Orangefs: Merge tag 'v4.4-rc1' into for-next
[deliverable/linux.git] / fs / orangefs / file.c
1 /*
2 * (C) 2001 Clemson University and The University of Chicago
3 *
4 * See COPYING in top-level directory.
5 */
6
7 /*
8 * Linux VFS file operations.
9 */
10
11 #include "protocol.h"
12 #include "pvfs2-kernel.h"
13 #include "pvfs2-bufmap.h"
14 #include <linux/fs.h>
15 #include <linux/pagemap.h>
16
17 #define wake_up_daemon_for_return(op) \
18 do { \
19 spin_lock(&op->lock); \
20 op->io_completed = 1; \
21 spin_unlock(&op->lock); \
22 wake_up_interruptible(&op->io_completion_waitq);\
23 } while (0)
24
25 /*
26 * Copy to client-core's address space from the buffers specified
27 * by the iovec upto total_size bytes.
28 * NOTE: the iovector can either contain addresses which
29 * can futher be kernel-space or user-space addresses.
30 * or it can pointers to struct page's
31 */
32 static int precopy_buffers(struct pvfs2_bufmap *bufmap,
33 int buffer_index,
34 struct iov_iter *iter,
35 size_t total_size)
36 {
37 int ret = 0;
38 /*
39 * copy data from application/kernel by pulling it out
40 * of the iovec.
41 */
42
43
44 if (total_size) {
45 ret = pvfs_bufmap_copy_from_iovec(bufmap,
46 iter,
47 buffer_index,
48 total_size);
49 if (ret < 0)
50 gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
51 __func__,
52 (long)ret);
53 }
54
55 if (ret < 0)
56 gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
57 __func__,
58 (long)ret);
59 return ret;
60 }
61
62 /*
63 * Copy from client-core's address space to the buffers specified
64 * by the iovec upto total_size bytes.
65 * NOTE: the iovector can either contain addresses which
66 * can futher be kernel-space or user-space addresses.
67 * or it can pointers to struct page's
68 */
69 static int postcopy_buffers(struct pvfs2_bufmap *bufmap,
70 int buffer_index,
71 struct iov_iter *iter,
72 size_t total_size)
73 {
74 int ret = 0;
75 /*
76 * copy data to application/kernel by pushing it out to
77 * the iovec. NOTE; target buffers can be addresses or
78 * struct page pointers.
79 */
80 if (total_size) {
81 ret = pvfs_bufmap_copy_to_iovec(bufmap,
82 iter,
83 buffer_index,
84 total_size);
85 if (ret < 0)
86 gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n",
87 __func__,
88 (long)ret);
89 }
90 return ret;
91 }
92
93 /*
94 * Post and wait for the I/O upcall to finish
95 */
96 static ssize_t wait_for_direct_io(enum PVFS_io_type type, struct inode *inode,
97 loff_t *offset, struct iov_iter *iter,
98 size_t total_size, loff_t readahead_size)
99 {
100 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
101 struct pvfs2_khandle *handle = &pvfs2_inode->refn.khandle;
102 struct pvfs2_bufmap *bufmap = NULL;
103 struct pvfs2_kernel_op_s *new_op = NULL;
104 int buffer_index = -1;
105 ssize_t ret;
106
107 new_op = op_alloc(PVFS2_VFS_OP_FILE_IO);
108 if (!new_op) {
109 ret = -ENOMEM;
110 goto out;
111 }
112 /* synchronous I/O */
113 new_op->upcall.req.io.async_vfs_io = PVFS_VFS_SYNC_IO;
114 new_op->upcall.req.io.readahead_size = readahead_size;
115 new_op->upcall.req.io.io_type = type;
116 new_op->upcall.req.io.refn = pvfs2_inode->refn;
117
118 populate_shared_memory:
119 /* get a shared buffer index */
120 ret = pvfs_bufmap_get(&bufmap, &buffer_index);
121 if (ret < 0) {
122 gossip_debug(GOSSIP_FILE_DEBUG,
123 "%s: pvfs_bufmap_get failure (%ld)\n",
124 __func__, (long)ret);
125 goto out;
126 }
127 gossip_debug(GOSSIP_FILE_DEBUG,
128 "%s(%pU): GET op %p -> buffer_index %d\n",
129 __func__,
130 handle,
131 new_op,
132 buffer_index);
133
134 new_op->uses_shared_memory = 1;
135 new_op->upcall.req.io.buf_index = buffer_index;
136 new_op->upcall.req.io.count = total_size;
137 new_op->upcall.req.io.offset = *offset;
138
139 gossip_debug(GOSSIP_FILE_DEBUG,
140 "%s(%pU): offset: %llu total_size: %zd\n",
141 __func__,
142 handle,
143 llu(*offset),
144 total_size);
145 /*
146 * Stage 1: copy the buffers into client-core's address space
147 * precopy_buffers only pertains to writes.
