Merge git://git.kernel.org/pub/scm/linux/kernel/git/sfrench/cifs-2.6
[pandora-kernel.git] / fs / cifs / file.c
1 /*
2  *   fs/cifs/file.c
3  *
4  *   vfs operations that deal with files
5  *
6  *   Copyright (C) International Business Machines  Corp., 2002,2010
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 <linux/mount.h>
34 #include <linux/slab.h>
35 #include <asm/div64.h>
36 #include "cifsfs.h"
37 #include "cifspdu.h"
38 #include "cifsglob.h"
39 #include "cifsproto.h"
40 #include "cifs_unicode.h"
41 #include "cifs_debug.h"
42 #include "cifs_fs_sb.h"
43 #include "fscache.h"
44
45 static inline int cifs_convert_flags(unsigned int flags)
46 {
47         if ((flags & O_ACCMODE) == O_RDONLY)
48                 return GENERIC_READ;
49         else if ((flags & O_ACCMODE) == O_WRONLY)
50                 return GENERIC_WRITE;
51         else if ((flags & O_ACCMODE) == O_RDWR) {
52                 /* GENERIC_ALL is too much permission to request
53                    can cause unnecessary access denied on create */
54                 /* return GENERIC_ALL; */
55                 return (GENERIC_READ | GENERIC_WRITE);
56         }
57
58         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
59                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
60                 FILE_READ_DATA);
61 }
62
63 static u32 cifs_posix_convert_flags(unsigned int flags)
64 {
65         u32 posix_flags = 0;
66
67         if ((flags & O_ACCMODE) == O_RDONLY)
68                 posix_flags = SMB_O_RDONLY;
69         else if ((flags & O_ACCMODE) == O_WRONLY)
70                 posix_flags = SMB_O_WRONLY;
71         else if ((flags & O_ACCMODE) == O_RDWR)
72                 posix_flags = SMB_O_RDWR;
73
74         if (flags & O_CREAT)
75                 posix_flags |= SMB_O_CREAT;
76         if (flags & O_EXCL)
77                 posix_flags |= SMB_O_EXCL;
78         if (flags & O_TRUNC)
79                 posix_flags |= SMB_O_TRUNC;
80         /* be safe and imply O_SYNC for O_DSYNC */
81         if (flags & O_DSYNC)
82                 posix_flags |= SMB_O_SYNC;
83         if (flags & O_DIRECTORY)
84                 posix_flags |= SMB_O_DIRECTORY;
85         if (flags & O_NOFOLLOW)
86                 posix_flags |= SMB_O_NOFOLLOW;
87         if (flags & O_DIRECT)
88                 posix_flags |= SMB_O_DIRECT;
89
90         return posix_flags;
91 }
92
93 static inline int cifs_get_disposition(unsigned int flags)
94 {
95         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
96                 return FILE_CREATE;
97         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
98                 return FILE_OVERWRITE_IF;
99         else if ((flags & O_CREAT) == O_CREAT)
100                 return FILE_OPEN_IF;
101         else if ((flags & O_TRUNC) == O_TRUNC)
102                 return FILE_OVERWRITE;
103         else
104                 return FILE_OPEN;
105 }
106
107 int cifs_posix_open(char *full_path, struct inode **pinode,
108                         struct super_block *sb, int mode, unsigned int f_flags,
109                         __u32 *poplock, __u16 *pnetfid, int xid)
110 {
111         int rc;
112         FILE_UNIX_BASIC_INFO *presp_data;
113         __u32 posix_flags = 0;
114         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
115         struct cifs_fattr fattr;
116         struct tcon_link *tlink;
117         struct cifsTconInfo *tcon;
118
119         cFYI(1, "posix open %s", full_path);
120
121         presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
122         if (presp_data == NULL)
123                 return -ENOMEM;
124
125         tlink = cifs_sb_tlink(cifs_sb);
126         if (IS_ERR(tlink)) {
127                 rc = PTR_ERR(tlink);
128                 goto posix_open_ret;
129         }
130
131         tcon = tlink_tcon(tlink);
132         mode &= ~current_umask();
133
134         posix_flags = cifs_posix_convert_flags(f_flags);
135         rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
136                              poplock, full_path, cifs_sb->local_nls,
137                              cifs_sb->mnt_cifs_flags &
138                                         CIFS_MOUNT_MAP_SPECIAL_CHR);
139         cifs_put_tlink(tlink);
140
141         if (rc)
142                 goto posix_open_ret;
143
144         if (presp_data->Type == cpu_to_le32(-1))
145                 goto posix_open_ret; /* open ok, caller does qpathinfo */
146
147         if (!pinode)
148                 goto posix_open_ret; /* caller does not need info */
149
150         cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
151
152         /* get new inode and set it up */
153         if (*pinode == NULL) {
154                 cifs_fill_uniqueid(sb, &fattr);
155                 *pinode = cifs_iget(sb, &fattr);
156                 if (!*pinode) {
157                         rc = -ENOMEM;
158                         goto posix_open_ret;
159                 }
160         } else {
161                 cifs_fattr_to_inode(*pinode, &fattr);
162         }
163
164 posix_open_ret:
165         kfree(presp_data);
166         return rc;
167 }
168
169 static int
170 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
171              struct cifsTconInfo *tcon, unsigned int f_flags, __u32 *poplock,
172              __u16 *pnetfid, int xid)
173 {
174         int rc;
175         int desiredAccess;
176         int disposition;
177         FILE_ALL_INFO *buf;
178
179         desiredAccess = cifs_convert_flags(f_flags);
180
181 /*********************************************************************
182  *  open flag mapping table:
183  *
184  *      POSIX Flag            CIFS Disposition
185  *      ----------            ----------------
186  *      O_CREAT               FILE_OPEN_IF
187  *      O_CREAT | O_EXCL      FILE_CREATE
188  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
189  *      O_TRUNC               FILE_OVERWRITE
190  *      none of the above     FILE_OPEN
191  *
192  *      Note that there is not a direct match between disposition
193  *      FILE_SUPERSEDE (ie create whether or not file exists although
194  *      O_CREAT | O_TRUNC is similar but truncates the existing
195  *      file rather than creating a new file as FILE_SUPERSEDE does
196  *      (which uses the attributes / metadata passed in on open call)
197  *?
198  *?  O_SYNC is a reasonable match to CIFS writethrough flag
199  *?  and the read write flags match reasonably.  O_LARGEFILE
200  *?  is irrelevant because largefile support is always used
201  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
202  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
203  *********************************************************************/
204
205         disposition = cifs_get_disposition(f_flags);
206
207         /* BB pass O_SYNC flag through on file attributes .. BB */
208
209         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
210         if (!buf)
211                 return -ENOMEM;
212
213         if (tcon->ses->capabilities & CAP_NT_SMBS)
214                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
215                          desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
216                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
217                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
218         else
219                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
220                         desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
221                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
222                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
223
224         if (rc)
225                 goto out;
226
227         if (tcon->unix_ext)
228                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
229                                               xid);
230         else
231                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
232                                          xid, pnetfid);
233
234 out:
235         kfree(buf);
236         return rc;
237 }
238
239 struct cifsFileInfo *
240 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
241                   struct tcon_link *tlink, __u32 oplock)
242 {
243         struct dentry *dentry = file->f_path.dentry;
244         struct inode *inode = dentry->d_inode;
245         struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
246         struct cifsFileInfo *pCifsFile;
247
248         pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
249         if (pCifsFile == NULL)
250                 return pCifsFile;
251
252         pCifsFile->count = 1;
253         pCifsFile->netfid = fileHandle;
254         pCifsFile->pid = current->tgid;
255         pCifsFile->uid = current_fsuid();
256         pCifsFile->dentry = dget(dentry);
257         pCifsFile->f_flags = file->f_flags;
258         pCifsFile->invalidHandle = false;
259         pCifsFile->tlink = cifs_get_tlink(tlink);
260         mutex_init(&pCifsFile->fh_mutex);
261         mutex_init(&pCifsFile->lock_mutex);
262         INIT_LIST_HEAD(&pCifsFile->llist);
263         INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
264
265         spin_lock(&cifs_file_list_lock);
266         list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
267         /* if readable file instance put first in list*/
268         if (file->f_mode & FMODE_READ)
269                 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
270         else
271                 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
272         spin_unlock(&cifs_file_list_lock);
273
274         cifs_set_oplock_level(pCifsInode, oplock);
275
276         file->private_data = pCifsFile;
277         return pCifsFile;
278 }
279
280 /*
281  * Release a reference on the file private data. This may involve closing
282  * the filehandle out on the server. Must be called without holding
283  * cifs_file_list_lock.
284  */
285 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
286 {
287         struct inode *inode = cifs_file->dentry->d_inode;
288         struct cifsTconInfo *tcon = tlink_tcon(cifs_file->tlink);
289         struct cifsInodeInfo *cifsi = CIFS_I(inode);
290         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
291         struct cifsLockInfo *li, *tmp;
292
293         spin_lock(&cifs_file_list_lock);
294         if (--cifs_file->count > 0) {
295                 spin_unlock(&cifs_file_list_lock);
296                 return;
297         }
298
299         /* remove it from the lists */
300         list_del(&cifs_file->flist);
301         list_del(&cifs_file->tlist);
302
303         if (list_empty(&cifsi->openFileList)) {
304                 cFYI(1, "closing last open instance for inode %p",
305                         cifs_file->dentry->d_inode);
306
307                 /* in strict cache mode we need invalidate mapping on the last
308                    close  because it may cause a error when we open this file
309                    again and get at least level II oplock */
310                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
311                         CIFS_I(inode)->invalid_mapping = true;
312
313                 cifs_set_oplock_level(cifsi, 0);
314         }
315         spin_unlock(&cifs_file_list_lock);
316
317         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
318                 int xid, rc;
319
320                 xid = GetXid();
321                 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
322                 FreeXid(xid);
323         }
324
325         /* Delete any outstanding lock records. We'll lose them when the file
326          * is closed anyway.
