_inc(&open_file->wrtPending);
1059                                 read_unlock(&GlobalSMBSeslock);
1060                                 return open_file;
1061                         } /* else might as well continue, and look for
1062                              another, or simply have the caller reopen it
1063                              again rather than trying to fix this handle */
1064                 } else /* write only file */
1065                         break; /* write only files are last so must be done */
1066         }
1067         read_unlock(&GlobalSMBSeslock);
1068         return NULL;
1069 }
1070 #endif
1071
1072 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1073 {
1074         struct cifsFileInfo *open_file;
1075         bool any_available = false;
1076         int rc;
1077
1078         /* Having a null inode here (because mapping->host was set to zero by
1079         the VFS or MM) should not happen but we had reports of on oops (due to
1080         it being zero) during stress testcases so we need to check for it */
1081
1082         if (cifs_inode == NULL) {
1083                 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1084                 dump_stack();
1085                 return NULL;
1086         }
1087
1088         read_lock(&GlobalSMBSeslock);
1089 refind_writable:
1090         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1091                 if (open_file->closePend ||
1092                     (!any_available && open_file->pid != current->tgid))
1093                         continue;
1094
1095                 if (open_file->pfile &&
1096                     ((open_file->pfile->f_flags & O_RDWR) ||
1097                      (open_file->pfile->f_flags & O_WRONLY))) {
1098                         atomic_inc(&open_file->wrtPending);
1099
1100                         if (!open_file->invalidHandle) {
1101                                 /* found a good writable file */
1102                                 read_unlock(&GlobalSMBSeslock);
1103                                 return open_file;
1104                         }
1105
1106                         read_unlock(&GlobalSMBSeslock);
1107                         /* Had to unlock since following call can block */
1108                         rc = cifs_reopen_file(open_file->pfile, false);
1109                         if (!rc) {
1110                                 if (!open_file->closePend)
1111                                         return open_file;
1112                                 else { /* start over in case this was deleted */
1113                                        /* since the list could be modified */
1114                                         read_lock(&GlobalSMBSeslock);
1115                                         atomic_dec(&open_file->wrtPending);
1116                                         goto refind_writable;
1117                                 }
1118                         }
1119
1120                         /* if it fails, try another handle if possible -
1121                         (we can not do this if closePending since
1122                         loop could be modified - in which case we
1123                         have to start at the beginning of the list
1124                         again. Note that it would be bad
1125                         to hold up writepages here (rather than
1126                         in caller) with continuous retries */
1127                         cFYI(1, ("wp failed on reopen file"));
1128                         read_lock(&GlobalSMBSeslock);
1129                         /* can not use this handle, no write
1130                            pending on this one after all */
1131                         atomic_dec(&open_file->wrtPending);
1132
1133                         if (open_file->closePend) /* list could have changed */
1134                                 goto refind_writable;
1135                         /* else we simply continue to the next entry. Thus
1136                            we do not loop on reopen errors.  If we
1137                            can not reopen the file, for example if we
1138                            reconnected to a server with another client
1139                            racing to delete or lock the file we would not
1140                            make progress if we restarted before the beginning
1141                            of the loop here. */
1142                 }
1143         }
1144         /* couldn't find useable FH with same pid, try any available */
1145         if (!any_available) {
1146                 any_available = true;
1147                 goto refind_writable;
1148         }
1149         read_unlock(&GlobalSMBSeslock);
1150         return NULL;
1151 }
1152
1153 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1154 {
1155         struct address_space *mapping = page->mapping;
1156         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1157         char *write_data;
1158         int rc = -EFAULT;
1159         int bytes_written = 0;
1160         struct cifs_sb_info *cifs_sb;
1161         struct cifsTconInfo *pTcon;
1162         struct inode *inode;
1163         struct cifsFileInfo *open_file;
1164
1165         if (!mapping || !mapping->host)
1166                 return -EFAULT;
1167
1168         inode = page->mapping->host;
1169         cifs_sb = CIFS_SB(inode->i_sb);
1170         pTcon = cifs_sb->tcon;
1171
1172         offset += (loff_t)from;
1173         write_data = kmap(page);
1174         write_data += from;
1175
1176         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1177                 kunmap(page);
1178                 return -EIO;
1179         }
1180
1181         /* racing with truncate? */
1182         if (offset > mapping->host->i_size) {
1183                 kunmap(page);
1184                 return 0; /* don't care */
1185         }
1186
1187         /* check to make sure that we are not extending the file */
1188         if (mapping->host->i_size - offset < (loff_t)to)