148 */
149 if (type == PVFS_IO_WRITE) {
150 ret = precopy_buffers(bufmap,
151 buffer_index,
152 iter,
153 total_size);
154 if (ret < 0)
155 goto out;
156 }
157
158 gossip_debug(GOSSIP_FILE_DEBUG,
159 "%s(%pU): Calling post_io_request with tag (%llu)\n",
160 __func__,
161 handle,
162 llu(new_op->tag));
163
164 /* Stage 2: Service the I/O operation */
165 ret = service_operation(new_op,
166 type == PVFS_IO_WRITE ?
167 "file_write" :
168 "file_read",
169 get_interruptible_flag(inode));
170
171 /*
172 * If service_operation() returns -EAGAIN #and# the operation was
173 * purged from pvfs2_request_list or htable_ops_in_progress, then
174 * we know that the client was restarted, causing the shared memory
175 * area to be wiped clean. To restart a write operation in this
176 * case, we must re-copy the data from the user's iovec to a NEW
177 * shared memory location. To restart a read operation, we must get
178 * a new shared memory location.
179 */
180 if (ret == -EAGAIN && op_state_purged(new_op)) {
181 pvfs_bufmap_put(bufmap, buffer_index);
182 gossip_debug(GOSSIP_FILE_DEBUG,
183 "%s:going to repopulate_shared_memory.\n",
184 __func__);
185 goto populate_shared_memory;
186 }
187
188 if (ret < 0) {
189 handle_io_error(); /* defined in pvfs2-kernel.h */
190 /*
191 * don't write an error to syslog on signaled operation
192 * termination unless we've got debugging turned on, as
193 * this can happen regularly (i.e. ctrl-c)
194 */
195 if (ret == -EINTR)
196 gossip_debug(GOSSIP_FILE_DEBUG,
197 "%s: returning error %ld\n", __func__,
198 (long)ret);
199 else
200 gossip_err("%s: error in %s handle %pU, returning %zd\n",
201 __func__,
202 type == PVFS_IO_READ ?
203 "read from" : "write to",
204 handle, ret);
205 goto out;
206 }
207
208 /*
209 * Stage 3: Post copy buffers from client-core's address space
210 * postcopy_buffers only pertains to reads.
211 */
212 if (type == PVFS_IO_READ) {
213 ret = postcopy_buffers(bufmap,
214 buffer_index,
215 iter,
216 new_op->downcall.resp.io.amt_complete);
217 if (ret < 0) {
218 /*
219 * put error codes in downcall so that handle_io_error()
220 * preserves it properly
221 */
222 new_op->downcall.status = ret;
223 handle_io_error();
224 goto out;
225 }
226 }
227 gossip_debug(GOSSIP_FILE_DEBUG,
228 "%s(%pU): Amount written as returned by the sys-io call:%d\n",
229 __func__,
230 handle,
231 (int)new_op->downcall.resp.io.amt_complete);
232
233 ret = new_op->downcall.resp.io.amt_complete;
234
235 /*
236 * tell the device file owner waiting on I/O that this read has
237 * completed and it can return now. in this exact case, on
238 * wakeup the daemon will free the op, so we *cannot* touch it
239 * after this.