327          */
328         mutex_lock(&cifs_file->lock_mutex);
329         list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
330                 list_del(&li->llist);
331                 kfree(li);
332         }
333         mutex_unlock(&cifs_file->lock_mutex);
334
335         cifs_put_tlink(cifs_file->tlink);
336         dput(cifs_file->dentry);
337         kfree(cifs_file);
338 }
339
340 int cifs_open(struct inode *inode, struct file *file)
341 {
342         int rc = -EACCES;
343         int xid;
344         __u32 oplock;
345         struct cifs_sb_info *cifs_sb;
346         struct cifsTconInfo *tcon;
347         struct tcon_link *tlink;
348         struct cifsFileInfo *pCifsFile = NULL;
349         char *full_path = NULL;
350         bool posix_open_ok = false;
351         __u16 netfid;
352
353         xid = GetXid();
354
355         cifs_sb = CIFS_SB(inode->i_sb);
356         tlink = cifs_sb_tlink(cifs_sb);
357         if (IS_ERR(tlink)) {
358                 FreeXid(xid);
359                 return PTR_ERR(tlink);
360         }
361         tcon = tlink_tcon(tlink);
362
363         full_path = build_path_from_dentry(file->f_path.dentry);
364         if (full_path == NULL) {
365                 rc = -ENOMEM;
366                 goto out;
367         }
368
369         cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
370                  inode, file->f_flags, full_path);
371
372         if (oplockEnabled)
373                 oplock = REQ_OPLOCK;
374         else
375                 oplock = 0;
376
377         if (!tcon->broken_posix_open && tcon->unix_ext &&
378             (tcon->ses->capabilities & CAP_UNIX) &&
379             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
380                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
381                 /* can not refresh inode info since size could be stale */
382                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
383                                 cifs_sb->mnt_file_mode /* ignored */,
384                                 file->f_flags, &oplock, &netfid, xid);
385                 if (rc == 0) {
386                         cFYI(1, "posix open succeeded");
387                         posix_open_ok = true;
388                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
389                         if (tcon->ses->serverNOS)
390                                 cERROR(1, "server %s of type %s returned"
391                                            " unexpected error on SMB posix open"
392                                            ", disabling posix open support."
393                                            " Check if server update available.",
394                                            tcon->ses->serverName,
395                                            tcon->ses->serverNOS);
396                         tcon->broken_posix_open = true;
397                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
398                          (rc != -EOPNOTSUPP)) /* path not found or net err */
399                         goto out;
400                 /* else fallthrough to retry open the old way on network i/o
401                    or DFS errors */
402         }
403
404         if (!posix_open_ok) {
405                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
406                                   file->f_flags, &oplock, &netfid, xid);
407                 if (rc)
408                         goto out;
409         }
410
411         pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
412         if (pCifsFile == NULL) {
413                 CIFSSMBClose(xid, tcon, netfid);
414                 rc = -ENOMEM;
415                 goto out;
416         }
417
418         cifs_fscache_set_inode_cookie(inode, file);
419
420         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
421                 /* time to set mode which we can not set earlier due to
422                    problems creating new read-only files */
423                 struct cifs_unix_set_info_args args = {
424                         .mode   = inode->i_mode,
425                         .uid    = NO_CHANGE_64,
426                         .gid    = NO_CHANGE_64,
427                         .ctime  = NO_CHANGE_64,
428                         .atime  = NO_CHANGE_64,
429                         .mtime  = NO_CHANGE_64,
430                         .device = 0,
431                 };
432                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
433                                         pCifsFile->pid);
434         }
435
436 out:
437         kfree(full_path);
438         FreeXid(xid);
439         cifs_put_tlink(tlink);
440         return rc;
441 }
442
443 /* Try to reacquire byte range locks that were released when session */
444 /* to server was lost */
445 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
446 {
447         int rc = 0;
448
449 /* BB list all locks open on this file and relock */
450
451         return rc;
452 }
453
454 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
455 {
456         int rc = -EACCES;
457         int xid;
458         __u32 oplock;
459         struct cifs_sb_info *cifs_sb;
460         struct cifsTconInfo *tcon;
461         struct cifsInodeInfo *pCifsInode;
462         struct inode *inode;
463         char *full_path = NULL;
464         int desiredAccess;
465         int disposition = FILE_OPEN;
466         __u16 netfid;
467
468         xid = GetXid();
469         mutex_lock(&pCifsFile->fh_mutex);
470         if (!pCifsFile->invalidHandle) {
471                 mutex_unlock(&pCifsFile->fh_mutex);
472                 rc = 0;
473                 FreeXid(xid);
474                 return rc;
475         }
476
477         inode = pCifsFile->dentry->d_inode;
478         cifs_sb = CIFS_SB(inode->i_sb);
479         tcon = tlink_tcon(pCifsFile->tlink);
480
481 /* can not grab rename sem here because various ops, including
482    those that already have the rename sem can end up causing writepage
483    to get called and if the server was down that means we end up here,
484    and we can never tell if the caller already has the rename_sem */
485         full_path = build_path_from_dentry(pCifsFile->dentry);
486         if (full_path == NULL) {
487                 rc = -ENOMEM;
488                 mutex_unlock(&pCifsFile->fh_mutex);
489                 FreeXid(xid);
490                 return rc;
491         }
492
493         cFYI(1, "inode = 0x%p file flags 0x%x for %s",
494                  inode, pCifsFile->f_flags, full_path);
495
496         if (oplockEnabled)
497                 oplock = REQ_OPLOCK;
498         else
499                 oplock = 0;
500
501         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
502             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
503                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
504
505                 /*
506                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
507                  * original open. Must mask them off for a reopen.
508                  */
509                 unsigned int oflags = pCifsFile->f_flags &
510                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
511
512                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
513                                 cifs_sb->mnt_file_mode /* ignored */,
514                                 oflags, &oplock, &netfid, xid);
515                 if (rc == 0) {
516                         cFYI(1, "posix reopen succeeded");
517                         goto reopen_success;
518                 }
519                 /* fallthrough to retry open the old way on errors, especially
520                    in the reconnect path it is important to retry hard */
521         }
522
523         desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
524
525         /* Can not refresh inode by passing in file_info buf to be returned
526            by SMBOpen and then calling get_inode_info with returned buf
527            since file might have write behind data that needs to be flushed
528            and server version of file size can be stale. If we knew for sure
529            that inode was not dirty locally we could do this */
530
531         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
532                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
533                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
534                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
535         if (rc) {
536                 mutex_unlock(&pCifsFile->fh_mutex);
537                 cFYI(1, "cifs_open returned 0x%x", rc);
538                 cFYI(1, "oplock: %d", oplock);
539                 goto reopen_error_exit;
540         }
541
542 reopen_success:
543         pCifsFile->netfid = netfid;
544         pCifsFile->invalidHandle = false;
545         mutex_unlock(&pCifsFile->fh_mutex);
546         pCifsInode = CIFS_I(inode);
547
548         if (can_flush) {
549                 rc = filemap_write_and_wait(inode->i_mapping);
550                 mapping_set_error(inode->i_mapping, rc);
551
552                 if (tcon->unix_ext)
553                         rc = cifs_get_inode_info_unix(&inode,
554                                 full_path, inode->i_sb, xid);
555                 else
556                         rc = cifs_get_inode_info(&inode,
557                                 full_path, NULL, inode->i_sb,
558                                 xid, NULL);
559         } /* else we are writing out data to server already
560              and could deadlock if we tried to flush data, and
561              since we do not know if we have data that would
562              invalidate the current end of file on the server
563              we can not go to the server to get the new inod
564              info */
565
566         cifs_set_oplock_level(pCifsInode, oplock);
567
568         cifs_relock_file(pCifsFile);
569
570 reopen_error_exit:
571         kfree(full_path);
572         FreeXid(xid);
573         return rc;
574 }
575
576 int cifs_close(struct inode *inode, struct file *file)
577 {
578         if (file->private_data != NULL) {
579                 cifsFileInfo_put(file->private_data);
580                 file->private_data = NULL;
581         }
582
583         /* return code from the ->release op is always ignored */
584         return 0;
585 }
586
587 int cifs_closedir(struct inode *inode, struct file *file)
588 {
589         int rc = 0;
590         int xid;
591         struct cifsFileInfo *pCFileStruct = file->private_data;
592         char *ptmp;
593
594         cFYI(1, "Closedir inode = 0x%p", inode);
595
596         xid = GetXid();
597
598         if (pCFileStruct) {
599                 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
600
601                 cFYI(1, "Freeing private data in close dir");
602                 spin_lock(&cifs_file_list_lock);
603                 if (!pCFileStruct->srch_inf.endOfSearch &&
604                     !pCFileStruct->invalidHandle) {
605                         pCFileStruct->invalidHandle = true;
606                         spin_unlock(&cifs_file_list_lock);
607                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
608                         cFYI(1, "Closing uncompleted readdir with rc %d",
609                                  rc);
610                         /* not much we can do if it fails anyway, ignore rc */
611                         rc = 0;
612                 } else
613                         spin_unlock(&cifs_file_list_lock);
614                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
615                 if (ptmp) {
616                         cFYI(1, "closedir free smb buf in srch struct");
617                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
618                         if (pCFileStruct->srch_inf.smallBuf)
619                                 cifs_small_buf_release(ptmp);
620                         else
621                                 cifs_buf_release(ptmp);
622                 }
623                 cifs_put_tlink(pCFileStruct->tlink);
624                 kfree(file->private_data);
625                 file->private_data = NULL;
626         }
627         /* BB can we lock the filestruct while this is going on? */
628         FreeXid(xid);
629         return rc;
630 }
631
632 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
633                                 __u64 offset, __u8 lockType)
634 {
635         struct cifsLockInfo *li =
636                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
637         if (li == NULL)
638                 return -ENOMEM;
639         li->offset = offset;
640         li->length = len;
641         li->type = lockType;
642         mutex_lock(&fid->lock_mutex);