1189                 to = (unsigned)(mapping->host->i_size - offset);
1190
1191         open_file = find_writable_file(CIFS_I(mapping->host));
1192         if (open_file) {
1193                 bytes_written = cifs_write(open_file->pfile, write_data,
1194                                            to-from, &offset);
1195                 atomic_dec(&open_file->wrtPending);
1196                 /* Does mm or vfs already set times? */
1197                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1198                 if ((bytes_written > 0) && (offset))
1199                         rc = 0;
1200                 else if (bytes_written < 0)
1201                         rc = bytes_written;
1202         } else {
1203                 cFYI(1, ("No writeable filehandles for inode"));
1204                 rc = -EIO;
1205         }
1206
1207         kunmap(page);
1208         return rc;
1209 }
1210
1211 static int cifs_writepages(struct address_space *mapping,
1212                            struct writeback_control *wbc)
1213 {
1214         struct backing_dev_info *bdi = mapping->backing_dev_info;
1215         unsigned int bytes_to_write;
1216         unsigned int bytes_written;
1217         struct cifs_sb_info *cifs_sb;
1218         int done = 0;
1219         pgoff_t end;
1220         pgoff_t index;
1221         int range_whole = 0;
1222         struct kvec *iov;
1223         int len;
1224         int n_iov = 0;
1225         pgoff_t next;
1226         int nr_pages;
1227         __u64 offset = 0;
1228         struct cifsFileInfo *open_file;
1229         struct page *page;
1230         struct pagevec pvec;
1231         int rc = 0;
1232         int scanned = 0;
1233         int xid;
1234
1235         cifs_sb = CIFS_SB(mapping->host->i_sb);
1236
1237         /*
1238          * If wsize is smaller that the page cache size, default to writing
1239          * one page at a time via cifs_writepage
1240          */
1241         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1242                 return generic_writepages(mapping, wbc);
1243
1244         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1245                 if (cifs_sb->tcon->ses->server->secMode &
1246                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1247                         if (!experimEnabled)
1248                                 return generic_writepages(mapping, wbc);
1249
1250         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1251         if (iov == NULL)
1252                 return generic_writepages(mapping, wbc);
1253
1254
1255         /*
1256          * BB: Is this meaningful for a non-block-device file system?
1257          * If it is, we should test it again after we do I/O
1258          */
1259         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1260                 wbc->encountered_congestion = 1;
1261                 kfree(iov);
1262                 return 0;
1263         }
1264
1265         xid = GetXid();
1266
1267         pagevec_init(&pvec, 0);
1268         if (wbc->range_cyclic) {
1269                 index = mapping->writeback_index; /* Start from prev offset */
1270                 end = -1;
1271         } else {
1272                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1273                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1274                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1275                         range_whole = 1;
1276                 scanned = 1;
1277         }
1278 retry:
1279         while (!done && (index <= end) &&
1280                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1281                         PAGECACHE_TAG_DIRTY,
1282                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1283                 int first;
1284                 unsigned int i;
1285
1286                 first = -1;
1287                 next = 0;
1288                 n_iov = 0;
1289                 bytes_to_write = 0;
1290
1291                 for (i = 0; i < nr_pages; i++) {
1292                         page = pvec.pages[i];
1293                         /*
1294                          * At this point we hold neither mapping->tree_lock nor
1295                          * lock on the page itself: the page may be truncated or
1296                          * invalidated (changing page->mapping to NULL), or even
1297                          * swizzled back from swapper_space to tmpfs file
1298                          * mapping
1299                          */
1300
1301                         if (first < 0)
1302                                 lock_page(page);
1303                         else if (!trylock_page(page))
1304                                 break;
1305
1306                         if (unlikely(page->mapping != mapping)) {
1307                                 unlock_page(page);
1308                                 break;
1309                         }
1310
1311                         if (!wbc->range_cyclic && page->index > end) {
1312                                 done = 1;
1313                                 unlock_page(page);
1314                                 break;
1315                         }
1316
1317                         if (next && (page->index != next)) {
1318                                 /* Not next consecutive page */
1319                                 unlock_page(page);
1320                                 break;
1321                         }
1322
1323                         if (wbc->sync_mode != WB_SYNC_NONE)
1324                                 wait_on_page_writeback(page);
1325
1326                         if (PageWriteback(page) ||
1327                                         !clear_page_dirty_for_io(page)) {
1328                                 unlock_page(page);
1329                                 break;
1330                         }
1331
1332                         /*
1333                          * This actually clears the dirty bit in the radix tree.
1334                          * See cifs_writepage() for more commentary.
