blob: ae92795ed9658d476a6ab229582086986fc2ad27 [file] [log] [blame]
Mike Marshall5db11c22015-07-17 10:38:12 -04001/*
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"
Mike Marshall575e9462015-12-04 12:56:14 -050012#include "orangefs-kernel.h"
13#include "orangefs-bufmap.h"
Mike Marshall5db11c22015-07-17 10:38:12 -040014#include <linux/fs.h>
15#include <linux/pagemap.h>
16
Mike Marshall5db11c22015-07-17 10:38:12 -040017/*
18 * Copy to client-core's address space from the buffers specified
19 * by the iovec upto total_size bytes.
20 * NOTE: the iovector can either contain addresses which
21 * can futher be kernel-space or user-space addresses.
22 * or it can pointers to struct page's
23 */
Al Virobf6bf602016-02-16 20:06:19 -050024static int precopy_buffers(int buffer_index,
Al Viroa5c126a2015-10-08 17:54:31 -040025 struct iov_iter *iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -040026 size_t total_size)
Mike Marshall5db11c22015-07-17 10:38:12 -040027{
28 int ret = 0;
Mike Marshall5db11c22015-07-17 10:38:12 -040029 /*
30 * copy data from application/kernel by pulling it out
31 * of the iovec.
32 */
Mike Marshall4d1c4402015-09-04 10:31:16 -040033
34
35 if (total_size) {
Al Virobf6bf602016-02-16 20:06:19 -050036 ret = orangefs_bufmap_copy_from_iovec(iter,
Yi Liu8bb8aef2015-11-24 15:12:14 -050037 buffer_index,
38 total_size);
Mike Marshall4d1c4402015-09-04 10:31:16 -040039 if (ret < 0)
40 gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
41 __func__,
42 (long)ret);
Mike Marshall4d1c4402015-09-04 10:31:16 -040043 }
44
Mike Marshall5db11c22015-07-17 10:38:12 -040045 if (ret < 0)
46 gossip_err("%s: Failed to copy-in buffers. Please make sure that the pvfs2-client is running. %ld\n",
47 __func__,
48 (long)ret);
49 return ret;
50}
51
52/*
53 * Copy from client-core's address space to the buffers specified
54 * by the iovec upto total_size bytes.
55 * NOTE: the iovector can either contain addresses which
56 * can futher be kernel-space or user-space addresses.
57 * or it can pointers to struct page's
58 */
Al Virobf6bf602016-02-16 20:06:19 -050059static int postcopy_buffers(int buffer_index,
Al Viro5f0e3c92015-10-08 17:52:44 -040060 struct iov_iter *iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -040061 size_t total_size)
Mike Marshall5db11c22015-07-17 10:38:12 -040062{
63 int ret = 0;
Mike Marshall5db11c22015-07-17 10:38:12 -040064 /*
65 * copy data to application/kernel by pushing it out to
66 * the iovec. NOTE; target buffers can be addresses or
67 * struct page pointers.
68 */
69 if (total_size) {
Al Virobf6bf602016-02-16 20:06:19 -050070 ret = orangefs_bufmap_copy_to_iovec(iter,
Yi Liu8bb8aef2015-11-24 15:12:14 -050071 buffer_index,
72 total_size);
Mike Marshall5db11c22015-07-17 10:38:12 -040073 if (ret < 0)
Mike Marshall4d1c4402015-09-04 10:31:16 -040074 gossip_err("%s: Failed to copy-out buffers. Please make sure that the pvfs2-client is running (%ld)\n",
Mike Marshall5db11c22015-07-17 10:38:12 -040075 __func__,
76 (long)ret);
77 }
78 return ret;
79}
80
81/*
82 * Post and wait for the I/O upcall to finish
83 */
Yi Liu8bb8aef2015-11-24 15:12:14 -050084static ssize_t wait_for_direct_io(enum ORANGEFS_io_type type, struct inode *inode,
Al Viro3c2fcfc2015-10-08 18:00:26 -040085 loff_t *offset, struct iov_iter *iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -040086 size_t total_size, loff_t readahead_size)
Mike Marshall5db11c22015-07-17 10:38:12 -040087{