643         list_add(&li->llist, &fid->llist);
644         mutex_unlock(&fid->lock_mutex);
645         return 0;
646 }
647
648 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
649 {
650         int rc, xid;
651         __u32 numLock = 0;
652         __u32 numUnlock = 0;
653         __u64 length;
654         bool wait_flag = false;
655         struct cifs_sb_info *cifs_sb;
656         struct cifsTconInfo *tcon;
657         __u16 netfid;
658         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
659         bool posix_locking = 0;
660
661         length = 1 + pfLock->fl_end - pfLock->fl_start;
662         rc = -EACCES;
663         xid = GetXid();
664
665         cFYI(1, "Lock parm: 0x%x flockflags: "
666                  "0x%x flocktype: 0x%x start: %lld end: %lld",
667                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
668                 pfLock->fl_end);
669
670         if (pfLock->fl_flags & FL_POSIX)
671                 cFYI(1, "Posix");
672         if (pfLock->fl_flags & FL_FLOCK)
673                 cFYI(1, "Flock");
674         if (pfLock->fl_flags & FL_SLEEP) {
675                 cFYI(1, "Blocking lock");
676                 wait_flag = true;
677         }
678         if (pfLock->fl_flags & FL_ACCESS)
679                 cFYI(1, "Process suspended by mandatory locking - "
680                          "not implemented yet");
681         if (pfLock->fl_flags & FL_LEASE)
682                 cFYI(1, "Lease on file - not implemented yet");
683         if (pfLock->fl_flags &
684             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
685                 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
686
687         if (pfLock->fl_type == F_WRLCK) {
688                 cFYI(1, "F_WRLCK ");
689                 numLock = 1;
690         } else if (pfLock->fl_type == F_UNLCK) {
691                 cFYI(1, "F_UNLCK");
692                 numUnlock = 1;
693                 /* Check if unlock includes more than
694                 one lock range */
695         } else if (pfLock->fl_type == F_RDLCK) {
696                 cFYI(1, "F_RDLCK");
697                 lockType |= LOCKING_ANDX_SHARED_LOCK;
698                 numLock = 1;
699         } else if (pfLock->fl_type == F_EXLCK) {
700                 cFYI(1, "F_EXLCK");
701                 numLock = 1;
702         } else if (pfLock->fl_type == F_SHLCK) {
703                 cFYI(1, "F_SHLCK");
704                 lockType |= LOCKING_ANDX_SHARED_LOCK;
705                 numLock = 1;
706         } else
707                 cFYI(1, "Unknown type of lock");
708
709         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
710         tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
711         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
712
713         if ((tcon->ses->capabilities & CAP_UNIX) &&
714             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
715             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
716                 posix_locking = 1;
717         /* BB add code here to normalize offset and length to
718         account for negative length which we can not accept over the
719         wire */
720         if (IS_GETLK(cmd)) {
721                 if (posix_locking) {
722                         int posix_lock_type;
723                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
724                                 posix_lock_type = CIFS_RDLCK;
725                         else
726                                 posix_lock_type = CIFS_WRLCK;
727                         rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
728                                         length, pfLock,
729                                         posix_lock_type, wait_flag);
730                         FreeXid(xid);
731                         return rc;
732                 }
733
734                 /* BB we could chain these into one lock request BB */
735                 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
736                                  0, 1, lockType, 0 /* wait flag */, 0);
737                 if (rc == 0) {
738                         rc = CIFSSMBLock(xid, tcon, netfid, length,
739                                          pfLock->fl_start, 1 /* numUnlock */ ,
740                                          0 /* numLock */ , lockType,
741                                          0 /* wait flag */, 0);
742                         pfLock->fl_type = F_UNLCK;
743                         if (rc != 0)
744                                 cERROR(1, "Error unlocking previously locked "
745                                            "range %d during test of lock", rc);
746                         rc = 0;
747
748                 } else {
749                         /* if rc == ERR_SHARING_VIOLATION ? */
750                         rc = 0;
751
752                         if (lockType & LOCKING_ANDX_SHARED_LOCK) {
753                                 pfLock->fl_type = F_WRLCK;
754                         } else {
755                                 rc = CIFSSMBLock(xid, tcon, netfid, length,
756                                         pfLock->fl_start, 0, 1,
757                                         lockType | LOCKING_ANDX_SHARED_LOCK,
758                                         0 /* wait flag */, 0);
759                                 if (rc == 0) {
760                                         rc = CIFSSMBLock(xid, tcon, netfid,
761                                                 length, pfLock->fl_start, 1, 0,
762                                                 lockType |
763                                                 LOCKING_ANDX_SHARED_LOCK,
764                                                 0 /* wait flag */, 0);
765                                         pfLock->fl_type = F_RDLCK;
766                                         if (rc != 0)
767                                                 cERROR(1, "Error unlocking "
768                                                 "previously locked range %d "
769                                                 "during test of lock", rc);
770                                         rc = 0;
771                                 } else {
772                                         pfLock->fl_type = F_WRLCK;
773                                         rc = 0;
774                                 }
775                         }
776                 }
777
778                 FreeXid(xid);
779                 return rc;
780         }
781
782         if (!numLock && !numUnlock) {
783                 /* if no lock or unlock then nothing
784                 to do since we do not know what it is */
785                 FreeXid(xid);
786                 return -EOPNOTSUPP;
787         }
788
789         if (posix_locking) {
790                 int posix_lock_type;
791                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
792                         posix_lock_type = CIFS_RDLCK;
793                 else
794                         posix_lock_type = CIFS_WRLCK;
795
796                 if (numUnlock == 1)
797                         posix_lock_type = CIFS_UNLCK;
798
799                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
800                                       length, pfLock,
801                                       posix_lock_type, wait_flag);
802         } else {
803                 struct cifsFileInfo *fid = file->private_data;
804
805                 if (numLock) {
806                         rc = CIFSSMBLock(xid, tcon, netfid, length,
807                                          pfLock->fl_start, 0, numLock, lockType,
808                                          wait_flag, 0);
809
810                         if (rc == 0) {
811                                 /* For Windows locks we must store them. */
812                                 rc = store_file_lock(fid, length,
813                                                 pfLock->fl_start, lockType);
814                         }
815                 } else if (numUnlock) {
816                         /* For each stored lock that this unlock overlaps
817                            completely, unlock it. */
818                         int stored_rc = 0;
819                         struct cifsLockInfo *li, *tmp;
820
821                         rc = 0;
822                         mutex_lock(&fid->lock_mutex);
823                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
824                                 if (pfLock->fl_start <= li->offset &&
825                                                 (pfLock->fl_start + length) >=
826                                                 (li->offset + li->length)) {
827                                         stored_rc = CIFSSMBLock(xid, tcon,
828                                                         netfid, li->length,
829                                                         li->offset, 1, 0,
830                                                         li->type, false, 0);
831                                         if (stored_rc)
832                                                 rc = stored_rc;
833                                         else {
834                                                 list_del(&li->llist);
835                                                 kfree(li);
836                                         }
837                                 }
838                         }
839                         mutex_unlock(&fid->lock_mutex);
840                 }
841         }
842
843         if (pfLock->fl_flags & FL_POSIX)
844                 posix_lock_file_wait(file, pfLock);
845         FreeXid(xid);
846         return rc;
847 }
848
849 /* update the file size (if needed) after a write */
850 void
851 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
852                       unsigned int bytes_written)
853 {
854         loff_t end_of_write = offset + bytes_written;
855
856         if (end_of_write > cifsi->server_eof)
857                 cifsi->server_eof = end_of_write;
858 }
859
860 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
861         size_t write_size, loff_t *poffset)
862 {
863         struct inode *inode = file->f_path.dentry->d_inode;
864         int rc = 0;
865         unsigned int bytes_written = 0;
866         unsigned int total_written;
867         struct cifs_sb_info *cifs_sb;
868         struct cifsTconInfo *pTcon;
869         int xid;
870         struct cifsFileInfo *open_file;
871         struct cifsInodeInfo *cifsi = CIFS_I(inode);
872
873         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
874
875         /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
876            *poffset, file->f_path.dentry->d_name.name); */
877
878         if (file->private_data == NULL)
879                 return -EBADF;
880
881         open_file = file->private_data;
882         pTcon = tlink_tcon(open_file->tlink);
883
884         rc = generic_write_checks(file, poffset, &write_size, 0);
885         if (rc)
886                 return rc;
887
888         xid = GetXid();
889
890         for (total_written = 0; write_size > total_written;
891              total_written += bytes_written) {
892                 rc = -EAGAIN;
893                 while (rc == -EAGAIN) {
894                         if (file->private_data == NULL) {
895                                 /* file has been closed on us */
896                                 FreeXid(xid);
897                         /* if we have gotten here we have written some data
898                            and blocked, and the file has been freed on us while
899                            we blocked so return what we managed to write */
900                                 return total_written;
901                         }
902                         if (open_file->invalidHandle) {
903                                 /* we could deadlock if we called
904                                    filemap_fdatawait from here so tell
905                                    reopen_file not to flush data to server
906                                    now */
907                                 rc = cifs_reopen_file(open_file, false);
908                                 if (rc != 0)
909                                         break;
910                         }
911
912                         rc = CIFSSMBWrite(xid, pTcon,
913                                 open_file->netfid,
914                                 min_t(const int, cifs_sb->wsize,
915                                       write_size - total_written),
916                                 *poffset, &bytes_written,
917                                 NULL, write_data + total_written, 0);
918                 }
919                 if (rc || (bytes_written == 0)) {