1335                          */
1336                         set_page_writeback(page);
1337
1338                         if (page_offset(page) >= mapping->host->i_size) {
1339                                 done = 1;
1340                                 unlock_page(page);
1341                                 end_page_writeback(page);
1342                                 break;
1343                         }
1344
1345                         /*
1346                          * BB can we get rid of this?  pages are held by pvec
1347                          */
1348                         page_cache_get(page);
1349
1350                         len = min(mapping->host->i_size - page_offset(page),
1351                                   (loff_t)PAGE_CACHE_SIZE);
1352
1353                         /* reserve iov[0] for the smb header */
1354                         n_iov++;
1355                         iov[n_iov].iov_base = kmap(page);
1356                         iov[n_iov].iov_len = len;
1357                         bytes_to_write += len;
1358
1359                         if (first < 0) {
1360                                 first = i;
1361                                 offset = page_offset(page);
1362                         }
1363                         next = page->index + 1;
1364                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1365                                 break;
1366                 }
1367                 if (n_iov) {
1368                         /* Search for a writable handle every time we call
1369                          * CIFSSMBWrite2.  We can't rely on the last handle
1370                          * we used to still be valid
1371                          */
1372                         open_file = find_writable_file(CIFS_I(mapping->host));
1373                         if (!open_file) {
1374                                 cERROR(1, ("No writable handles for inode"));
1375                                 rc = -EBADF;
1376                         } else {
1377                                 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1378                                                    open_file->netfid,
1379                                                    bytes_to_write, offset,
1380                                                    &bytes_written, iov, n_iov,
1381                                                    CIFS_LONG_OP);
1382                                 atomic_dec(&open_file->wrtPending);
1383                                 if (rc || bytes_written < bytes_to_write) {
1384                                         cERROR(1, ("Write2 ret %d, wrote %d",
1385                                                   rc, bytes_written));
1386                                         /* BB what if continued retry is
1387                                            requested via mount flags? */
1388                                         if (rc == -ENOSPC)
1389                                                 set_bit(AS_ENOSPC, &mapping->flags);
1390                                         else
1391                                                 set_bit(AS_EIO, &mapping->flags);
1392                                 } else {
1393                                         cifs_stats_bytes_written(cifs_sb->tcon,
1394                                                                  bytes_written);
1395                                 }
1396                         }
1397                         for (i = 0; i < n_iov; i++) {
1398                                 page = pvec.pages[first + i];
1399                                 /* Should we also set page error on
1400                                 success rc but too little data written? */
1401                                 /* BB investigate retry logic on temporary
1402                                 server crash cases and how recovery works
1403                                 when page marked as error */
1404                                 if (rc)
1405                                         SetPageError(page);
1406                                 kunmap(page);
1407                                 unlock_page(page);
1408                                 end_page_writeback(page);
1409                                 page_cache_release(page);
1410                         }
1411                         if ((wbc->nr_to_write -= n_iov) <= 0)
1412                                 done = 1;
1413                         index = next;
1414                 } else
1415                         /* Need to re-find the pages we skipped */
1416                         index = pvec.pages[0]->index + 1;
1417
1418                 pagevec_release(&pvec);
1419         }
1420         if (!scanned && !done) {
1421                 /*
1422                  * We hit the last page and there is more work to be done: wrap
1423                  * back to the start of the file
1424                  */
1425                 scanned = 1;
1426                 index = 0;
1427                 goto retry;
1428         }
1429         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1430                 mapping->writeback_index = index;
1431
1432         FreeXid(xid);
1433         kfree(iov);
1434         return rc;
1435 }
1436
1437 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1438 {
1439         int rc = -EFAULT;
1440         int xid;
1441
1442         xid = GetXid();
1443 /* BB add check for wbc flags */
1444         page_cache_get(page);
1445         if (!PageUptodate(page))
1446                 cFYI(1, ("ppw - page not up to date"));
1447
1448         /*
1449          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1450          *
1451          * A writepage() implementation always needs to do either this,
1452          * or re-dirty the page with "redirty_page_for_writepage()" in
1453          * the case of a failure.
1454          *
1455          * Just unlocking the page will cause the radix tree tag-bits
1456          * to fail to update with the state of the page correctly.