Yi Liu8bb8aef2015-11-24 15:12:14 -050088 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
89 struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
Yi Liu8bb8aef2015-11-24 15:12:14 -050090 struct orangefs_kernel_op_s *new_op = NULL;
Al Viro7b9761a2016-02-07 01:25:06 -050091 struct iov_iter saved = *iter;
Mike Marshall5db11c22015-07-17 10:38:12 -040092 int buffer_index = -1;
93 ssize_t ret;
94
Yi Liu8bb8aef2015-11-24 15:12:14 -050095 new_op = op_alloc(ORANGEFS_VFS_OP_FILE_IO);
Al Viroed42fe02016-01-22 19:47:47 -050096 if (!new_op)
97 return -ENOMEM;
98
Mike Marshall5db11c22015-07-17 10:38:12 -040099 /* synchronous I/O */
Mike Marshall5db11c22015-07-17 10:38:12 -0400100 new_op->upcall.req.io.readahead_size = readahead_size;
101 new_op->upcall.req.io.io_type = type;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500102 new_op->upcall.req.io.refn = orangefs_inode->refn;
Mike Marshall5db11c22015-07-17 10:38:12 -0400103
104populate_shared_memory:
105 /* get a shared buffer index */
Al Virob8a99a82016-02-16 20:10:26 -0500106 buffer_index = orangefs_bufmap_get();
107 if (buffer_index < 0) {
108 ret = buffer_index;
Mike Marshall5db11c22015-07-17 10:38:12 -0400109 gossip_debug(GOSSIP_FILE_DEBUG,
Al Virob8a99a82016-02-16 20:10:26 -0500110 "%s: orangefs_bufmap_get failure (%zd)\n",
111 __func__, ret);
Mike Marshall5db11c22015-07-17 10:38:12 -0400112 goto out;
113 }
114 gossip_debug(GOSSIP_FILE_DEBUG,
115 "%s(%pU): GET op %p -> buffer_index %d\n",
116 __func__,
117 handle,
118 new_op,
119 buffer_index);
120
121 new_op->uses_shared_memory = 1;
122 new_op->upcall.req.io.buf_index = buffer_index;
123 new_op->upcall.req.io.count = total_size;
124 new_op->upcall.req.io.offset = *offset;
125
126 gossip_debug(GOSSIP_FILE_DEBUG,
Al Viro3c2fcfc2015-10-08 18:00:26 -0400127 "%s(%pU): offset: %llu total_size: %zd\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400128 __func__,
129 handle,
Mike Marshall5db11c22015-07-17 10:38:12 -0400130 llu(*offset),
131 total_size);
132 /*
133 * Stage 1: copy the buffers into client-core's address space
134 * precopy_buffers only pertains to writes.
135 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500136 if (type == ORANGEFS_IO_WRITE) {
Al Virobf6bf602016-02-16 20:06:19 -0500137 ret = precopy_buffers(buffer_index,
Al Viro3c2fcfc2015-10-08 18:00:26 -0400138 iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -0400139 total_size);
Mike Marshall5db11c22015-07-17 10:38:12 -0400140 if (ret < 0)
141 goto out;
142 }
143
144 gossip_debug(GOSSIP_FILE_DEBUG,
145 "%s(%pU): Calling post_io_request with tag (%llu)\n",
146 __func__,
147 handle,
148 llu(new_op->tag));
149
150 /* Stage 2: Service the I/O operation */
151 ret = service_operation(new_op,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500152 type == ORANGEFS_IO_WRITE ?
Mike Marshall5db11c22015-07-17 10:38:12 -0400153 "file_write" :
154 "file_read",
155 get_interruptible_flag(inode));
156
157 /*
158 * If service_operation() returns -EAGAIN #and# the operation was
Yi Liu8bb8aef2015-11-24 15:12:14 -0500159 * purged from orangefs_request_list or htable_ops_in_progress, then
Mike Marshall5db11c22015-07-17 10:38:12 -0400160 * we know that the client was restarted, causing the shared memory
161 * area to be wiped clean. To restart a write operation in this
162 * case, we must re-copy the data from the user's iovec to a NEW
163 * shared memory location. To restart a read operation, we must get
164 * a new shared memory location.