920                         if (total_written)
921                                 break;
922                         else {
923                                 FreeXid(xid);
924                                 return rc;
925                         }
926                 } else {
927                         cifs_update_eof(cifsi, *poffset, bytes_written);
928                         *poffset += bytes_written;
929                 }
930         }
931
932         cifs_stats_bytes_written(pTcon, total_written);
933
934 /* Do not update local mtime - server will set its actual value on write
935  *      inode->i_ctime = inode->i_mtime =
936  *              current_fs_time(inode->i_sb);*/
937         if (total_written > 0) {
938                 spin_lock(&inode->i_lock);
939                 if (*poffset > inode->i_size)
940                         i_size_write(inode, *poffset);
941                 spin_unlock(&inode->i_lock);
942         }
943         mark_inode_dirty_sync(inode);
944
945         FreeXid(xid);
946         return total_written;
947 }
948
949 static ssize_t cifs_write(struct cifsFileInfo *open_file,
950                           const char *write_data, size_t write_size,
951                           loff_t *poffset)
952 {
953         int rc = 0;
954         unsigned int bytes_written = 0;
955         unsigned int total_written;
956         struct cifs_sb_info *cifs_sb;
957         struct cifsTconInfo *pTcon;
958         int xid;
959         struct dentry *dentry = open_file->dentry;
960         struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
961
962         cifs_sb = CIFS_SB(dentry->d_sb);
963
964         cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
965            *poffset, dentry->d_name.name);
966
967         pTcon = tlink_tcon(open_file->tlink);
968
969         xid = GetXid();
970
971         for (total_written = 0; write_size > total_written;
972              total_written += bytes_written) {
973                 rc = -EAGAIN;
974                 while (rc == -EAGAIN) {
975                         struct kvec iov[2];
976                         unsigned int len;
977
978                         if (open_file->invalidHandle) {
979                                 /* we could deadlock if we called
980                                    filemap_fdatawait from here so tell
981                                    reopen_file not to flush data to
982                                    server now */
983                                 rc = cifs_reopen_file(open_file, false);
984                                 if (rc != 0)
985                                         break;
986                         }
987
988                         len = min((size_t)cifs_sb->wsize,
989                                   write_size - total_written);
990                         /* iov[0] is reserved for smb header */
991                         iov[1].iov_base = (char *)write_data + total_written;
992                         iov[1].iov_len = len;
993                         rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid, len,
994                                            *poffset, &bytes_written, iov, 1, 0);
995                 }
996                 if (rc || (bytes_written == 0)) {
997                         if (total_written)
998                                 break;
999                         else {
1000                                 FreeXid(xid);
1001                                 return rc;
1002                         }
1003                 } else {
1004                         cifs_update_eof(cifsi, *poffset, bytes_written);
1005                         *poffset += bytes_written;
1006                 }
1007         }
1008
1009         cifs_stats_bytes_written(pTcon, total_written);
1010
1011         if (total_written > 0) {
1012                 spin_lock(&dentry->d_inode->i_lock);
1013                 if (*poffset > dentry->d_inode->i_size)
1014                         i_size_write(dentry->d_inode, *poffset);
1015                 spin_unlock(&dentry->d_inode->i_lock);
1016         }
1017         mark_inode_dirty_sync(dentry->d_inode);
1018         FreeXid(xid);
1019         return total_written;
1020 }
1021
1022 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1023                                         bool fsuid_only)
1024 {
1025         struct cifsFileInfo *open_file = NULL;
1026         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1027
1028         /* only filter by fsuid on multiuser mounts */
1029         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1030                 fsuid_only = false;
1031
1032         spin_lock(&cifs_file_list_lock);
1033         /* we could simply get the first_list_entry since write-only entries
1034            are always at the end of the list but since the first entry might
1035            have a close pending, we go through the whole list */
1036         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1037                 if (fsuid_only && open_file->uid != current_fsuid())
1038                         continue;
1039                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1040                         if (!open_file->invalidHandle) {
1041                                 /* found a good file */
1042                                 /* lock it so it will not be closed on us */
1043                                 cifsFileInfo_get(open_file);
1044                                 spin_unlock(&cifs_file_list_lock);
1045                                 return open_file;
1046                         } /* else might as well continue, and look for
1047                              another, or simply have the caller reopen it
1048                              again rather than trying to fix this handle */
1049                 } else /* write only file */
1050                         break; /* write only files are last so must be done */
1051         }
1052         spin_unlock(&cifs_file_list_lock);
1053         return NULL;
1054 }
1055
1056 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1057                                         bool fsuid_only)
1058 {
1059         struct cifsFileInfo *open_file;
1060         struct cifs_sb_info *cifs_sb;
1061         bool any_available = false;
1062         int rc;
1063
1064         /* Having a null inode here (because mapping->host was set to zero by
1065         the VFS or MM) should not happen but we had reports of on oops (due to
1066         it being zero) during stress testcases so we need to check for it */
1067
1068         if (cifs_inode == NULL) {
1069                 cERROR(1, "Null inode passed to cifs_writeable_file");
1070                 dump_stack();
1071                 return NULL;
1072         }
1073
1074         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1075
1076         /* only filter by fsuid on multiuser mounts */
1077         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1078                 fsuid_only = false;
1079
1080         spin_lock(&cifs_file_list_lock);
1081 refind_writable:
1082         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1083                 if (!any_available && open_file->pid != current->tgid)
1084                         continue;
1085                 if (fsuid_only && open_file->uid != current_fsuid())
1086                         continue;
1087                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1088                         cifsFileInfo_get(open_file);
1089
1090                         if (!open_file->invalidHandle) {
1091                                 /* found a good writable file */
1092                                 spin_unlock(&cifs_file_list_lock);
1093                                 return open_file;
1094                         }
1095
1096                         spin_unlock(&cifs_file_list_lock);
1097
1098                         /* Had to unlock since following call can block */
1099                         rc = cifs_reopen_file(open_file, false);
1100                         if (!rc)
1101                                 return open_file;
1102
1103                         /* if it fails, try another handle if possible */
1104                         cFYI(1, "wp failed on reopen file");
1105                         cifsFileInfo_put(open_file);
1106
1107                         spin_lock(&cifs_file_list_lock);
1108
1109                         /* else we simply continue to the next entry. Thus
1110                            we do not loop on reopen errors.  If we
1111                            can not reopen the file, for example if we
1112                            reconnected to a server with another client
1113                            racing to delete or lock the file we would not
1114                            make progress if we restarted before the beginning
1115                            of the loop here. */
1116                 }
1117         }
1118         /* couldn't find useable FH with same pid, try any available */
1119         if (!any_available) {
1120                 any_available = true;
1121                 goto refind_writable;
1122         }
1123         spin_unlock(&cifs_file_list_lock);
1124         return NULL;
1125 }
1126
1127 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1128 {
1129         struct address_space *mapping = page->mapping;
1130         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1131         char *write_data;
1132         int rc = -EFAULT;
1133         int bytes_written = 0;
1134         struct inode *inode;
1135         struct cifsFileInfo *open_file;
1136
1137         if (!mapping || !mapping->host)
1138                 return -EFAULT;
1139
1140         inode = page->mapping->host;
1141
1142         offset += (loff_t)from;
1143         write_data = kmap(page);
1144         write_data += from;
1145
1146         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1147                 kunmap(page);
1148                 return -EIO;
1149         }
1150
1151         /* racing with truncate? */
1152         if (offset > mapping->host->i_size) {
1153                 kunmap(page);
1154                 return 0; /* don't care */
1155         }
1156
1157         /* check to make sure that we are not extending the file */
1158         if (mapping->host->i_size - offset < (loff_t)to)
1159                 to = (unsigned)(mapping->host->i_size - offset);
1160
1161         open_file = find_writable_file(CIFS_I(mapping->host), false);
1162         if (open_file) {
1163                 bytes_written = cifs_write(open_file, write_data,
1164                                            to - from, &offset);
1165                 cifsFileInfo_put(open_file);
1166                 /* Does mm or vfs already set times? */
1167                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1168                 if ((bytes_written > 0) && (offset))
1169                         rc = 0;
1170                 else if (bytes_written < 0)
1171                         rc = bytes_written;
1172         } else {
1173                 cFYI(1, "No writeable filehandles for inode");
1174                 rc = -EIO;
1175         }
1176
1177         kunmap(page);
1178         return rc;
1179 }
1180
1181 static int cifs_writepages(struct address_space *mapping,
1182                            struct writeback_control *wbc)
1183 {
1184         unsigned int bytes_to_write;
1185         unsigned int bytes_written;
1186         struct cifs_sb_info *cifs_sb;
1187         int done = 0;
1188         pgoff_t end;
1189         pgoff_t index;
1190         int range_whole = 0;
1191         struct kvec *iov;
1192         int len;
1193         int n_iov = 0;
1194         pgoff_t next;
1195         int nr_pages;
1196         __u64 offset = 0;
1197         struct cifsFileInfo *open_file;
1198         struct cifsTconInfo *tcon;
1199         struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1200         struct page *page;
1201         struct pagevec pvec;
1202         int rc = 0;
1203         int scanned = 0;
1204         int xid;
1205
1206         cifs_sb = CIFS_SB(mapping->host->i_sb);
1207
1208         /*
1209          * If wsize is smaller that the page cache size, default to writing
1210          * one page at a time via cifs_writepage
1211          */
1212         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1213                 return generic_writepages(mapping, wbc);
1214
1215         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1216         if (iov == NULL)
1217                 return generic_writepages(mapping, wbc);
1218
1219         /*
1220          * if there's no open file, then this is likely to fail too,
1221          * but it'll at least handle the return. Maybe it should be
1222          * a BUG() instead?