1457          */
1458         set_page_writeback(page);
1459         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1460         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1461         unlock_page(page);
1462         end_page_writeback(page);
1463         page_cache_release(page);
1464         FreeXid(xid);
1465         return rc;
1466 }
1467
1468 static int cifs_write_end(struct file *file, struct address_space *mapping,
1469                         loff_t pos, unsigned len, unsigned copied,
1470                         struct page *page, void *fsdata)
1471 {
1472         int rc;
1473         struct inode *inode = mapping->host;
1474
1475         cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1476                  page, pos, copied));
1477
1478         if (PageChecked(page)) {
1479                 if (copied == len)
1480                         SetPageUptodate(page);
1481                 ClearPageChecked(page);
1482         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1483                 SetPageUptodate(page);
1484
1485         if (!PageUptodate(page)) {
1486                 char *page_data;
1487                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1488                 int xid;
1489
1490                 xid = GetXid();
1491                 /* this is probably better than directly calling
1492                    partialpage_write since in this function the file handle is
1493                    known which we might as well leverage */
1494                 /* BB check if anything else missing out of ppw
1495                    such as updating last write time */
1496                 page_data = kmap(page);
1497                 rc = cifs_write(file, page_data + offset, copied, &pos);
1498                 /* if (rc < 0) should we set writebehind rc? */
1499                 kunmap(page);
1500
1501                 FreeXid(xid);
1502         } else {
1503                 rc = copied;
1504                 pos += copied;
1505                 set_page_dirty(page);
1506         }
1507
1508         if (rc > 0) {
1509                 spin_lock(&inode->i_lock);
1510                 if (pos > inode->i_size)
1511                         i_size_write(inode, pos);
1512                 spin_unlock(&inode->i_lock);
1513         }
1514
1515         unlock_page(page);
1516         page_cache_release(page);
1517
1518         return rc;
1519 }
1520
1521 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1522 {
1523         int xid;
1524         int rc = 0;
1525         struct inode *inode = file->f_path.dentry->d_inode;
1526
1527         xid = GetXid();
1528
1529         cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1530                 dentry->d_name.name, datasync));
1531
1532         rc = filemap_write_and_wait(inode->i_mapping);
1533         if (rc == 0) {
1534                 rc = CIFS_I(inode)->write_behind_rc;
1535                 CIFS_I(inode)->write_behind_rc = 0;
1536         }
1537         FreeXid(xid);
1538         return rc;
1539 }
1540
1541 /* static void cifs_sync_page(struct page *page)
1542 {
1543         struct address_space *mapping;
1544         struct inode *inode;
1545         unsigned long index = page->index;
1546         unsigned int rpages = 0;
1547         int rc = 0;
1548
1549         cFYI(1, ("sync page %p",page));
1550         mapping = page->mapping;
1551         if (!mapping)
1552                 return 0;
1553         inode = mapping->host;
1554         if (!inode)
1555                 return; */
1556
1557 /*      fill in rpages then
1558         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1559
1560 /*      cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1561
1562 #if 0
1563         if (rc < 0)
1564                 return rc;
1565         return 0;
1566 #endif
1567 } */
1568
1569 /*
1570  * As file closes, flush all cached write data for this inode checking
1571  * for write behind errors.
1572  */
1573 int cifs_flush(struct file *file, fl_owner_t id)
1574 {
1575         struct inode *inode = file->f_path.dentry->d_inode;
1576         int rc = 0;
1577
1578         /* Rather than do the steps manually:
1579            lock the inode for writing
1580            loop through pages looking for write behind data (dirty pages)
1581            coalesce into contiguous 16K (or smaller) chunks to write to server
1582            send to server (prefer in parallel)
1583            deal with writebehind errors
1584            unlock inode for writing
1585            filemapfdatawrite appears easier for the time being */
1586
1587         rc = filemap_fdatawrite(inode->i_mapping);
1588         /* reset wb rc if we were able to write out dirty pages */
1589         if (!rc) {
1590                 rc = CIFS_I(inode)->write_behind_rc;
1591                 CIFS_I(inode)->write_behind_rc = 0;
1592         }
1593
1594         cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1595
1596         return rc;
1597 }
1598
1599 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1600         size_t read_size, loff_t *poffset)
1601 {
1602         int rc = -EACCES;
1603         unsigned int bytes_read = 0;
1604         unsigned int total_read = 0;
1605         unsigned int current_read_size;
1606         struct cifs_sb_info *cifs_sb;
1607         struct cifsTconInfo *pTcon;
1608         int xid;
1609         struct cifsFileInfo *open_file;
1610         char *smb_read_data;
1611         char __user *current_offset;
1612         struct smb_com_read_rsp *pSMBr;
1613
1614         xid = GetXid();
1615         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1616         pTcon = cifs_sb->tcon;
1617
1618         if (file->private_data == NULL) {
1619                 FreeXid(xid);
1620                 return -EBADF;
1621         }
1622         open_file = (struct cifsFileInfo *)file->private_data;
