1 // SPDX-License-Identifier: GPL-2.0
3 * (C) 2001 Clemson University and The University of Chicago
5 * See COPYING in top-level directory.
9 * Linux VFS file operations.
13 #include "orangefs-kernel.h"
14 #include "orangefs-bufmap.h"
16 #include <linux/pagemap.h>
18 static int flush_racache(struct inode
*inode
)
20 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
21 struct orangefs_kernel_op_s
*new_op
;
24 gossip_debug(GOSSIP_UTILS_DEBUG
,
25 "%s: %pU: Handle is %pU | fs_id %d\n", __func__
,
26 get_khandle_from_ino(inode
), &orangefs_inode
->refn
.khandle
,
27 orangefs_inode
->refn
.fs_id
);
29 new_op
= op_alloc(ORANGEFS_VFS_OP_RA_FLUSH
);
32 new_op
->upcall
.req
.ra_cache_flush
.refn
= orangefs_inode
->refn
;
34 ret
= service_operation(new_op
, "orangefs_flush_racache",
35 get_interruptible_flag(inode
));
37 gossip_debug(GOSSIP_UTILS_DEBUG
, "%s: got return value of %d\n",
45 * Post and wait for the I/O upcall to finish
47 static ssize_t
wait_for_direct_io(enum ORANGEFS_io_type type
, struct inode
*inode
,
48 loff_t
*offset
, struct iov_iter
*iter
,
49 size_t total_size
, loff_t readahead_size
)
51 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
52 struct orangefs_khandle
*handle
= &orangefs_inode
->refn
.khandle
;
53 struct orangefs_kernel_op_s
*new_op
= NULL
;
54 int buffer_index
= -1;
57 new_op
= op_alloc(ORANGEFS_VFS_OP_FILE_IO
);
62 new_op
->upcall
.req
.io
.readahead_size
= readahead_size
;
63 new_op
->upcall
.req
.io
.io_type
= type
;
64 new_op
->upcall
.req
.io
.refn
= orangefs_inode
->refn
;
66 populate_shared_memory
:
67 /* get a shared buffer index */
68 buffer_index
= orangefs_bufmap_get();
69 if (buffer_index
< 0) {
71 gossip_debug(GOSSIP_FILE_DEBUG
,
72 "%s: orangefs_bufmap_get failure (%zd)\n",
76 gossip_debug(GOSSIP_FILE_DEBUG
,
77 "%s(%pU): GET op %p -> buffer_index %d\n",
83 new_op
->uses_shared_memory
= 1;
84 new_op
->upcall
.req
.io
.buf_index
= buffer_index
;
85 new_op
->upcall
.req
.io
.count
= total_size
;
86 new_op
->upcall
.req
.io
.offset
= *offset
;
88 gossip_debug(GOSSIP_FILE_DEBUG
,
89 "%s(%pU): offset: %llu total_size: %zd\n",
95 * Stage 1: copy the buffers into client-core's address space
97 if (type
== ORANGEFS_IO_WRITE
&& total_size
) {
98 ret
= orangefs_bufmap_copy_from_iovec(iter
, buffer_index
,
101 gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
102 __func__
, (long)ret
);
107 gossip_debug(GOSSIP_FILE_DEBUG
,
108 "%s(%pU): Calling post_io_request with tag (%llu)\n",
113 /* Stage 2: Service the I/O operation */
114 ret
= service_operation(new_op
,
115 type
== ORANGEFS_IO_WRITE
?
118 get_interruptible_flag(inode
));
121 * If service_operation() returns -EAGAIN #and# the operation was
122 * purged from orangefs_request_list or htable_ops_in_progress, then
123 * we know that the client was restarted, causing the shared memory
124 * area to be wiped clean. To restart a write operation in this
125 * case, we must re-copy the data from the user's iovec to a NEW
126 * shared memory location. To restart a read operation, we must get
127 * a new shared memory location.