240 */
241 wake_up_daemon_for_return(new_op);
242 new_op = NULL;
243
244 out:
245 if (buffer_index >= 0) {
246 pvfs_bufmap_put(bufmap, buffer_index);
247 gossip_debug(GOSSIP_FILE_DEBUG,
248 "%s(%pU): PUT buffer_index %d\n",
249 __func__, handle, buffer_index);
250 buffer_index = -1;
251 }
252 if (new_op) {
253 op_release(new_op);
254 new_op = NULL;
255 }
256 return ret;
257 }
258
259 /*
260 * Common entry point for read/write/readv/writev
261 * This function will dispatch it to either the direct I/O
262 * or buffered I/O path depending on the mount options and/or
263 * augmented/extended metadata attached to the file.
264 * Note: File extended attributes override any mount options.
265 */
266 static ssize_t do_readv_writev(enum PVFS_io_type type, struct file *file,
267 loff_t *offset, struct iov_iter *iter)
268 {
269 struct inode *inode = file->f_mapping->host;
270 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
271 struct pvfs2_khandle *handle = &pvfs2_inode->refn.khandle;
272 size_t count = iov_iter_count(iter);
273 ssize_t total_count = 0;
274 ssize_t ret = -EINVAL;
275
276 gossip_debug(GOSSIP_FILE_DEBUG,
277 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
278 __func__,
279 handle,
280 (int)count);
281
282 if (type == PVFS_IO_WRITE) {
283 gossip_debug(GOSSIP_FILE_DEBUG,
284 "%s(%pU): proceeding with offset : %llu, "
285 "size %d\n",
286 __func__,
287 handle,
288 llu(*offset),
289 (int)count);
290 }
291
292 if (count == 0) {
293 ret = 0;
294 goto out;
295 }
296
297 while (iov_iter_count(iter)) {
298 size_t each_count = iov_iter_count(iter);
299 size_t amt_complete;
300
301 /* how much to transfer in this loop iteration */
302 if (each_count > pvfs_bufmap_size_query())
303 each_count = pvfs_bufmap_size_query();
304
305 gossip_debug(GOSSIP_FILE_DEBUG,
306 "%s(%pU): size of each_count(%d)\n",
307 __func__,
308 handle,
309 (int)each_count);
310 gossip_debug(GOSSIP_FILE_DEBUG,
311 "%s(%pU): BEFORE wait_for_io: offset is %d\n",
312 __func__,
313 handle,
314 (int)*offset);
315
316 ret = wait_for_direct_io(type, inode, offset, iter,
317 each_count, 0);
318 gossip_debug(GOSSIP_FILE_DEBUG,
319 "%s(%pU): return from wait_for_io:%d\n",
320 __func__,
321 handle,
322 (int)ret);
323
324 if (ret < 0)
325 goto out;
326
327 *offset += ret;
328 total_count += ret;
329 amt_complete = ret;
330
331 gossip_debug(GOSSIP_FILE_DEBUG,
332 "%s(%pU): AFTER wait_for_io: offset is %d\n",
333 __func__,
334 handle,
335 (int)*offset);
336
337 /*
338 * if we got a short I/O operations,
339 * fall out and return what we got so far
340 */
341 if (amt_complete < each_count)
342 break;
343 } /*end while */
344
345 if (total_count > 0)
346 ret = total_count;
347 out:
348 if (ret > 0) {
349 if (type == PVFS_IO_READ) {
350 file_accessed(file);
351 } else {
352 SetMtimeFlag(pvfs2_inode);
353 inode->i_mtime = CURRENT_TIME;
354 mark_inode_dirty_sync(inode);
355 }
356 }
357
358 gossip_debug(GOSSIP_FILE_DEBUG,
359 "%s(%pU): Value(%d) returned.\n",
360 __func__,
361 handle,
362 (int)ret);
363
364 return ret;
365 }
366
367 /*
368 * Read data from a specified offset in a file (referenced by inode).
369 * Data may be placed either in a user or kernel buffer.