1623
1624         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1625                 cFYI(1, ("attempting read on write only file instance"));
1626
1627         for (total_read = 0, current_offset = read_data;
1628              read_size > total_read;
1629              total_read += bytes_read, current_offset += bytes_read) {
1630                 current_read_size = min_t(const int, read_size - total_read,
1631                                           cifs_sb->rsize);
1632                 rc = -EAGAIN;
1633                 smb_read_data = NULL;
1634                 while (rc == -EAGAIN) {
1635                         int buf_type = CIFS_NO_BUFFER;
1636                         if ((open_file->invalidHandle) &&
1637                             (!open_file->closePend)) {
1638                                 rc = cifs_reopen_file(file, true);
1639                                 if (rc != 0)
1640                                         break;
1641                         }
1642                         rc = CIFSSMBRead(xid, pTcon,
1643                                          open_file->netfid,
1644                                          current_read_size, *poffset,
1645                                          &bytes_read, &smb_read_data,
1646                                          &buf_type);
1647                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1648                         if (smb_read_data) {
1649                                 if (copy_to_user(current_offset,
1650                                                 smb_read_data +
1651                                                 4 /* RFC1001 length field */ +
1652                                                 le16_to_cpu(pSMBr->DataOffset),
1653                                                 bytes_read))
1654                                         rc = -EFAULT;
1655
1656                                 if (buf_type == CIFS_SMALL_BUFFER)
1657                                         cifs_small_buf_release(smb_read_data);
1658                                 else if (buf_type == CIFS_LARGE_BUFFER)
1659                                         cifs_buf_release(smb_read_data);
1660                                 smb_read_data = NULL;
1661                         }
1662                 }
1663                 if (rc || (bytes_read == 0)) {
1664                         if (total_read) {
1665                                 break;
1666                         } else {
1667                                 FreeXid(xid);
1668                                 return rc;
1669                         }
1670                 } else {
1671                         cifs_stats_bytes_read(pTcon, bytes_read);
1672                         *poffset += bytes_read;
1673                 }
1674         }
1675         FreeXid(xid);
1676         return total_read;
1677 }
1678
1679
1680 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1681         loff_t *poffset)
1682 {
1683         int rc = -EACCES;
1684         unsigned int bytes_read = 0;
1685         unsigned int total_read;
1686         unsigned int current_read_size;
1687         struct cifs_sb_info *cifs_sb;
1688         struct cifsTconInfo *pTcon;
1689         int xid;
1690         char *current_offset;
1691         struct cifsFileInfo *open_file;
1692         int buf_type = CIFS_NO_BUFFER;
1693
1694         xid = GetXid();
1695         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1696         pTcon = cifs_sb->tcon;
1697
1698         if (file->private_data == NULL) {
1699                 FreeXid(xid);
1700                 return -EBADF;
1701         }
1702         open_file = (struct cifsFileInfo *)file->private_data;
1703
1704         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1705                 cFYI(1, ("attempting read on write only file instance"));
1706
1707         for (total_read = 0, current_offset = read_data;
1708              read_size > total_read;
1709              total_read += bytes_read, current_offset += bytes_read) {
1710                 current_read_size = min_t(const int, read_size - total_read,
1711                                           cifs_sb->rsize);
1712                 /* For windows me and 9x we do not want to request more
1713                 than it negotiated since it will refuse the read then */
1714                 if ((pTcon->ses) &&
1715                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1716                         current_read_size = min_t(const int, current_read_size,
1717                                         pTcon->ses->server->maxBuf - 128);
1718                 }
1719                 rc = -EAGAIN;
1720                 while (rc == -EAGAIN) {
1721                         if ((open_file->invalidHandle) &&
1722                             (!open_file->closePend)) {
1723                                 rc = cifs_reopen_file(file, true);
1724                                 if (rc != 0)
1725                                         break;
1726                         }
1727                         rc = CIFSSMBRead(xid, pTcon,
1728                                          open_file->netfid,
1729                                          current_read_size, *poffset,
1730                                          &bytes_read, &current_offset,
1731                                          &buf_type);
1732                 }
1733                 if (rc || (bytes_read == 0)) {
1734                         if (total_read) {
1735                                 break;
1736                         } else {
1737                                 FreeXid(xid);
1738                                 return rc;
1739                         }
1740                 } else {
1741                         cifs_stats_bytes_read(pTcon, total_read);