129 if (ret
== -EAGAIN
&& op_state_purged(new_op
)) {
130 orangefs_bufmap_put(buffer_index
);
132 if (type
== ORANGEFS_IO_WRITE
)
133 iov_iter_revert(iter
, total_size
);
134 gossip_debug(GOSSIP_FILE_DEBUG
,
135 "%s:going to repopulate_shared_memory.\n",
137 goto populate_shared_memory
;
143 * We can't return EINTR if any data was written,
144 * it's not POSIX. It is minimally acceptable
145 * to give a partial write, the way NFS does.
147 * It would be optimal to return all or nothing,
148 * but if a userspace write is bigger than
149 * an IO buffer, and the interrupt occurs
150 * between buffer writes, that would not be
153 switch (new_op
->op_state
- OP_VFS_STATE_GIVEN_UP
) {
155 * If the op was waiting when the interrupt
156 * occurred, then the client-core did not
159 case OP_VFS_STATE_WAITING
:
166 * If the op was in progress when the interrupt
167 * occurred, then the client-core was able to
170 case OP_VFS_STATE_INPROGR
:
174 gossip_err("%s: unexpected op state :%d:.\n",
180 gossip_debug(GOSSIP_FILE_DEBUG
,
181 "%s: got EINTR, state:%d: %p\n",
186 gossip_err("%s: error in %s handle %pU, returning %zd\n",
188 type
== ORANGEFS_IO_READ
?
189 "read from" : "write to",
192 if (orangefs_cancel_op_in_progress(new_op
))
199 * Stage 3: Post copy buffers from client-core's address space
201 if (type
== ORANGEFS_IO_READ
&& new_op
->downcall
.resp
.io
.amt_complete
) {
203 * NOTE: the iovector can either contain addresses which
204 * can futher be kernel-space or user-space addresses.
205 * or it can pointers to struct page's
207 ret
= orangefs_bufmap_copy_to_iovec(iter
, buffer_index
,
208 new_op
->downcall
.resp
.io
.amt_complete
);
210 gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n",
211 __func__
, (long)ret
);
215 gossip_debug(GOSSIP_FILE_DEBUG
,
216 "%s(%pU): Amount %s, returned by the sys-io call:%d\n",
219 type
== ORANGEFS_IO_READ
? "read" : "written",
220 (int)new_op
->downcall
.resp
.io
.amt_complete
);
222 ret
= new_op
->downcall
.resp
.io
.amt_complete
;
225 if (buffer_index
>= 0) {
226 orangefs_bufmap_put(buffer_index
);
227 gossip_debug(GOSSIP_FILE_DEBUG
,
228 "%s(%pU): PUT buffer_index %d\n",
229 __func__
, handle
, buffer_index
);
237 * Common entry point for read/write/readv/writev
238 * This function will dispatch it to either the direct I/O
239 * or buffered I/O path depending on the mount options and/or
240 * augmented/extended metadata attached to the file.
241 * Note: File extended attributes override any mount options.