165 */
166 if (ret == -EAGAIN && op_state_purged(new_op)) {
Al Viro1357d062016-02-11 21:34:52 -0500167 orangefs_bufmap_put(buffer_index);
Al Viroe17be9f2016-02-06 14:59:38 -0500168 buffer_index = -1;
Al Viro7b9761a2016-02-07 01:25:06 -0500169 if (type == ORANGEFS_IO_WRITE)
170 *iter = saved;
Mike Marshall5db11c22015-07-17 10:38:12 -0400171 gossip_debug(GOSSIP_FILE_DEBUG,
172 "%s:going to repopulate_shared_memory.\n",
173 __func__);
174 goto populate_shared_memory;
175 }
176
177 if (ret < 0) {
Mike Marshall162ada72016-03-09 13:12:37 -0500178 if (ret == -EINTR) {
179 /*
180 * We can't return EINTR if any data was written,
181 * it's not POSIX. It is minimally acceptable
182 * to give a partial write, the way NFS does.
183 *
184 * It would be optimal to return all or nothing,
185 * but if a userspace write is bigger than
186 * an IO buffer, and the interrupt occurs
187 * between buffer writes, that would not be
188 * possible.
189 */
190 switch (new_op->op_state - OP_VFS_STATE_GIVEN_UP) {
191 /*
192 * If the op was waiting when the interrupt
193 * occurred, then the client-core did not
194 * trigger the write.
195 */
196 case OP_VFS_STATE_WAITING:
197 if (*offset == 0)
198 ret = -EINTR;
199 else
200 ret = 0;
201 break;
202 /*
203 * If the op was in progress when the interrupt
204 * occurred, then the client-core was able to
205 * trigger the write.
206 */
207 case OP_VFS_STATE_INPROGR:
208 ret = total_size;
209 break;
210 default:
211 gossip_err("%s: unexpected op state :%d:.\n",
212 __func__,
213 new_op->op_state);
214 ret = 0;
215 break;
216 }
Mike Marshall5db11c22015-07-17 10:38:12 -0400217 gossip_debug(GOSSIP_FILE_DEBUG,
Mike Marshall162ada72016-03-09 13:12:37 -0500218 "%s: got EINTR, state:%d: %p\n",
219 __func__,
220 new_op->op_state,
221 new_op);
222 } else {
Mike Marshall5db11c22015-07-17 10:38:12 -0400223 gossip_err("%s: error in %s handle %pU, returning %zd\n",
224 __func__,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500225 type == ORANGEFS_IO_READ ?
Mike Marshall5db11c22015-07-17 10:38:12 -0400226 "read from" : "write to",
227 handle, ret);
Mike Marshall162ada72016-03-09 13:12:37 -0500228 }
Al Viro78699e22016-02-11 23:07:19 -0500229 if (orangefs_cancel_op_in_progress(new_op))
230 return ret;
231
Al Viro897c5df2016-02-13 21:06:50 -0500232 goto out;
Mike Marshall5db11c22015-07-17 10:38:12 -0400233 }
234
235 /*
236 * Stage 3: Post copy buffers from client-core's address space
237 * postcopy_buffers only pertains to reads.
238 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500239 if (type == ORANGEFS_IO_READ) {
Al Virobf6bf602016-02-16 20:06:19 -0500240 ret = postcopy_buffers(buffer_index,
Al Viro3c2fcfc2015-10-08 18:00:26 -0400241 iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -0400242 new_op->downcall.resp.io.amt_complete);
Al Viroc0eae8c2016-02-11 21:28:52 -0500243 if (ret < 0)
Al Viro897c5df2016-02-13 21:06:50 -0500244 goto out;
Mike Marshall5db11c22015-07-17 10:38:12 -0400245 }
246 gossip_debug(GOSSIP_FILE_DEBUG,
Mike Marshall9d9e7ba2016-03-03 13:46:48 -0500247 "%s(%pU): Amount %s, returned by the sys-io call:%d\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400248 __func__,
249 handle,
Mike Marshall9d9e7ba2016-03-03 13:46:48 -0500250 type == ORANGEFS_IO_READ ? "read" : "written",
Mike Marshall5db11c22015-07-17 10:38:12 -0400251 (int)new_op->downcall.resp.io.amt_complete);
252
253 ret = new_op->downcall.resp.io.amt_complete;
254
Mike Marshall5db11c22015-07-17 10:38:12 -0400255out:
256 if (buffer_index >= 0) {
Al Viro1357d062016-02-11 21:34:52 -0500257 orangefs_bufmap_put(buffer_index);
Mike Marshall5db11c22015-07-17 10:38:12 -0400258 gossip_debug(GOSSIP_FILE_DEBUG,
259 "%s(%pU): PUT buffer_index %d\n",
260 __func__, handle, buffer_index);
261 buffer_index = -1;
262 }
Al Viroed42fe02016-01-22 19:47:47 -0500263 op_release(new_op);
Mike Marshall5db11c22015-07-17 10:38:12 -0400264 return ret;
265}
266
267/*
Mike Marshall5db11c22015-07-17 10:38:12 -0400268 * Common entry point for read/write/readv/writev
269 * This function will dispatch it to either the direct I/O
270 * or buffered I/O path depending on the mount options and/or
271 * augmented/extended metadata attached to the file.