1223          */
1224         open_file = find_writable_file(CIFS_I(mapping->host), false);
1225         if (!open_file) {
1226                 kfree(iov);
1227                 return generic_writepages(mapping, wbc);
1228         }
1229
1230         tcon = tlink_tcon(open_file->tlink);
1231         cifsFileInfo_put(open_file);
1232
1233         xid = GetXid();
1234
1235         pagevec_init(&pvec, 0);
1236         if (wbc->range_cyclic) {
1237                 index = mapping->writeback_index; /* Start from prev offset */
1238                 end = -1;
1239         } else {
1240                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1241                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1242                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1243                         range_whole = 1;
1244                 scanned = 1;
1245         }
1246 retry:
1247         while (!done && (index <= end) &&
1248                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1249                         PAGECACHE_TAG_DIRTY,
1250                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1251                 int first;
1252                 unsigned int i;
1253
1254                 first = -1;
1255                 next = 0;
1256                 n_iov = 0;
1257                 bytes_to_write = 0;
1258
1259                 for (i = 0; i < nr_pages; i++) {
1260                         page = pvec.pages[i];
1261                         /*
1262                          * At this point we hold neither mapping->tree_lock nor
1263                          * lock on the page itself: the page may be truncated or
1264                          * invalidated (changing page->mapping to NULL), or even
1265                          * swizzled back from swapper_space to tmpfs file
1266                          * mapping
1267                          */
1268
1269                         if (first < 0)
1270                                 lock_page(page);
1271                         else if (!trylock_page(page))
1272                                 break;
1273
1274                         if (unlikely(page->mapping != mapping)) {
1275                                 unlock_page(page);
1276                                 break;
1277                         }
1278
1279                         if (!wbc->range_cyclic && page->index > end) {
1280                                 done = 1;
1281                                 unlock_page(page);
1282                                 break;
1283                         }
1284
1285                         if (next && (page->index != next)) {
1286                                 /* Not next consecutive page */
1287                                 unlock_page(page);
1288                                 break;
1289                         }
1290
1291                         if (wbc->sync_mode != WB_SYNC_NONE)
1292                                 wait_on_page_writeback(page);
1293
1294                         if (PageWriteback(page) ||
1295                                         !clear_page_dirty_for_io(page)) {
1296                                 unlock_page(page);
1297                                 break;
1298                         }
1299
1300                         /*
1301                          * This actually clears the dirty bit in the radix tree.
1302                          * See cifs_writepage() for more commentary.
1303                          */
1304                         set_page_writeback(page);
1305
1306                         if (page_offset(page) >= mapping->host->i_size) {
1307                                 done = 1;
1308                                 unlock_page(page);
1309                                 end_page_writeback(page);
1310                                 break;
1311                         }
1312
1313                         /*
1314                          * BB can we get rid of this?  pages are held by pvec
1315                          */
1316                         page_cache_get(page);
1317
1318                         len = min(mapping->host->i_size - page_offset(page),
1319                                   (loff_t)PAGE_CACHE_SIZE);
1320
1321                         /* reserve iov[0] for the smb header */
1322                         n_iov++;
1323                         iov[n_iov].iov_base = kmap(page);
1324                         iov[n_iov].iov_len = len;
1325                         bytes_to_write += len;
1326
1327                         if (first < 0) {
1328                                 first = i;
1329                                 offset = page_offset(page);
1330                         }
1331                         next = page->index + 1;
1332                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1333                                 break;
1334                 }
1335                 if (n_iov) {
1336 retry_write:
1337                         open_file = find_writable_file(CIFS_I(mapping->host),
1338                                                         false);
1339                         if (!open_file) {
1340                                 cERROR(1, "No writable handles for inode");
1341                                 rc = -EBADF;
1342                         } else {
1343                                 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1344                                                    bytes_to_write, offset,
1345                                                    &bytes_written, iov, n_iov,
1346                                                    0);
1347                                 cifsFileInfo_put(open_file);
1348                         }
1349
1350                         cFYI(1, "Write2 rc=%d, wrote=%u", rc, bytes_written);
1351
1352                         /*
1353                          * For now, treat a short write as if nothing got
1354                          * written. A zero length write however indicates
1355                          * ENOSPC or EFBIG. We have no way to know which
1356                          * though, so call it ENOSPC for now. EFBIG would
1357                          * get translated to AS_EIO anyway.
1358                          *
1359                          * FIXME: make it take into account the data that did
1360                          *        get written
1361                          */
1362                         if (rc == 0) {
1363                                 if (bytes_written == 0)
1364                                         rc = -ENOSPC;
1365                                 else if (bytes_written < bytes_to_write)
1366                                         rc = -EAGAIN;
1367                         }
1368
1369                         /* retry on data-integrity flush */
1370                         if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
1371                                 goto retry_write;
1372
1373                         /* fix the stats and EOF */
1374                         if (bytes_written > 0) {
1375                                 cifs_stats_bytes_written(tcon, bytes_written);
1376                                 cifs_update_eof(cifsi, offset, bytes_written);
1377                         }
1378
1379                         for (i = 0; i < n_iov; i++) {
1380                                 page = pvec.pages[first + i];
1381                                 /* on retryable write error, redirty page */
1382                                 if (rc == -EAGAIN)
1383                                         redirty_page_for_writepage(wbc, page);
1384                                 else if (rc != 0)
1385                                         SetPageError(page);
1386                                 kunmap(page);
1387                                 unlock_page(page);
1388                                 end_page_writeback(page);
1389                                 page_cache_release(page);
1390                         }
1391
1392                         if (rc != -EAGAIN)
1393                                 mapping_set_error(mapping, rc);
1394                         else
1395                                 rc = 0;
1396
1397                         if ((wbc->nr_to_write -= n_iov) <= 0)
1398                                 done = 1;
1399                         index = next;
1400                 } else
1401                         /* Need to re-find the pages we skipped */
1402                         index = pvec.pages[0]->index + 1;
1403
1404                 pagevec_release(&pvec);
1405         }
1406         if (!scanned && !done) {
1407                 /*
1408                  * We hit the last page and there is more work to be done: wrap
1409                  * back to the start of the file
1410                  */
1411                 scanned = 1;
1412                 index = 0;
1413                 goto retry;
1414         }
1415         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1416                 mapping->writeback_index = index;
1417
1418         FreeXid(xid);
1419         kfree(iov);
1420         return rc;
1421 }
1422
1423 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1424 {
1425         int rc = -EFAULT;
1426         int xid;
1427
1428         xid = GetXid();
1429 /* BB add check for wbc flags */
1430         page_cache_get(page);
1431         if (!PageUptodate(page))
1432                 cFYI(1, "ppw - page not up to date");
1433
1434         /*
1435          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1436          *
1437          * A writepage() implementation always needs to do either this,
1438          * or re-dirty the page with "redirty_page_for_writepage()" in
1439          * the case of a failure.
1440          *
1441          * Just unlocking the page will cause the radix tree tag-bits
1442          * to fail to update with the state of the page correctly.
1443          */
1444         set_page_writeback(page);
1445         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1446         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1447         unlock_page(page);
1448         end_page_writeback(page);
1449         page_cache_release(page);
1450         FreeXid(xid);
1451         return rc;
1452 }
1453
1454 static int cifs_write_end(struct file *file, struct address_space *mapping,
1455                         loff_t pos, unsigned len, unsigned copied,
1456                         struct page *page, void *fsdata)
1457 {
1458         int rc;
1459         struct inode *inode = mapping->host;
1460
1461         cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1462                  page, pos, copied);
1463
1464         if (PageChecked(page)) {
1465                 if (copied == len)
1466                         SetPageUptodate(page);
1467                 ClearPageChecked(page);
1468         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1469                 SetPageUptodate(page);
1470
1471         if (!PageUptodate(page)) {
1472                 char *page_data;
1473                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1474                 int xid;
1475
1476                 xid = GetXid();
1477                 /* this is probably better than directly calling
1478                    partialpage_write since in this function the file handle is
1479                    known which we might as well leverage */
1480                 /* BB check if anything else missing out of ppw
1481                    such as updating last write time */
1482                 page_data = kmap(page);
1483                 rc = cifs_write(file->private_data, page_data + offset,
1484                                 copied, &pos);
1485                 /* if (rc < 0) should we set writebehind rc? */
1486                 kunmap(page);
1487
1488                 FreeXid(xid);
1489         } else {
1490                 rc = copied;
1491                 pos += copied;
1492                 set_page_dirty(page);
1493         }
1494
1495         if (rc > 0) {
1496                 spin_lock(&inode->i_lock);
1497                 if (pos > inode->i_size)
1498                         i_size_write(inode, pos);
1499                 spin_unlock(&inode->i_lock);
1500         }
1501
1502         unlock_page(page);
1503         page_cache_release(page);
1504
1505         return rc;
1506 }
1507
1508 int cifs_strict_fsync(struct file *file, int datasync)
1509 {
1510         int xid;
1511         int rc = 0;
1512         struct cifsTconInfo *tcon;
1513         struct cifsFileInfo *smbfile = file->private_data;
1514         struct inode *inode = file->f_path.dentry->d_inode;
1515         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1516
1517         xid = GetXid();
1518
1519         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1520                 file->f_path.dentry->d_name.name, datasync);
1521
1522         if (!CIFS_I(inode)->clientCanCacheRead)
1523                 cifs_invalidate_mapping(inode);
1524
1525         tcon = tlink_tcon(smbfile->tlink);
1526         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1527                 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1528
1529         FreeXid(xid);
1530         return rc;
1531 }
1532
1533 int cifs_fsync(struct file *file, int datasync)
1534 {
1535         int xid;
1536         int rc = 0;
1537         struct cifsTconInfo *tcon;
1538         struct cifsFileInfo *smbfile = file->private_data;
1539         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1540
1541         xid = GetXid();
1542
1543         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1544                 file->f_path.dentry->d_name.name, datasync);
1545
1546         tcon = tlink_tcon(smbfile->tlink);
1547         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1548                 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1549
1550         FreeXid(xid);
1551         return rc;
1552 }
1553
1554 /*
1555  * As file closes, flush all cached write data for this inode checking
1556  * for write behind errors.