243 static ssize_t
do_readv_writev(enum ORANGEFS_io_type type
, struct file
*file
,
244 loff_t
*offset
, struct iov_iter
*iter
)
246 struct inode
*inode
= file
->f_mapping
->host
;
247 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
248 struct orangefs_khandle
*handle
= &orangefs_inode
->refn
.khandle
;
249 size_t count
= iov_iter_count(iter
);
250 ssize_t total_count
= 0;
251 ssize_t ret
= -EINVAL
;
253 gossip_debug(GOSSIP_FILE_DEBUG
,
254 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
259 if (type
== ORANGEFS_IO_WRITE
) {
260 gossip_debug(GOSSIP_FILE_DEBUG
,
261 "%s(%pU): proceeding with offset : %llu, "
274 while (iov_iter_count(iter
)) {
275 size_t each_count
= iov_iter_count(iter
);
278 /* how much to transfer in this loop iteration */
279 if (each_count
> orangefs_bufmap_size_query())
280 each_count
= orangefs_bufmap_size_query();
282 gossip_debug(GOSSIP_FILE_DEBUG
,
283 "%s(%pU): size of each_count(%d)\n",
287 gossip_debug(GOSSIP_FILE_DEBUG
,
288 "%s(%pU): BEFORE wait_for_io: offset is %d\n",
293 ret
= wait_for_direct_io(type
, inode
, offset
, iter
,
295 gossip_debug(GOSSIP_FILE_DEBUG
,
296 "%s(%pU): return from wait_for_io:%d\n",
308 gossip_debug(GOSSIP_FILE_DEBUG
,
309 "%s(%pU): AFTER wait_for_io: offset is %d\n",
315 * if we got a short I/O operations,
316 * fall out and return what we got so far
318 if (amt_complete
< each_count
)
326 if (type
== ORANGEFS_IO_READ
) {
329 file_update_time(file
);
331 * Must invalidate to ensure write loop doesn't
332 * prevent kernel from reading updated
333 * attribute. Size probably changed because of
334 * the write, and other clients could update
335 * any other attribute.
337 orangefs_inode
->getattr_time
= jiffies
- 1;
341 gossip_debug(GOSSIP_FILE_DEBUG
,
342 "%s(%pU): Value(%d) returned.\n",
351 * Read data from a specified offset in a file (referenced by inode).
352 * Data may be placed either in a user or kernel buffer.
354 ssize_t
orangefs_inode_read(struct inode
*inode
,
355 struct iov_iter
*iter
,
357 loff_t readahead_size
)
359 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
360 size_t count
= iov_iter_count(iter
);
362 ssize_t ret
= -EINVAL
;
364 orangefs_stats
.reads
++;
366 bufmap_size
= orangefs_bufmap_size_query();
367 if (count
> bufmap_size
) {
368 gossip_debug(GOSSIP_FILE_DEBUG
,
369 "%s: count is too large (%zd/%zd)!\n",
370 __func__
, count
, bufmap_size
);
374 gossip_debug(GOSSIP_FILE_DEBUG
,
375 "%s(%pU) %zd@%llu\n",
377 &orangefs_inode
->refn
.khandle
,
381 ret
= wait_for_direct_io(ORANGEFS_IO_READ
, inode
, offset
, iter
,
382 count
, readahead_size
);
386 gossip_debug(GOSSIP_FILE_DEBUG
,
387 "%s(%pU): Value(%zd) returned.\n",
389 &orangefs_inode
->refn
.khandle
,
395 static ssize_t
orangefs_file_read_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
397 struct file
*file
= iocb
->ki_filp
;
398 loff_t pos
= iocb
->ki_pos
;
401 BUG_ON(iocb
->private);
403 gossip_debug(GOSSIP_FILE_DEBUG
, "orangefs_file_read_iter\n");
405 orangefs_stats
.reads
++;
407 rc
= do_readv_writev(ORANGEFS_IO_READ
, file
, &pos
, iter
);
413 static ssize_t
orangefs_file_write_iter(struct kiocb
*iocb
, struct iov_iter
*iter
)
415 struct file
*file
= iocb
->ki_filp
;
419 BUG_ON(iocb
->private);
421 gossip_debug(GOSSIP_FILE_DEBUG
, "orangefs_file_write_iter\n");
423 inode_lock(file
->f_mapping
->host
);
425 /* Make sure generic_write_checks sees an up to date inode size. */
426 if (file
->f_flags
& O_APPEND
) {
427 rc
= orangefs_inode_getattr(file
->f_mapping
->host
, 0, 1,
432 gossip_err("%s: orangefs_inode_getattr failed, "
433 "rc:%zd:.\n", __func__
, rc
);
438 rc
= generic_write_checks(iocb
, iter
);
441 gossip_err("%s: generic_write_checks failed, rc:%zd:.\n",
447 * if we are appending, generic_write_checks would have updated
448 * pos to the end of the file, so we will wait till now to set
453 rc
= do_readv_writev(ORANGEFS_IO_WRITE
,
458 gossip_err("%s: do_readv_writev failed, rc:%zd:.\n",
464 orangefs_stats
.writes
++;
468 inode_unlock(file
->f_mapping
->host
);
473 * Perform a miscellaneous operation on a file.