370 */
371 ssize_t pvfs2_inode_read(struct inode *inode,
372 struct iov_iter *iter,
373 loff_t *offset,
374 loff_t readahead_size)
375 {
376 struct pvfs2_inode_s *pvfs2_inode = PVFS2_I(inode);
377 size_t count = iov_iter_count(iter);
378 size_t bufmap_size;
379 ssize_t ret = -EINVAL;
380
381 g_pvfs2_stats.reads++;
382
383 bufmap_size = pvfs_bufmap_size_query();
384 if (count > bufmap_size) {
385 gossip_debug(GOSSIP_FILE_DEBUG,
386 "%s: count is too large (%zd/%zd)!\n",
387 __func__, count, bufmap_size);
388 return -EINVAL;
389 }
390
391 gossip_debug(GOSSIP_FILE_DEBUG,
392 "%s(%pU) %zd@%llu\n",
393 __func__,
394 &pvfs2_inode->refn.khandle,
395 count,
396 llu(*offset));
397
398 ret = wait_for_direct_io(PVFS_IO_READ, inode, offset, iter,
399 count, readahead_size);
400 if (ret > 0)
401 *offset += ret;
402
403 gossip_debug(GOSSIP_FILE_DEBUG,
404 "%s(%pU): Value(%zd) returned.\n",
405 __func__,
406 &pvfs2_inode->refn.khandle,
407 ret);
408
409 return ret;
410 }
411
412 static ssize_t pvfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
413 {
414 struct file *file = iocb->ki_filp;
415 loff_t pos = *(&iocb->ki_pos);
416 ssize_t rc = 0;
417
418 BUG_ON(iocb->private);
419
420 gossip_debug(GOSSIP_FILE_DEBUG, "pvfs2_file_read_iter\n");
421
422 g_pvfs2_stats.reads++;
423
424 rc = do_readv_writev(PVFS_IO_READ, file, &pos, iter);
425 iocb->ki_pos = pos;
426
427 return rc;
428 }
429
430 static ssize_t pvfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
431 {
432 struct file *file = iocb->ki_filp;
433 loff_t pos;
434 ssize_t rc;
435
436 BUG_ON(iocb->private);
437
438 gossip_debug(GOSSIP_FILE_DEBUG, "pvfs2_file_write_iter\n");
439
440 mutex_lock(&file->f_mapping->host->i_mutex);
441
442 /* Make sure generic_write_checks sees an up to date inode size. */
443 if (file->f_flags & O_APPEND) {
444 rc = pvfs2_inode_getattr(file->f_mapping->host,
445 PVFS_ATTR_SYS_SIZE);
446 if (rc) {
447 gossip_err("%s: pvfs2_inode_getattr failed, rc:%zd:.\n",
448 __func__, rc);
449 goto out;
450 }
451 }
452
453 if (file->f_pos > i_size_read(file->f_mapping->host))
454 pvfs2_i_size_write(file->f_mapping->host, file->f_pos);
455
456 rc = generic_write_checks(iocb, iter);
457
458 if (rc <= 0) {
459 gossip_err("%s: generic_write_checks failed, rc:%zd:.\n",
460 __func__, rc);
461 goto out;
462 }
463
464 /*
465 * if we are appending, generic_write_checks would have updated
466 * pos to the end of the file, so we will wait till now to set
467 * pos...
468 */
469 pos = *(&iocb->ki_pos);
470
471 rc = do_readv_writev(PVFS_IO_WRITE,
472 file,
473 &pos,
474 iter);
475 if (rc < 0) {
476 gossip_err("%s: do_readv_writev failed, rc:%zd:.\n",
477 __func__, rc);
478 goto out;
479 }
480
481 iocb->ki_pos = pos;
482 g_pvfs2_stats.writes++;
483
484 out:
485
486 mutex_unlock(&file->f_mapping->host->i_mutex);
487 return rc;
488 }
489
490 /*
491 * Perform a miscellaneous operation on a file.