1557  */
1558 int cifs_flush(struct file *file, fl_owner_t id)
1559 {
1560         struct inode *inode = file->f_path.dentry->d_inode;
1561         int rc = 0;
1562
1563         if (file->f_mode & FMODE_WRITE)
1564                 rc = filemap_write_and_wait(inode->i_mapping);
1565
1566         cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1567
1568         return rc;
1569 }
1570
1571 static int
1572 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1573 {
1574         int rc = 0;
1575         unsigned long i;
1576
1577         for (i = 0; i < num_pages; i++) {
1578                 pages[i] = alloc_page(__GFP_HIGHMEM);
1579                 if (!pages[i]) {
1580                         /*
1581                          * save number of pages we have already allocated and
1582                          * return with ENOMEM error
1583                          */
1584                         num_pages = i;
1585                         rc = -ENOMEM;
1586                         goto error;
1587                 }
1588         }
1589
1590         return rc;
1591
1592 error:
1593         for (i = 0; i < num_pages; i++)
1594                 put_page(pages[i]);
1595         return rc;
1596 }
1597
1598 static inline
1599 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1600 {
1601         size_t num_pages;
1602         size_t clen;
1603
1604         clen = min_t(const size_t, len, wsize);
1605         num_pages = clen / PAGE_CACHE_SIZE;
1606         if (clen % PAGE_CACHE_SIZE)
1607                 num_pages++;
1608
1609         if (cur_len)
1610                 *cur_len = clen;
1611
1612         return num_pages;
1613 }
1614
1615 static ssize_t
1616 cifs_iovec_write(struct file *file, const struct iovec *iov,
1617                  unsigned long nr_segs, loff_t *poffset)
1618 {
1619         unsigned int written;
1620         unsigned long num_pages, npages, i;
1621         size_t copied, len, cur_len;
1622         ssize_t total_written = 0;
1623         struct kvec *to_send;
1624         struct page **pages;
1625         struct iov_iter it;
1626         struct inode *inode;
1627         struct cifsFileInfo *open_file;
1628         struct cifsTconInfo *pTcon;
1629         struct cifs_sb_info *cifs_sb;
1630         int xid, rc;
1631
1632         len = iov_length(iov, nr_segs);
1633         if (!len)
1634                 return 0;
1635
1636         rc = generic_write_checks(file, poffset, &len, 0);
1637         if (rc)
1638                 return rc;
1639
1640         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1641         num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1642
1643         pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1644         if (!pages)
1645                 return -ENOMEM;
1646
1647         to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1648         if (!to_send) {
1649                 kfree(pages);
1650                 return -ENOMEM;
1651         }
1652
1653         rc = cifs_write_allocate_pages(pages, num_pages);
1654         if (rc) {
1655                 kfree(pages);
1656                 kfree(to_send);
1657                 return rc;
1658         }
1659
1660         xid = GetXid();
1661         open_file = file->private_data;
1662         pTcon = tlink_tcon(open_file->tlink);
1663         inode = file->f_path.dentry->d_inode;
1664
1665         iov_iter_init(&it, iov, nr_segs, len, 0);
1666         npages = num_pages;
1667
1668         do {
1669                 size_t save_len = cur_len;
1670                 for (i = 0; i < npages; i++) {
1671                         copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1672                         copied = iov_iter_copy_from_user(pages[i], &it, 0,
1673                                                          copied);
1674                         cur_len -= copied;
1675                         iov_iter_advance(&it, copied);
1676                         to_send[i+1].iov_base = kmap(pages[i]);
1677                         to_send[i+1].iov_len = copied;
1678                 }
1679
1680                 cur_len = save_len - cur_len;
1681
1682                 do {
1683                         if (open_file->invalidHandle) {
1684                                 rc = cifs_reopen_file(open_file, false);
1685                                 if (rc != 0)
1686                                         break;
1687                         }
1688                         rc = CIFSSMBWrite2(xid, pTcon, open_file->netfid,
1689                                            cur_len, *poffset, &written,
1690                                            to_send, npages, 0);
1691                 } while (rc == -EAGAIN);
1692
1693                 for (i = 0; i < npages; i++)
1694                         kunmap(pages[i]);
1695
1696                 if (written) {
1697                         len -= written;
1698                         total_written += written;
1699                         cifs_update_eof(CIFS_I(inode), *poffset, written);
1700                         *poffset += written;
1701                 } else if (rc < 0) {
1702                         if (!total_written)
1703                                 total_written = rc;
1704                         break;
1705                 }
1706
1707                 /* get length and number of kvecs of the next write */
1708                 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1709         } while (len > 0);
1710
1711         if (total_written > 0) {
1712                 spin_lock(&inode->i_lock);
1713                 if (*poffset > inode->i_size)
1714                         i_size_write(inode, *poffset);
1715                 spin_unlock(&inode->i_lock);
1716         }
1717
1718         cifs_stats_bytes_written(pTcon, total_written);
1719         mark_inode_dirty_sync(inode);
1720
1721         for (i = 0; i < num_pages; i++)
1722                 put_page(pages[i]);
1723         kfree(to_send);
1724         kfree(pages);
1725         FreeXid(xid);
1726         return total_written;
1727 }
1728
1729 static ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1730                                 unsigned long nr_segs, loff_t pos)
1731 {
1732         ssize_t written;
1733         struct inode *inode;
1734
1735         inode = iocb->ki_filp->f_path.dentry->d_inode;
1736
1737         /*
1738          * BB - optimize the way when signing is disabled. We can drop this
1739          * extra memory-to-memory copying and use iovec buffers for constructing
1740          * write request.
1741          */
1742
1743         written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1744         if (written > 0) {
1745                 CIFS_I(inode)->invalid_mapping = true;
1746                 iocb->ki_pos = pos;
1747         }
1748
1749         return written;
1750 }
1751
1752 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1753                            unsigned long nr_segs, loff_t pos)
1754 {
1755         struct inode *inode;
1756
1757         inode = iocb->ki_filp->f_path.dentry->d_inode;
1758
1759         if (CIFS_I(inode)->clientCanCacheAll)
1760                 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1761
1762         /*
1763          * In strict cache mode we need to write the data to the server exactly
1764          * from the pos to pos+len-1 rather than flush all affected pages
1765          * because it may cause a error with mandatory locks on these pages but
1766          * not on the region from pos to ppos+len-1.
1767          */
1768
1769         return cifs_user_writev(iocb, iov, nr_segs, pos);
1770 }
1771
1772 static ssize_t
1773 cifs_iovec_read(struct file *file, const struct iovec *iov,
1774                  unsigned long nr_segs, loff_t *poffset)
1775 {
1776         int rc;
1777         int xid;
1778         ssize_t total_read;
1779         unsigned int bytes_read = 0;
1780         size_t len, cur_len;
1781         int iov_offset = 0;
1782         struct cifs_sb_info *cifs_sb;
1783         struct cifsTconInfo *pTcon;
1784         struct cifsFileInfo *open_file;
1785         struct smb_com_read_rsp *pSMBr;
1786         char *read_data;
1787
1788         if (!nr_segs)
1789                 return 0;
1790
1791         len = iov_length(iov, nr_segs);
1792         if (!len)
1793                 return 0;
1794
1795         xid = GetXid();
1796         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1797
1798         open_file = file->private_data;
1799         pTcon = tlink_tcon(open_file->tlink);
1800
1801         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1802                 cFYI(1, "attempting read on write only file instance");
1803
1804         for (total_read = 0; total_read < len; total_read += bytes_read) {
1805                 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1806                 rc = -EAGAIN;
1807                 read_data = NULL;
1808
1809                 while (rc == -EAGAIN) {
1810                         int buf_type = CIFS_NO_BUFFER;
1811                         if (open_file->invalidHandle) {
1812                                 rc = cifs_reopen_file(open_file, true);
1813                                 if (rc != 0)
1814                                         break;
1815                         }
1816                         rc = CIFSSMBRead(xid, pTcon, open_file->netfid,
1817                                          cur_len, *poffset, &bytes_read,
1818                                          &read_data, &buf_type);
1819                         pSMBr = (struct smb_com_read_rsp *)read_data;
1820                         if (read_data) {
1821                                 char *data_offset = read_data + 4 +
1822                                                 le16_to_cpu(pSMBr->DataOffset);
1823                                 if (memcpy_toiovecend(iov, data_offset,
1824                                                       iov_offset, bytes_read))
1825                                         rc = -EFAULT;
1826                                 if (buf_type == CIFS_SMALL_BUFFER)
1827                                         cifs_small_buf_release(read_data);
1828                                 else if (buf_type == CIFS_LARGE_BUFFER)
1829                                         cifs_buf_release(read_data);
1830                                 read_data = NULL;
1831                                 iov_offset += bytes_read;
1832                         }
1833                 }
1834
1835                 if (rc || (bytes_read == 0)) {
1836                         if (total_read) {
1837                                 break;
1838                         } else {
1839                                 FreeXid(xid);
1840                                 return rc;
1841                         }
1842                 } else {
1843                         cifs_stats_bytes_read(pTcon, bytes_read);
1844                         *poffset += bytes_read;
1845                 }
1846         }
1847
1848         FreeXid(xid);
1849         return total_read;
1850 }
1851
1852 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1853                        size_t read_size, loff_t *poffset)
1854 {
1855         struct iovec iov;
1856         iov.iov_base = read_data;
1857         iov.iov_len = read_size;
1858
1859         return cifs_iovec_read(file, &iov, 1, poffset);
1860 }
1861
1862 static ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1863                                unsigned long nr_segs, loff_t pos)
1864 {
1865         ssize_t read;
1866
1867         read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1868         if (read > 0)
1869                 iocb->ki_pos = pos;
1870
1871         return read;
1872 }
1873
1874 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1875                           unsigned long nr_segs, loff_t pos)
1876 {
1877         struct inode *inode;
1878
1879         inode = iocb->ki_filp->f_path.dentry->d_inode;
1880
1881         if (CIFS_I(inode)->clientCanCacheRead)
1882                 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1883
1884         /*
1885          * In strict cache mode we need to read from the server all the time
1886          * if we don't have level II oplock because the server can delay mtime
1887          * change - so we can't make a decision about inode invalidating.
1888          * And we can also fail with pagereading if there are mandatory locks
1889          * on pages affected by this read but not on the region from pos to
1890          * pos+len-1.