1742                         *poffset += bytes_read;
1743                 }
1744         }
1745         FreeXid(xid);
1746         return total_read;
1747 }
1748
1749 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1750 {
1751         struct dentry *dentry = file->f_path.dentry;
1752         int rc, xid;
1753
1754         xid = GetXid();
1755         rc = cifs_revalidate(dentry);
1756         if (rc) {
1757                 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1758                 FreeXid(xid);
1759                 return rc;
1760         }
1761         rc = generic_file_mmap(file, vma);
1762         FreeXid(xid);
1763         return rc;
1764 }
1765
1766
1767 static void cifs_copy_cache_pages(struct address_space *mapping,
1768         struct list_head *pages, int bytes_read, char *data,
1769         struct pagevec *plru_pvec)
1770 {
1771         struct page *page;
1772         char *target;
1773
1774         while (bytes_read > 0) {
1775                 if (list_empty(pages))
1776                         break;
1777
1778                 page = list_entry(pages->prev, struct page, lru);
1779                 list_del(&page->lru);
1780
1781                 if (add_to_page_cache(page, mapping, page->index,
1782                                       GFP_KERNEL)) {
1783                         page_cache_release(page);
1784                         cFYI(1, ("Add page cache failed"));
1785                         data += PAGE_CACHE_SIZE;
1786                         bytes_read -= PAGE_CACHE_SIZE;
1787                         continue;
1788                 }
1789
1790                 target = kmap_atomic(page, KM_USER0);
1791
1792                 if (PAGE_CACHE_SIZE > bytes_read) {
1793                         memcpy(target, data, bytes_read);
1794                         /* zero the tail end of this partial page */
1795                         memset(target + bytes_read, 0,
1796                                PAGE_CACHE_SIZE - bytes_read);
1797                         bytes_read = 0;
1798                 } else {
1799                         memcpy(target, data, PAGE_CACHE_SIZE);
1800                         bytes_read -= PAGE_CACHE_SIZE;
1801                 }
1802                 kunmap_atomic(target, KM_USER0);
1803
1804                 flush_dcache_page(page);
1805                 SetPageUptodate(page);
1806                 unlock_page(page);
1807                 if (!pagevec_add(plru_pvec, page))
1808                         __pagevec_lru_add_file(plru_pvec);
1809                 data += PAGE_CACHE_SIZE;
1810         }
1811         return;
1812 }
1813
1814 static int cifs_readpages(struct file *file, struct address_space *mapping,
1815         struct list_head *page_list, unsigned num_pages)
1816 {
1817         int rc = -EACCES;
1818         int xid;
1819         loff_t offset;
1820         struct page *page;
1821         struct cifs_sb_info *cifs_sb;
1822         struct cifsTconInfo *pTcon;
1823         unsigned int bytes_read = 0;
1824         unsigned int read_size, i;
1825         char *smb_read_data = NULL;
1826         struct smb_com_read_rsp *pSMBr;
1827         struct pagevec lru_pvec;
1828         struct cifsFileInfo *open_file;
1829         int buf_type = CIFS_NO_BUFFER;
1830
1831         xid = GetXid();
1832         if (file->private_data == NULL) {
1833                 FreeXid(xid);
1834                 return -EBADF;
1835         }
1836         open_file = (struct cifsFileInfo *)file->private_data;
1837         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1838         pTcon = cifs_sb->tcon;
1839
1840         pagevec_init(&lru_pvec, 0);
1841         cFYI(DBG2, ("rpages: num pages %d", num_pages));
1842         for (i = 0; i < num_pages; ) {
1843                 unsigned contig_pages;
1844                 struct page *tmp_page;
1845                 unsigned long expected_index;
1846
1847                 if (list_empty(page_list))
1848                         break;
1849
1850                 page = list_entry(page_list->prev, struct page, lru);
1851                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1852
1853                 /* count adjacent pages that we will read into */
1854                 contig_pages = 0;
1855                 expected_index =
1856                         list_entry(page_list->prev, struct page, lru)->index;
1857                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1858                         if (tmp_page->index == expected_index) {
1859                                 contig_pages++;
1860                                 expected_index++;
1861                         } else
1862                                 break;
1863                 }
1864                 if (contig_pages + i >  num_pages)
1865                         contig_pages = num_pages - i;
1866
1867                 /* for reads over a certain size could initiate async
1868                    read ahead */
1869
1870                 read_size = contig_pages * PAGE_CACHE_SIZE;
1871                 /* Read size needs to be in multiples of one page */
1872                 read_size = min_t(const unsigned int, read_size,
1873                                   cifs_sb->rsize & PAGE_CACHE_MASK);
1874                 cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
1875                                 read_size, contig_pages));
1876                 rc = -EAGAIN;
1877                 while (rc == -EAGAIN) {
1878                         if ((open_file->invalidHandle) &&
1879                             (!open_file->closePend)) {
1880                                 rc = cifs_reopen_file(file, true);
1881                                 if (rc != 0)
1882                                         break;
1883                         }
1884