492 */
493 static long pvfs2_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
494 {
495 int ret = -ENOTTY;
496 __u64 val = 0;
497 unsigned long uval;
498
499 gossip_debug(GOSSIP_FILE_DEBUG,
500 "pvfs2_ioctl: called with cmd %d\n",
501 cmd);
502
503 /*
504 * we understand some general ioctls on files, such as the immutable
505 * and append flags
506 */
507 if (cmd == FS_IOC_GETFLAGS) {
508 val = 0;
509 ret = pvfs2_xattr_get_default(file->f_path.dentry,
510 "user.pvfs2.meta_hint",
511 &val,
512 sizeof(val),
513 0);
514 if (ret < 0 && ret != -ENODATA)
515 return ret;
516 else if (ret == -ENODATA)
517 val = 0;
518 uval = val;
519 gossip_debug(GOSSIP_FILE_DEBUG,
520 "pvfs2_ioctl: FS_IOC_GETFLAGS: %llu\n",
521 (unsigned long long)uval);
522 return put_user(uval, (int __user *)arg);
523 } else if (cmd == FS_IOC_SETFLAGS) {
524 ret = 0;
525 if (get_user(uval, (int __user *)arg))
526 return -EFAULT;
527 /*
528 * PVFS_MIRROR_FL is set internally when the mirroring mode
529 * is turned on for a file. The user is not allowed to turn
530 * on this bit, but the bit is present if the user first gets
531 * the flags and then updates the flags with some new
532 * settings. So, we ignore it in the following edit. bligon.
533 */
534 if ((uval & ~PVFS_MIRROR_FL) &
535 (~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL))) {
536 gossip_err("pvfs2_ioctl: the FS_IOC_SETFLAGS only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n");
537 return -EINVAL;
538 }
539 val = uval;
540 gossip_debug(GOSSIP_FILE_DEBUG,
541 "pvfs2_ioctl: FS_IOC_SETFLAGS: %llu\n",
542 (unsigned long long)val);
543 ret = pvfs2_xattr_set_default(file->f_path.dentry,
544 "user.pvfs2.meta_hint",
545 &val,
546 sizeof(val),
547 0,
548 0);
549 }
550
551 return ret;
552 }
553
554 /*
555 * Memory map a region of a file.
556 */
557 static int pvfs2_file_mmap(struct file *file, struct vm_area_struct *vma)
558 {
559 gossip_debug(GOSSIP_FILE_DEBUG,
560 "pvfs2_file_mmap: called on %s\n",
561 (file ?
562 (char *)file->f_path.dentry->d_name.name :
563 (char *)"Unknown"));
564
565 /* set the sequential readahead hint */
566 vma->vm_flags |= VM_SEQ_READ;
567 vma->vm_flags &= ~VM_RAND_READ;
568
569 /* Use readonly mmap since we cannot support writable maps. */
570 return generic_file_readonly_mmap(file, vma);
571 }
572
573 #define mapping_nrpages(idata) ((idata)->nrpages)
574
575 /*
576 * Called to notify the module that there are no more references to
577 * this file (i.e. no processes have it open).
578 *
579 * \note Not called when each file is closed.
580 */
581 static int pvfs2_file_release(struct inode *inode, struct file *file)
582 {
583 gossip_debug(GOSSIP_FILE_DEBUG,
584 "pvfs2_file_release: called on %s\n",
585 file->f_path.dentry->d_name.name);
586
587 pvfs2_flush_inode(inode);
588
589 /*
590 * remove all associated inode pages from the page cache and mmap
591 * readahead cache (if any); this forces an expensive refresh of
592 * data for the next caller of mmap (or 'get_block' accesses)
593 */
594 if (file->f_path.dentry->d_inode &&
595 file->f_path.dentry->d_inode->i_mapping &&
596 mapping_nrpages(&file->f_path.dentry->d_inode->i_data))
597 truncate_inode_pages(file->f_path.dentry->d_inode->i_mapping,
598 0);
599 return 0;
600 }
601
602 /*
603 * Push all data for a specific file onto permanent storage.