475 static long orangefs_ioctl(struct file
*file
, unsigned int cmd
, unsigned long arg
)
481 gossip_debug(GOSSIP_FILE_DEBUG
,
482 "orangefs_ioctl: called with cmd %d\n",
486 * we understand some general ioctls on files, such as the immutable
489 if (cmd
== FS_IOC_GETFLAGS
) {
491 ret
= orangefs_inode_getxattr(file_inode(file
),
492 "user.pvfs2.meta_hint",
494 if (ret
< 0 && ret
!= -ENODATA
)
496 else if (ret
== -ENODATA
)
499 gossip_debug(GOSSIP_FILE_DEBUG
,
500 "orangefs_ioctl: FS_IOC_GETFLAGS: %llu\n",
501 (unsigned long long)uval
);
502 return put_user(uval
, (int __user
*)arg
);
503 } else if (cmd
== FS_IOC_SETFLAGS
) {
505 if (get_user(uval
, (int __user
*)arg
))
508 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode
509 * is turned on for a file. The user is not allowed to turn
510 * on this bit, but the bit is present if the user first gets
511 * the flags and then updates the flags with some new
512 * settings. So, we ignore it in the following edit. bligon.
514 if ((uval
& ~ORANGEFS_MIRROR_FL
) &
515 (~(FS_IMMUTABLE_FL
| FS_APPEND_FL
| FS_NOATIME_FL
))) {
516 gossip_err("orangefs_ioctl: the FS_IOC_SETFLAGS only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n");
520 gossip_debug(GOSSIP_FILE_DEBUG
,
521 "orangefs_ioctl: FS_IOC_SETFLAGS: %llu\n",
522 (unsigned long long)val
);
523 ret
= orangefs_inode_setxattr(file_inode(file
),
524 "user.pvfs2.meta_hint",
525 &val
, sizeof(val
), 0);
531 static int orangefs_fault(struct vm_fault
*vmf
)
533 struct file
*file
= vmf
->vma
->vm_file
;
535 rc
= orangefs_inode_getattr(file
->f_mapping
->host
, 0, 1,
540 gossip_err("%s: orangefs_inode_getattr failed, "
541 "rc:%d:.\n", __func__
, rc
);
544 return filemap_fault(vmf
);
547 const struct vm_operations_struct orangefs_file_vm_ops
= {
548 .fault
= orangefs_fault
,
549 .map_pages
= filemap_map_pages
,
550 .page_mkwrite
= filemap_page_mkwrite
,
554 * Memory map a region of a file.
556 static int orangefs_file_mmap(struct file
*file
, struct vm_area_struct
*vma
)
558 gossip_debug(GOSSIP_FILE_DEBUG
,
559 "orangefs_file_mmap: called on %s\n",
561 (char *)file
->f_path
.dentry
->d_name
.name
:
564 if ((vma
->vm_flags
& VM_SHARED
) && (vma
->vm_flags
& VM_MAYWRITE
))
567 /* set the sequential readahead hint */
568 vma
->vm_flags
|= VM_SEQ_READ
;
569 vma
->vm_flags
&= ~VM_RAND_READ
;
572 vma
->vm_ops
= &orangefs_file_vm_ops
;
576 #define mapping_nrpages(idata) ((idata)->nrpages)
579 * Called to notify the module that there are no more references to
580 * this file (i.e. no processes have it open).
582 * \note Not called when each file is closed.