272 * Note: File extended attributes override any mount options.
273 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500274static ssize_t do_readv_writev(enum ORANGEFS_io_type type, struct file *file,
Al Viro0071ed12015-10-08 18:22:08 -0400275 loff_t *offset, struct iov_iter *iter)
Mike Marshall5db11c22015-07-17 10:38:12 -0400276{
277 struct inode *inode = file->f_mapping->host;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500278 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
279 struct orangefs_khandle *handle = &orangefs_inode->refn.khandle;
Al Viro0071ed12015-10-08 18:22:08 -0400280 size_t count = iov_iter_count(iter);
Al Virodc4067f2015-10-08 18:17:26 -0400281 ssize_t total_count = 0;
282 ssize_t ret = -EINVAL;
Mike Marshall5db11c22015-07-17 10:38:12 -0400283
284 gossip_debug(GOSSIP_FILE_DEBUG,
285 "%s-BEGIN(%pU): count(%d) after estimate_max_iovecs.\n",
286 __func__,
287 handle,
288 (int)count);
289
Yi Liu8bb8aef2015-11-24 15:12:14 -0500290 if (type == ORANGEFS_IO_WRITE) {
Mike Marshall5db11c22015-07-17 10:38:12 -0400291 gossip_debug(GOSSIP_FILE_DEBUG,
292 "%s(%pU): proceeding with offset : %llu, "
293 "size %d\n",
294 __func__,
295 handle,
296 llu(*offset),
297 (int)count);
298 }
299
300 if (count == 0) {
301 ret = 0;
302 goto out;
303 }
304
Al Viro0071ed12015-10-08 18:22:08 -0400305 while (iov_iter_count(iter)) {
306 size_t each_count = iov_iter_count(iter);
Mike Marshall5db11c22015-07-17 10:38:12 -0400307 size_t amt_complete;
308
309 /* how much to transfer in this loop iteration */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500310 if (each_count > orangefs_bufmap_size_query())
311 each_count = orangefs_bufmap_size_query();
Mike Marshall5db11c22015-07-17 10:38:12 -0400312
313 gossip_debug(GOSSIP_FILE_DEBUG,
314 "%s(%pU): size of each_count(%d)\n",
315 __func__,
316 handle,
317 (int)each_count);
318 gossip_debug(GOSSIP_FILE_DEBUG,
319 "%s(%pU): BEFORE wait_for_io: offset is %d\n",
320 __func__,
321 handle,
322 (int)*offset);
323
Al Viro0071ed12015-10-08 18:22:08 -0400324 ret = wait_for_direct_io(type, inode, offset, iter,
Al Viro3c2fcfc2015-10-08 18:00:26 -0400325 each_count, 0);
Mike Marshall5db11c22015-07-17 10:38:12 -0400326 gossip_debug(GOSSIP_FILE_DEBUG,
327 "%s(%pU): return from wait_for_io:%d\n",
328 __func__,
329 handle,
330 (int)ret);
331
332 if (ret < 0)
333 goto out;
334
Mike Marshall5db11c22015-07-17 10:38:12 -0400335 *offset += ret;
336 total_count += ret;
337 amt_complete = ret;
338
339 gossip_debug(GOSSIP_FILE_DEBUG,
340 "%s(%pU): AFTER wait_for_io: offset is %d\n",
341 __func__,
342 handle,
343 (int)*offset);
344
345 /*
346 * if we got a short I/O operations,
347 * fall out and return what we got so far
348 */
349 if (amt_complete < each_count)
350 break;
351 } /*end while */
352
Al Viro6d4c1a32016-02-16 20:15:43 -0500353out:
Mike Marshall5db11c22015-07-17 10:38:12 -0400354 if (total_count > 0)
355 ret = total_count;
Mike Marshall5db11c22015-07-17 10:38:12 -0400356 if (ret > 0) {
Yi Liu8bb8aef2015-11-24 15:12:14 -0500357 if (type == ORANGEFS_IO_READ) {
Mike Marshall5db11c22015-07-17 10:38:12 -0400358 file_accessed(file);
359 } else {
Yi Liu8bb8aef2015-11-24 15:12:14 -0500360 SetMtimeFlag(orangefs_inode);
Mike Marshall5db11c22015-07-17 10:38:12 -0400361 inode->i_mtime = CURRENT_TIME;
362 mark_inode_dirty_sync(inode);
363 }
364 }
365
366 gossip_debug(GOSSIP_FILE_DEBUG,
367 "%s(%pU): Value(%d) returned.\n",
368 __func__,
369 handle,
370 (int)ret);
371
372 return ret;
373}
374
375/*
376 * Read data from a specified offset in a file (referenced by inode).