1885                         rc = CIFSSMBRead(xid, pTcon,
1886                                          open_file->netfid,
1887                                          read_size, offset,
1888                                          &bytes_read, &smb_read_data,
1889                                          &buf_type);
1890                         /* BB more RC checks ? */
1891                         if (rc == -EAGAIN) {
1892                                 if (smb_read_data) {
1893                                         if (buf_type == CIFS_SMALL_BUFFER)
1894                                                 cifs_small_buf_release(smb_read_data);
1895                                         else if (buf_type == CIFS_LARGE_BUFFER)
1896                                                 cifs_buf_release(smb_read_data);
1897                                         smb_read_data = NULL;
1898                                 }
1899                         }
1900                 }
1901                 if ((rc < 0) || (smb_read_data == NULL)) {
1902                         cFYI(1, ("Read error in readpages: %d", rc));
1903                         break;
1904                 } else if (bytes_read > 0) {
1905                         task_io_account_read(bytes_read);
1906                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1907                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
1908                                 smb_read_data + 4 /* RFC1001 hdr */ +
1909                                 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
1910
1911                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
1912                         cifs_stats_bytes_read(pTcon, bytes_read);
1913                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
1914                                 i++; /* account for partial page */
1915
1916                                 /* server copy of file can have smaller size
1917                                    than client */
1918                                 /* BB do we need to verify this common case ?
1919                                    this case is ok - if we are at server EOF
1920                                    we will hit it on next read */
1921
1922                                 /* break; */
1923                         }
1924                 } else {
1925                         cFYI(1, ("No bytes read (%d) at offset %lld . "
1926                                  "Cleaning remaining pages from readahead list",
1927                                  bytes_read, offset));
1928                         /* BB turn off caching and do new lookup on
1929                            file size at server? */
1930                         break;
1931                 }
1932                 if (smb_read_data) {
1933                         if (buf_type == CIFS_SMALL_BUFFER)
1934                                 cifs_small_buf_release(smb_read_data);
1935                         else if (buf_type == CIFS_LARGE_BUFFER)
1936                                 cifs_buf_release(smb_read_data);
1937                         smb_read_data = NULL;
1938                 }
1939                 bytes_read = 0;
1940         }
1941
1942         pagevec_lru_add_file(&lru_pvec);
1943
1944 /* need to free smb_read_data buf before exit */
1945         if (smb_read_data) {
1946                 if (buf_type == CIFS_SMALL_BUFFER)
1947                         cifs_small_buf_release(smb_read_data);
1948                 else if (buf_type == CIFS_LARGE_BUFFER)
1949                         cifs_buf_release(smb_read_data);
1950                 smb_read_data = NULL;
1951         }
1952
1953         FreeXid(xid);
1954         return rc;
1955 }
1956
1957 static int cifs_readpage_worker(struct file *file, struct page *page,
1958         loff_t *poffset)
1959 {
1960         char *read_data;
1961         int rc;
1962
1963         page_cache_get(page);
1964         read_data = kmap(page);
1965         /* for reads over a certain size could initiate async read ahead */
1966
1967         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
1968
1969         if (rc < 0)
1970                 goto io_error;
1971         else
1972                 cFYI(1, ("Bytes read %d", rc));
1973
1974         file->f_path.dentry->d_inode->i_atime =
1975                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
1976
1977         if (PAGE_CACHE_SIZE > rc)
1978                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
1979
1980         flush_dcache_page(page);
1981         SetPageUptodate(page);
1982         rc = 0;
1983
1984 io_error:
1985         kunmap(page);
1986         page_cache_release(page);
1987         return rc;
1988 }
1989
1990 static int cifs_readpage(struct file *file, struct page *page)
1991 {
1992         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1993         int rc = -EACCES;
1994         int xid;
1995
1996         xid = GetXid();
1997
1998         if (file->private_data == NULL) {
1999                 FreeXid(xid);
2000                 return -EBADF;
2001         }
2002
2003         cFYI(1, ("readpage %p at offset %d 0x%x\n",
2004                  page, (int)offset, (int)offset));
2005
2006         rc = cifs_readpage_worker(file, page, &offset);
2007
2008         unlock_page(page);
2009
2010         FreeXid(xid);
2011         return rc;
2012 }
2013
2014 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2015 {
2016         struct cifsFileInfo *open_file;
2017
2018         read_lock(&GlobalSMBSeslock);
2019         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2020                 if (open_file->closePend)
2021                         continue;
2022                 if (open_file->pfile &&
2023                     ((open_file->pfile->f_flags & O_RDWR) ||