584 static int orangefs_file_release(struct inode
*inode
, struct file
*file
)
586 gossip_debug(GOSSIP_FILE_DEBUG
,
587 "orangefs_file_release: called on %pD\n",
591 * remove all associated inode pages from the page cache and
592 * readahead cache (if any); this forces an expensive refresh of
593 * data for the next caller of mmap (or 'get_block' accesses)
595 if (file_inode(file
) &&
596 file_inode(file
)->i_mapping
&&
597 mapping_nrpages(&file_inode(file
)->i_data
)) {
598 if (orangefs_features
& ORANGEFS_FEATURE_READAHEAD
) {
599 gossip_debug(GOSSIP_INODE_DEBUG
,
600 "calling flush_racache on %pU\n",
601 get_khandle_from_ino(inode
));
602 flush_racache(inode
);
603 gossip_debug(GOSSIP_INODE_DEBUG
,
604 "flush_racache finished\n");
606 truncate_inode_pages(file_inode(file
)->i_mapping
,
613 * Push all data for a specific file onto permanent storage.
615 static int orangefs_fsync(struct file
*file
,
621 struct orangefs_inode_s
*orangefs_inode
=
622 ORANGEFS_I(file_inode(file
));
623 struct orangefs_kernel_op_s
*new_op
= NULL
;
625 new_op
= op_alloc(ORANGEFS_VFS_OP_FSYNC
);
628 new_op
->upcall
.req
.fsync
.refn
= orangefs_inode
->refn
;
630 ret
= service_operation(new_op
,
632 get_interruptible_flag(file_inode(file
)));
634 gossip_debug(GOSSIP_FILE_DEBUG
,
635 "orangefs_fsync got return value of %d\n",
643 * Change the file pointer position for an instance of an open file.
645 * \note If .llseek is overriden, we must acquire lock as described in
646 * Documentation/filesystems/Locking.
648 * Future upgrade could support SEEK_DATA and SEEK_HOLE but would
649 * require much changes to the FS
651 static loff_t
orangefs_file_llseek(struct file
*file
, loff_t offset
, int origin
)
654 struct inode
*inode
= file_inode(file
);
656 if (origin
== SEEK_END
) {
658 * revalidate the inode's file size.
659 * NOTE: We are only interested in file size here,
660 * so we set mask accordingly.
662 ret
= orangefs_inode_getattr(file
->f_mapping
->host
, 0, 1,
667 gossip_debug(GOSSIP_FILE_DEBUG
,
668 "%s:%s:%d calling make bad inode\n",
676 gossip_debug(GOSSIP_FILE_DEBUG
,
677 "orangefs_file_llseek: offset is %ld | origin is %d"
678 " | inode size is %lu\n",
681 (unsigned long)i_size_read(inode
));
683 return generic_file_llseek(file
, offset
, origin
);
687 * Support local locks (locks that only this kernel knows about)
688 * if Orangefs was mounted -o local_lock.
690 static int orangefs_lock(struct file
*filp
, int cmd
, struct file_lock
*fl
)
694 if (ORANGEFS_SB(file_inode(filp
)->i_sb
)->flags
& ORANGEFS_OPT_LOCAL_LOCK
) {
695 if (cmd
== F_GETLK
) {
697 posix_test_lock(filp
, fl
);
699 rc
= posix_lock_file(filp
, fl
, NULL
);
706 /** ORANGEFS implementation of VFS file operations */
707 const struct file_operations orangefs_file_operations
= {
708 .llseek
= orangefs_file_llseek
,
709 .read_iter
= orangefs_file_read_iter
,
710 .write_iter
= orangefs_file_write_iter
,
711 .lock
= orangefs_lock
,
712 .unlocked_ioctl
= orangefs_ioctl
,
713 .mmap
= orangefs_file_mmap
,
714 .open
= generic_file_open
,
715 .release
= orangefs_file_release
,
716 .fsync
= orangefs_fsync
,