377 * Data may be placed either in a user or kernel buffer.
378 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500379ssize_t orangefs_inode_read(struct inode *inode,
380 struct iov_iter *iter,
381 loff_t *offset,
382 loff_t readahead_size)
Mike Marshall5db11c22015-07-17 10:38:12 -0400383{
Yi Liu8bb8aef2015-11-24 15:12:14 -0500384 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
Al Viro74f68fc2015-10-08 18:31:05 -0400385 size_t count = iov_iter_count(iter);
Mike Marshall5db11c22015-07-17 10:38:12 -0400386 size_t bufmap_size;
Mike Marshall5db11c22015-07-17 10:38:12 -0400387 ssize_t ret = -EINVAL;
388
Yi Liu8bb8aef2015-11-24 15:12:14 -0500389 g_orangefs_stats.reads++;
Mike Marshall5db11c22015-07-17 10:38:12 -0400390
Yi Liu8bb8aef2015-11-24 15:12:14 -0500391 bufmap_size = orangefs_bufmap_size_query();
Mike Marshall5db11c22015-07-17 10:38:12 -0400392 if (count > bufmap_size) {
393 gossip_debug(GOSSIP_FILE_DEBUG,
394 "%s: count is too large (%zd/%zd)!\n",
395 __func__, count, bufmap_size);
396 return -EINVAL;
397 }
398
399 gossip_debug(GOSSIP_FILE_DEBUG,
400 "%s(%pU) %zd@%llu\n",
401 __func__,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500402 &orangefs_inode->refn.khandle,
Mike Marshall5db11c22015-07-17 10:38:12 -0400403 count,
404 llu(*offset));
405
Yi Liu8bb8aef2015-11-24 15:12:14 -0500406 ret = wait_for_direct_io(ORANGEFS_IO_READ, inode, offset, iter,
Mike Marshall4d1c4402015-09-04 10:31:16 -0400407 count, readahead_size);
Mike Marshall5db11c22015-07-17 10:38:12 -0400408 if (ret > 0)
409 *offset += ret;
410
411 gossip_debug(GOSSIP_FILE_DEBUG,
412 "%s(%pU): Value(%zd) returned.\n",
413 __func__,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500414 &orangefs_inode->refn.khandle,
Mike Marshall5db11c22015-07-17 10:38:12 -0400415 ret);
416
417 return ret;
418}
419
Yi Liu8bb8aef2015-11-24 15:12:14 -0500420static ssize_t orangefs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Mike Marshall5db11c22015-07-17 10:38:12 -0400421{
422 struct file *file = iocb->ki_filp;
423 loff_t pos = *(&iocb->ki_pos);
424 ssize_t rc = 0;
Mike Marshall5db11c22015-07-17 10:38:12 -0400425
426 BUG_ON(iocb->private);
427
Yi Liu8bb8aef2015-11-24 15:12:14 -0500428 gossip_debug(GOSSIP_FILE_DEBUG, "orangefs_file_read_iter\n");
Mike Marshall5db11c22015-07-17 10:38:12 -0400429
Yi Liu8bb8aef2015-11-24 15:12:14 -0500430 g_orangefs_stats.reads++;
Mike Marshall5db11c22015-07-17 10:38:12 -0400431
Yi Liu8bb8aef2015-11-24 15:12:14 -0500432 rc = do_readv_writev(ORANGEFS_IO_READ, file, &pos, iter);