1891          */
1892
1893         return cifs_user_readv(iocb, iov, nr_segs, pos);
1894 }
1895
1896 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1897                          loff_t *poffset)
1898 {
1899         int rc = -EACCES;
1900         unsigned int bytes_read = 0;
1901         unsigned int total_read;
1902         unsigned int current_read_size;
1903         struct cifs_sb_info *cifs_sb;
1904         struct cifsTconInfo *pTcon;
1905         int xid;
1906         char *current_offset;
1907         struct cifsFileInfo *open_file;
1908         int buf_type = CIFS_NO_BUFFER;
1909
1910         xid = GetXid();
1911         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1912
1913         if (file->private_data == NULL) {
1914                 rc = -EBADF;
1915                 FreeXid(xid);
1916                 return rc;
1917         }
1918         open_file = file->private_data;
1919         pTcon = tlink_tcon(open_file->tlink);
1920
1921         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1922                 cFYI(1, "attempting read on write only file instance");
1923
1924         for (total_read = 0, current_offset = read_data;
1925              read_size > total_read;
1926              total_read += bytes_read, current_offset += bytes_read) {
1927                 current_read_size = min_t(const int, read_size - total_read,
1928                                           cifs_sb->rsize);
1929                 /* For windows me and 9x we do not want to request more
1930                 than it negotiated since it will refuse the read then */
1931                 if ((pTcon->ses) &&
1932                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1933                         current_read_size = min_t(const int, current_read_size,
1934                                         pTcon->ses->server->maxBuf - 128);
1935                 }
1936                 rc = -EAGAIN;
1937                 while (rc == -EAGAIN) {
1938                         if (open_file->invalidHandle) {
1939                                 rc = cifs_reopen_file(open_file, true);
1940                                 if (rc != 0)
1941                                         break;
1942                         }
1943                         rc = CIFSSMBRead(xid, pTcon,
1944                                          open_file->netfid,
1945                                          current_read_size, *poffset,
1946                                          &bytes_read, &current_offset,
1947                                          &buf_type);
1948                 }
1949                 if (rc || (bytes_read == 0)) {
1950                         if (total_read) {
1951                                 break;
1952                         } else {
1953                                 FreeXid(xid);
1954                                 return rc;
1955                         }
1956                 } else {
1957                         cifs_stats_bytes_read(pTcon, total_read);
1958                         *poffset += bytes_read;
1959                 }
1960         }
1961         FreeXid(xid);
1962         return total_read;
1963 }
1964
1965 /*
1966  * If the page is mmap'ed into a process' page tables, then we need to make
1967  * sure that it doesn't change while being written back.
1968  */
1969 static int
1970 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1971 {
1972         struct page *page = vmf->page;
1973
1974         lock_page(page);
1975         return VM_FAULT_LOCKED;
1976 }
1977
1978 static struct vm_operations_struct cifs_file_vm_ops = {
1979         .fault = filemap_fault,
1980         .page_mkwrite = cifs_page_mkwrite,
1981 };
1982
1983 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
1984 {
1985         int rc, xid;
1986         struct inode *inode = file->f_path.dentry->d_inode;
1987
1988         xid = GetXid();
1989
1990         if (!CIFS_I(inode)->clientCanCacheRead)
1991                 cifs_invalidate_mapping(inode);
1992
1993         rc = generic_file_mmap(file, vma);
1994         if (rc == 0)
1995                 vma->vm_ops = &cifs_file_vm_ops;
1996         FreeXid(xid);
1997         return rc;
1998 }
1999
2000 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2001 {
2002         int rc, xid;
2003
2004         xid = GetXid();
2005         rc = cifs_revalidate_file(file);
2006         if (rc) {
2007                 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2008                 FreeXid(xid);
2009                 return rc;
2010         }
2011         rc = generic_file_mmap(file, vma);
2012         if (rc == 0)
2013                 vma->vm_ops = &cifs_file_vm_ops;
2014         FreeXid(xid);
2015         return rc;
2016 }
2017
2018
2019 static void cifs_copy_cache_pages(struct address_space *mapping,
2020         struct list_head *pages, int bytes_read, char *data)
2021 {
2022         struct page *page;
2023         char *target;
2024
2025         while (bytes_read > 0) {
2026                 if (list_empty(pages))
2027                         break;
2028
2029                 page = list_entry(pages->prev, struct page, lru);
2030                 list_del(&page->lru);
2031
2032                 if (add_to_page_cache_lru(page, mapping, page->index,
2033                                       GFP_KERNEL)) {
2034                         page_cache_release(page);
2035                         cFYI(1, "Add page cache failed");
2036                         data += PAGE_CACHE_SIZE;
2037                         bytes_read -= PAGE_CACHE_SIZE;
2038                         continue;
2039                 }
2040                 page_cache_release(page);
2041
2042                 target = kmap_atomic(page, KM_USER0);
2043
2044                 if (PAGE_CACHE_SIZE > bytes_read) {
2045                         memcpy(target, data, bytes_read);
2046                         /* zero the tail end of this partial page */
2047                         memset(target + bytes_read, 0,
2048                                PAGE_CACHE_SIZE - bytes_read);
2049                         bytes_read = 0;
2050                 } else {
2051                         memcpy(target, data, PAGE_CACHE_SIZE);
2052                         bytes_read -= PAGE_CACHE_SIZE;
2053                 }
2054                 kunmap_atomic(target, KM_USER0);
2055
2056                 flush_dcache_page(page);
2057                 SetPageUptodate(page);
2058                 unlock_page(page);
2059                 data += PAGE_CACHE_SIZE;
2060
2061                 /* add page to FS-Cache */
2062                 cifs_readpage_to_fscache(mapping->host, page);
2063         }
2064         return;
2065 }
2066
2067 static int cifs_readpages(struct file *file, struct address_space *mapping,
2068         struct list_head *page_list, unsigned num_pages)
2069 {
2070         int rc = -EACCES;
2071         int xid;
2072         loff_t offset;
2073         struct page *page;
2074         struct cifs_sb_info *cifs_sb;
2075         struct cifsTconInfo *pTcon;
2076         unsigned int bytes_read = 0;
2077         unsigned int read_size, i;
2078         char *smb_read_data = NULL;
2079         struct smb_com_read_rsp *pSMBr;
2080         struct cifsFileInfo *open_file;
2081         int buf_type = CIFS_NO_BUFFER;
2082
2083         xid = GetXid();
2084         if (file->private_data == NULL) {
2085                 rc = -EBADF;
2086                 FreeXid(xid);
2087                 return rc;
2088         }
2089         open_file = file->private_data;
2090         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2091         pTcon = tlink_tcon(open_file->tlink);
2092
2093         /*
2094          * Reads as many pages as possible from fscache. Returns -ENOBUFS
2095          * immediately if the cookie is negative
2096          */
2097         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2098                                          &num_pages);
2099         if (rc == 0)
2100                 goto read_complete;
2101
2102         cFYI(DBG2, "rpages: num pages %d", num_pages);
2103         for (i = 0; i < num_pages; ) {
2104                 unsigned contig_pages;
2105                 struct page *tmp_page;
2106                 unsigned long expected_index;
2107
2108                 if (list_empty(page_list))
2109                         break;
2110
2111                 page = list_entry(page_list->prev, struct page, lru);
2112                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2113
2114                 /* count adjacent pages that we will read into */
2115                 contig_pages = 0;
2116                 expected_index =
2117                         list_entry(page_list->prev, struct page, lru)->index;
2118                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2119                         if (tmp_page->index == expected_index) {
2120                                 contig_pages++;
2121                                 expected_index++;
2122                         } else
2123                                 break;
2124                 }
2125                 if (contig_pages + i >  num_pages)
2126                         contig_pages = num_pages - i;
2127
2128                 /* for reads over a certain size could initiate async
2129                    read ahead */
2130
2131                 read_size = contig_pages * PAGE_CACHE_SIZE;
2132                 /* Read size needs to be in multiples of one page */
2133                 read_size = min_t(const unsigned int, read_size,
2134                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2135                 cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
2136                                 read_size, contig_pages);
2137                 rc = -EAGAIN;
2138                 while (rc == -EAGAIN) {
2139                         if (open_file->invalidHandle) {
2140                                 rc = cifs_reopen_file(open_file, true);
2141                                 if (rc != 0)
2142                                         break;
2143                         }
2144
2145                         rc = CIFSSMBRead(xid, pTcon,
2146                                          open_file->netfid,
2147                                          read_size, offset,
2148                                          &bytes_read, &smb_read_data,
2149                                          &buf_type);
2150                         /* BB more RC checks ? */
2151                         if (rc == -EAGAIN) {
2152                                 if (smb_read_data) {
2153                                         if (buf_type == CIFS_SMALL_BUFFER)
2154                                                 cifs_small_buf_release(smb_read_data);
2155                                         else if (buf_type == CIFS_LARGE_BUFFER)
2156                                                 cifs_buf_release(smb_read_data);
2157                                         smb_read_data = NULL;
2158                                 }
2159                         }
2160                 }
2161                 if ((rc < 0) || (smb_read_data == NULL)) {
2162                         cFYI(1, "Read error in readpages: %d", rc);
2163                         break;
2164                 } else if (bytes_read > 0) {
2165                         task_io_account_read(bytes_read);
2166                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2167                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2168                                 smb_read_data + 4 /* RFC1001 hdr */ +
2169                                 le16_to_cpu(pSMBr->DataOffset));
2170
2171                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2172                         cifs_stats_bytes_read(pTcon, bytes_read);
2173                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2174                                 i++; /* account for partial page */
2175
2176                                 /* server copy of file can have smaller size
2177                                    than client */
2178                                 /* BB do we need to verify this common case ?