2024                      (open_file->pfile->f_flags & O_WRONLY))) {
2025                         read_unlock(&GlobalSMBSeslock);
2026                         return 1;
2027                 }
2028         }
2029         read_unlock(&GlobalSMBSeslock);
2030         return 0;
2031 }
2032
2033 /* We do not want to update the file size from server for inodes
2034    open for write - to avoid races with writepage extending
2035    the file - in the future we could consider allowing
2036    refreshing the inode only on increases in the file size
2037    but this is tricky to do without racing with writebehind
2038    page caching in the current Linux kernel design */
2039 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2040 {
2041         if (!cifsInode)
2042                 return true;
2043
2044         if (is_inode_writable(cifsInode)) {
2045                 /* This inode is open for write at least once */
2046                 struct cifs_sb_info *cifs_sb;
2047
2048                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2049                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2050                         /* since no page cache to corrupt on directio
2051                         we can change size safely */
2052                         return true;
2053                 }
2054
2055                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2056                         return true;
2057
2058                 return false;
2059         } else
2060                 return true;
2061 }
2062
2063 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2064                         loff_t pos, unsigned len, unsigned flags,
2065                         struct page **pagep, void **fsdata)
2066 {
2067         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2068         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2069         loff_t page_start = pos & PAGE_MASK;
2070         loff_t i_size;
2071         struct page *page;
2072         int rc = 0;
2073
2074         cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2075
2076         page = __grab_cache_page(mapping, index);
2077         if (!page) {
2078                 rc = -ENOMEM;
2079                 goto out;
2080         }
2081
2082         if (PageUptodate(page))
2083                 goto out;
2084
2085         /*
2086          * If we write a full page it will be up to date, no need to read from
2087          * the server. If the write is short, we'll end up doing a sync write
2088          * instead.
2089          */
2090         if (len == PAGE_CACHE_SIZE)
2091                 goto out;
2092
2093         /*
2094          * optimize away the read when we have an oplock, and we're not
2095          * expecting to use any of the data we'd be reading in. That
2096          * is, when the page lies beyond the EOF, or straddles the EOF
2097          * and the write will cover all of the existing data.
2098          */
2099         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2100                 i_size = i_size_read(mapping->host);
2101                 if (page_start >= i_size ||
2102                     (offset == 0 && (pos + len) >= i_size)) {
2103                         zero_user_segments(page, 0, offset,
2104                                            offset + len,
2105                                            PAGE_CACHE_SIZE);
2106                         /*
2107                          * PageChecked means that the parts of the page
2108                          * to which we're not writing are considered up
2109                          * to date. Once the data is copied to the
2110                          * page, it can be set uptodate.
2111                          */
2112                         SetPageChecked(page);
2113                         goto out;
2114                 }
2115         }
2116
2117         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2118                 /*
2119                  * might as well read a page, it is fast enough. If we get
2120                  * an error, we don't need to return it. cifs_write_end will
2121                  * do a sync write instead since PG_uptodate isn't set.
2122                  */
2123                 cifs_readpage_worker(file, page, &page_start);
2124         } else {
2125                 /* we could try using another file handle if there is one -
2126                    but how would we lock it to prevent close of that handle
2127                    racing with this read? In any case
2128                    this will be written out by write_end so is fine */
2129         }
2130 out:
2131         *pagep = page;
2132         return rc;
2133 }
2134
2135 const struct address_space_operations cifs_addr_ops = {
2136         .readpage = cifs_readpage,
2137         .readpages = cifs_readpages,
2138         .writepage = cifs_writepage,
2139         .writepages = cifs_writepages,
2140         .write_begin = cifs_write_begin,
2141         .write_end = cifs_write_end,
2142         .set_page_dirty = __set_page_dirty_nobuffers,
2143         /* .sync_page = cifs_sync_page, */
2144         /* .direct_IO = */
2145 };
2146
2147 /*
2148  * cifs_readpages requires the server to support a buffer large enough to
2149  * contain the header plus one complete page of data.  Otherwise, we need
2150  * to leave cifs_readpages out of the address space operations.
2151  */
2152 const struct address_space_operations cifs_addr_ops_smallbuf = {
2153         .readpage = cifs_readpage,
2154         .writepage = cifs_writepage,
2155         .writepages = cifs_writepages,
2156         .write_begin = cifs_write_begin,
2157         .write_end = cifs_write_end,
2158         .set_page_dirty = __set_page_dirty_nobuffers,
2159         /* .sync_page = cifs_sync_page, */
2160         /* .direct_IO = */
2161 };