Mike Marshall5db11c22015-07-17 10:38:12 -0400433 iocb->ki_pos = pos;
434
435 return rc;
436}
437
Yi Liu8bb8aef2015-11-24 15:12:14 -0500438static ssize_t orangefs_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
Mike Marshall5db11c22015-07-17 10:38:12 -0400439{
440 struct file *file = iocb->ki_filp;
Mike Marshall3f1b6942015-11-13 13:05:11 -0500441 loff_t pos;
Mike Marshall5db11c22015-07-17 10:38:12 -0400442 ssize_t rc;
443
444 BUG_ON(iocb->private);
445
Yi Liu8bb8aef2015-11-24 15:12:14 -0500446 gossip_debug(GOSSIP_FILE_DEBUG, "orangefs_file_write_iter\n");
Mike Marshall5db11c22015-07-17 10:38:12 -0400447
448 mutex_lock(&file->f_mapping->host->i_mutex);
449
450 /* Make sure generic_write_checks sees an up to date inode size. */
451 if (file->f_flags & O_APPEND) {
Martin Brandenburge2f7f0d2016-03-15 12:33:20 -0400452 rc = orangefs_inode_getattr(file->f_mapping->host, 0, 1);
453 if (rc == -ESTALE)
454 rc = -EIO;
Mike Marshall5db11c22015-07-17 10:38:12 -0400455 if (rc) {
Martin Brandenburge2f7f0d2016-03-15 12:33:20 -0400456 gossip_err("%s: orangefs_inode_getattr failed, "
457 "rc:%zd:.\n", __func__, rc);
Mike Marshall5db11c22015-07-17 10:38:12 -0400458 goto out;
459 }
460 }
461
462 if (file->f_pos > i_size_read(file->f_mapping->host))
Yi Liu8bb8aef2015-11-24 15:12:14 -0500463 orangefs_i_size_write(file->f_mapping->host, file->f_pos);
Mike Marshall5db11c22015-07-17 10:38:12 -0400464
465 rc = generic_write_checks(iocb, iter);
466
467 if (rc <= 0) {
468 gossip_err("%s: generic_write_checks failed, rc:%zd:.\n",
469 __func__, rc);
470 goto out;
471 }
472
Mike Marshall3f1b6942015-11-13 13:05:11 -0500473 /*
474 * if we are appending, generic_write_checks would have updated
475 * pos to the end of the file, so we will wait till now to set
476 * pos...
477 */
478 pos = *(&iocb->ki_pos);
479
Yi Liu8bb8aef2015-11-24 15:12:14 -0500480 rc = do_readv_writev(ORANGEFS_IO_WRITE,
Mike Marshall5db11c22015-07-17 10:38:12 -0400481 file,
482 &pos,
Al Viro0071ed12015-10-08 18:22:08 -0400483 iter);
Mike Marshall5db11c22015-07-17 10:38:12 -0400484 if (rc < 0) {
485 gossip_err("%s: do_readv_writev failed, rc:%zd:.\n",
486 __func__, rc);
487 goto out;
488 }
489
490 iocb->ki_pos = pos;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500491 g_orangefs_stats.writes++;
Mike Marshall5db11c22015-07-17 10:38:12 -0400492
493out:
494
495 mutex_unlock(&file->f_mapping->host->i_mutex);
496 return rc;
497}
498
499/*
500 * Perform a miscellaneous operation on a file.