2179                                    this case is ok - if we are at server EOF
2180                                    we will hit it on next read */
2181
2182                                 /* break; */
2183                         }
2184                 } else {
2185                         cFYI(1, "No bytes read (%d) at offset %lld . "
2186                                 "Cleaning remaining pages from readahead list",
2187                                 bytes_read, offset);
2188                         /* BB turn off caching and do new lookup on
2189                            file size at server? */
2190                         break;
2191                 }
2192                 if (smb_read_data) {
2193                         if (buf_type == CIFS_SMALL_BUFFER)
2194                                 cifs_small_buf_release(smb_read_data);
2195                         else if (buf_type == CIFS_LARGE_BUFFER)
2196                                 cifs_buf_release(smb_read_data);
2197                         smb_read_data = NULL;
2198                 }
2199                 bytes_read = 0;
2200         }
2201
2202 /* need to free smb_read_data buf before exit */
2203         if (smb_read_data) {
2204                 if (buf_type == CIFS_SMALL_BUFFER)
2205                         cifs_small_buf_release(smb_read_data);
2206                 else if (buf_type == CIFS_LARGE_BUFFER)
2207                         cifs_buf_release(smb_read_data);
2208                 smb_read_data = NULL;
2209         }
2210
2211 read_complete:
2212         FreeXid(xid);
2213         return rc;
2214 }
2215
2216 static int cifs_readpage_worker(struct file *file, struct page *page,
2217         loff_t *poffset)
2218 {
2219         char *read_data;
2220         int rc;
2221
2222         /* Is the page cached? */
2223         rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2224         if (rc == 0)
2225                 goto read_complete;
2226
2227         page_cache_get(page);
2228         read_data = kmap(page);
2229         /* for reads over a certain size could initiate async read ahead */
2230
2231         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2232
2233         if (rc < 0)
2234                 goto io_error;
2235         else
2236                 cFYI(1, "Bytes read %d", rc);
2237
2238         file->f_path.dentry->d_inode->i_atime =
2239                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2240
2241         if (PAGE_CACHE_SIZE > rc)
2242                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2243
2244         flush_dcache_page(page);
2245         SetPageUptodate(page);
2246
2247         /* send this page to the cache */
2248         cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2249
2250         rc = 0;
2251
2252 io_error:
2253         kunmap(page);
2254         page_cache_release(page);
2255
2256 read_complete:
2257         return rc;
2258 }
2259
2260 static int cifs_readpage(struct file *file, struct page *page)
2261 {
2262         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2263         int rc = -EACCES;
2264         int xid;
2265
2266         xid = GetXid();
2267
2268         if (file->private_data == NULL) {
2269                 rc = -EBADF;
2270                 FreeXid(xid);
2271                 return rc;
2272         }
2273
2274         cFYI(1, "readpage %p at offset %d 0x%x\n",
2275                  page, (int)offset, (int)offset);
2276
2277         rc = cifs_readpage_worker(file, page, &offset);
2278
2279         unlock_page(page);
2280
2281         FreeXid(xid);
2282         return rc;
2283 }
2284
2285 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2286 {
2287         struct cifsFileInfo *open_file;
2288
2289         spin_lock(&cifs_file_list_lock);
2290         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2291                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2292                         spin_unlock(&cifs_file_list_lock);
2293                         return 1;
2294                 }
2295         }
2296         spin_unlock(&cifs_file_list_lock);
2297         return 0;
2298 }
2299
2300 /* We do not want to update the file size from server for inodes
2301    open for write - to avoid races with writepage extending
2302    the file - in the future we could consider allowing
2303    refreshing the inode only on increases in the file size
2304    but this is tricky to do without racing with writebehind
2305    page caching in the current Linux kernel design */
2306 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2307 {
2308         if (!cifsInode)
2309                 return true;
2310
2311         if (is_inode_writable(cifsInode)) {
2312                 /* This inode is open for write at least once */
2313                 struct cifs_sb_info *cifs_sb;
2314
2315                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2316                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2317                         /* since no page cache to corrupt on directio
2318                         we can change size safely */
2319                         return true;
2320                 }
2321
2322                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2323                         return true;
2324
2325                 return false;
2326         } else
2327                 return true;
2328 }
2329
2330 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2331                         loff_t pos, unsigned len, unsigned flags,
2332                         struct page **pagep, void **fsdata)
2333 {
2334         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2335         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2336         loff_t page_start = pos & PAGE_MASK;
2337         loff_t i_size;
2338         struct page *page;
2339         int rc = 0;
2340
2341         cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2342
2343         page = grab_cache_page_write_begin(mapping, index, flags);
2344         if (!page) {
2345                 rc = -ENOMEM;
2346                 goto out;
2347         }
2348
2349         if (PageUptodate(page))
2350                 goto out;
2351
2352         /*
2353          * If we write a full page it will be up to date, no need to read from
2354          * the server. If the write is short, we'll end up doing a sync write
2355          * instead.
2356          */
2357         if (len == PAGE_CACHE_SIZE)
2358                 goto out;
2359
2360         /*
2361          * optimize away the read when we have an oplock, and we're not
2362          * expecting to use any of the data we'd be reading in. That
2363          * is, when the page lies beyond the EOF, or straddles the EOF
2364          * and the write will cover all of the existing data.
2365          */
2366         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2367                 i_size = i_size_read(mapping->host);
2368                 if (page_start >= i_size ||
2369                     (offset == 0 && (pos + len) >= i_size)) {
2370                         zero_user_segments(page, 0, offset,
2371                                            offset + len,
2372                                            PAGE_CACHE_SIZE);
2373                         /*
2374                          * PageChecked means that the parts of the page
2375                          * to which we're not writing are considered up
2376                          * to date. Once the data is copied to the
2377                          * page, it can be set uptodate.
2378                          */
2379                         SetPageChecked(page);
2380                         goto out;
2381                 }
2382         }
2383
2384         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2385                 /*
2386                  * might as well read a page, it is fast enough. If we get
2387                  * an error, we don't need to return it. cifs_write_end will
2388                  * do a sync write instead since PG_uptodate isn't set.
2389                  */
2390                 cifs_readpage_worker(file, page, &page_start);
2391         } else {
2392                 /* we could try using another file handle if there is one -
2393                    but how would we lock it to prevent close of that handle
2394                    racing with this read? In any case
2395                    this will be written out by write_end so is fine */
2396         }
2397 out:
2398         *pagep = page;
2399         return rc;
2400 }
2401
2402 static int cifs_release_page(struct page *page, gfp_t gfp)
2403 {
2404         if (PagePrivate(page))
2405                 return 0;
2406
2407         return cifs_fscache_release_page(page, gfp);
2408 }
2409
2410 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2411 {
2412         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2413
2414         if (offset == 0)
2415                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2416 }
2417
2418 void cifs_oplock_break(struct work_struct *work)
2419 {
2420         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2421                                                   oplock_break);
2422         struct inode *inode = cfile->dentry->d_inode;
2423         struct cifsInodeInfo *cinode = CIFS_I(inode);
2424         int rc = 0;
2425
2426         if (inode && S_ISREG(inode->i_mode)) {
2427                 if (cinode->clientCanCacheRead)
2428                         break_lease(inode, O_RDONLY);
2429                 else
2430                         break_lease(inode, O_WRONLY);
2431                 rc = filemap_fdatawrite(inode->i_mapping);
2432                 if (cinode->clientCanCacheRead == 0) {
2433                         rc = filemap_fdatawait(inode->i_mapping);
2434                         mapping_set_error(inode->i_mapping, rc);
2435                         invalidate_remote_inode(inode);
2436                 }
2437                 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2438         }
2439
2440         /*
2441          * releasing stale oplock after recent reconnect of smb session using
2442          * a now incorrect file handle is not a data integrity issue but do
2443          * not bother sending an oplock release if session to server still is
2444          * disconnected since oplock already released by the server
2445          */
2446         if (!cfile->oplock_break_cancelled) {
2447                 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2448                                  0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false,
2449                                  cinode->clientCanCacheRead ? 1 : 0);
2450                 cFYI(1, "Oplock release rc = %d", rc);
2451         }
2452
2453         /*
2454          * We might have kicked in before is_valid_oplock_break()
2455          * finished grabbing reference for us.  Make sure it's done by
2456          * waiting for cifs_file_list_lock.
2457          */
2458         spin_lock(&cifs_file_list_lock);
2459         spin_unlock(&cifs_file_list_lock);
2460
2461         cifs_oplock_break_put(cfile);
2462 }
2463
2464 /* must be called while holding cifs_file_list_lock */
2465 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2466 {
2467         cifs_sb_active(cfile->dentry->d_sb);
2468         cifsFileInfo_get(cfile);
2469 }
2470
2471 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2472 {
2473         struct super_block *sb = cfile->dentry->d_sb;
2474
2475         cifsFileInfo_put(cfile);
2476         cifs_sb_deactive(sb);
2477 }
2478
2479 const struct address_space_operations cifs_addr_ops = {
2480         .readpage = cifs_readpage,
2481         .readpages = cifs_readpages,
2482         .writepage = cifs_writepage,
2483         .writepages = cifs_writepages,
2484         .write_begin = cifs_write_begin,
2485         .write_end = cifs_write_end,
2486         .set_page_dirty = __set_page_dirty_nobuffers,
2487         .releasepage = cifs_release_page,
2488         .invalidatepage = cifs_invalidate_page,
2489         /* .direct_IO = */
2490 };
2491
2492 /*
2493  * cifs_readpages requires the server to support a buffer large enough to
2494  * contain the header plus one complete page of data.  Otherwise, we need
2495  * to leave cifs_readpages out of the address space operations.
2496  */
2497 const struct address_space_operations cifs_addr_ops_smallbuf = {
2498         .readpage = cifs_readpage,
2499         .writepage = cifs_writepage,
2500         .writepages = cifs_writepages,
2501         .write_begin = cifs_write_begin,
2502         .write_end = cifs_write_end,
2503         .set_page_dirty = __set_page_dirty_nobuffers,
2504         .releasepage = cifs_release_page,
2505         .invalidatepage = cifs_invalidate_page,
2506         /* .direct_IO = */
2507 };