604 */
605 static int pvfs2_fsync(struct file *file,
606 loff_t start,
607 loff_t end,
608 int datasync)
609 {
610 int ret = -EINVAL;
611 struct pvfs2_inode_s *pvfs2_inode =
612 PVFS2_I(file->f_path.dentry->d_inode);
613 struct pvfs2_kernel_op_s *new_op = NULL;
614
615 /* required call */
616 filemap_write_and_wait_range(file->f_mapping, start, end);
617
618 new_op = op_alloc(PVFS2_VFS_OP_FSYNC);
619 if (!new_op)
620 return -ENOMEM;
621 new_op->upcall.req.fsync.refn = pvfs2_inode->refn;
622
623 ret = service_operation(new_op,
624 "pvfs2_fsync",
625 get_interruptible_flag(file->f_path.dentry->d_inode));
626
627 gossip_debug(GOSSIP_FILE_DEBUG,
628 "pvfs2_fsync got return value of %d\n",
629 ret);
630
631 op_release(new_op);
632
633 pvfs2_flush_inode(file->f_path.dentry->d_inode);
634 return ret;
635 }
636
637 /*
638 * Change the file pointer position for an instance of an open file.
639 *
640 * \note If .llseek is overriden, we must acquire lock as described in
641 * Documentation/filesystems/Locking.
642 *
643 * Future upgrade could support SEEK_DATA and SEEK_HOLE but would
644 * require much changes to the FS
645 */
646 static loff_t pvfs2_file_llseek(struct file *file, loff_t offset, int origin)
647 {
648 int ret = -EINVAL;
649 struct inode *inode = file->f_path.dentry->d_inode;
650
651 if (!inode) {
652 gossip_err("pvfs2_file_llseek: invalid inode (NULL)\n");
653 return ret;
654 }
655
656 if (origin == PVFS2_SEEK_END) {
657 /*
658 * revalidate the inode's file size.
659 * NOTE: We are only interested in file size here,
660 * so we set mask accordingly.
661 */
662 ret = pvfs2_inode_getattr(inode, PVFS_ATTR_SYS_SIZE);
663 if (ret) {
664 gossip_debug(GOSSIP_FILE_DEBUG,
665 "%s:%s:%d calling make bad inode\n",
666 __FILE__,
667 __func__,
668 __LINE__);
669 pvfs2_make_bad_inode(inode);
670 return ret;
671 }
672 }
673
674 gossip_debug(GOSSIP_FILE_DEBUG,
675 "pvfs2_file_llseek: offset is %ld | origin is %d"
676 " | inode size is %lu\n",
677 (long)offset,
678 origin,
679 (unsigned long)file->f_path.dentry->d_inode->i_size);
680
681 return generic_file_llseek(file, offset, origin);
682 }
683
684 /*
685 * Support local locks (locks that only this kernel knows about)
686 * if Orangefs was mounted -o local_lock.
687 */
688 static int pvfs2_lock(struct file *filp, int cmd, struct file_lock *fl)
689 {
690 int rc = -EINVAL;
691
692 if (PVFS2_SB(filp->f_inode->i_sb)->flags & PVFS2_OPT_LOCAL_LOCK) {
693 if (cmd == F_GETLK) {
694 rc = 0;
695 posix_test_lock(filp, fl);
696 } else {
697 rc = posix_lock_file(filp, fl, NULL);
698 }
699 }
700
701 return rc;
702 }
703
704 /** PVFS2 implementation of VFS file operations */
705 const struct file_operations pvfs2_file_operations = {
706 .llseek = pvfs2_file_llseek,
707 .read_iter = pvfs2_file_read_iter,
708 .write_iter = pvfs2_file_write_iter,
709 .lock = pvfs2_lock,
710 .unlocked_ioctl = pvfs2_ioctl,
711 .mmap = pvfs2_file_mmap,
712 .open = generic_file_open,
713 .release = pvfs2_file_release,
714 .fsync = pvfs2_fsync,
715 };
This page took 0.044025 seconds and 6 git commands to generate.