501 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500502static long orangefs_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
Mike Marshall5db11c22015-07-17 10:38:12 -0400503{
504 int ret = -ENOTTY;
505 __u64 val = 0;
506 unsigned long uval;
507
508 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500509 "orangefs_ioctl: called with cmd %d\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400510 cmd);
511
512 /*
513 * we understand some general ioctls on files, such as the immutable
514 * and append flags
515 */
516 if (cmd == FS_IOC_GETFLAGS) {
517 val = 0;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500518 ret = orangefs_inode_getxattr(file_inode(file),
519 ORANGEFS_XATTR_NAME_DEFAULT_PREFIX,
520 "user.pvfs2.meta_hint",
521 &val, sizeof(val));
Mike Marshall5db11c22015-07-17 10:38:12 -0400522 if (ret < 0 && ret != -ENODATA)
523 return ret;
524 else if (ret == -ENODATA)
525 val = 0;
526 uval = val;
527 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500528 "orangefs_ioctl: FS_IOC_GETFLAGS: %llu\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400529 (unsigned long long)uval);
530 return put_user(uval, (int __user *)arg);
531 } else if (cmd == FS_IOC_SETFLAGS) {
532 ret = 0;
533 if (get_user(uval, (int __user *)arg))
534 return -EFAULT;
535 /*
Yi Liu8bb8aef2015-11-24 15:12:14 -0500536 * ORANGEFS_MIRROR_FL is set internally when the mirroring mode
Mike Marshall5db11c22015-07-17 10:38:12 -0400537 * is turned on for a file. The user is not allowed to turn
538 * on this bit, but the bit is present if the user first gets
539 * the flags and then updates the flags with some new
540 * settings. So, we ignore it in the following edit. bligon.
541 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500542 if ((uval & ~ORANGEFS_MIRROR_FL) &
Mike Marshall5db11c22015-07-17 10:38:12 -0400543 (~(FS_IMMUTABLE_FL | FS_APPEND_FL | FS_NOATIME_FL))) {
Yi Liu8bb8aef2015-11-24 15:12:14 -0500544 gossip_err("orangefs_ioctl: the FS_IOC_SETFLAGS only supports setting one of FS_IMMUTABLE_FL|FS_APPEND_FL|FS_NOATIME_FL\n");
Mike Marshall5db11c22015-07-17 10:38:12 -0400545 return -EINVAL;
546 }
547 val = uval;
548 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500549 "orangefs_ioctl: FS_IOC_SETFLAGS: %llu\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400550 (unsigned long long)val);
Yi Liu8bb8aef2015-11-24 15:12:14 -0500551 ret = orangefs_inode_setxattr(file_inode(file),
552 ORANGEFS_XATTR_NAME_DEFAULT_PREFIX,
553 "user.pvfs2.meta_hint",
554 &val, sizeof(val), 0);
Mike Marshall5db11c22015-07-17 10:38:12 -0400555 }
556
557 return ret;
558}
559
560/*
561 * Memory map a region of a file.
562 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500563static int orangefs_file_mmap(struct file *file, struct vm_area_struct *vma)
Mike Marshall5db11c22015-07-17 10:38:12 -0400564{
565 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500566 "orangefs_file_mmap: called on %s\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400567 (file ?
568 (char *)file->f_path.dentry->d_name.name :
569 (char *)"Unknown"));
570
571 /* set the sequential readahead hint */
572 vma->vm_flags |= VM_SEQ_READ;
573 vma->vm_flags &= ~VM_RAND_READ;
Martin Brandenburg35390802015-09-30 13:11:54 -0400574
575 /* Use readonly mmap since we cannot support writable maps. */
576 return generic_file_readonly_mmap(file, vma);
Mike Marshall5db11c22015-07-17 10:38:12 -0400577}
578
579#define mapping_nrpages(idata) ((idata)->nrpages)
580
581/*
582 * Called to notify the module that there are no more references to
583 * this file (i.e. no processes have it open).
584 *
585 * \note Not called when each file is closed.
586 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500587static int orangefs_file_release(struct inode *inode, struct file *file)
Mike Marshall5db11c22015-07-17 10:38:12 -0400588{
589 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500590 "orangefs_file_release: called on %s\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400591 file->f_path.dentry->d_name.name);
592
Yi Liu8bb8aef2015-11-24 15:12:14 -0500593 orangefs_flush_inode(inode);
Mike Marshall5db11c22015-07-17 10:38:12 -0400594
595 /*
Mike Marshall54804942015-10-05 13:44:24 -0400596 * remove all associated inode pages from the page cache and mmap
597 * readahead cache (if any); this forces an expensive refresh of
598 * data for the next caller of mmap (or 'get_block' accesses)
Mike Marshall5db11c22015-07-17 10:38:12 -0400599 */
600 if (file->f_path.dentry->d_inode &&
601 file->f_path.dentry->d_inode->i_mapping &&
602 mapping_nrpages(&file->f_path.dentry->d_inode->i_data))
603 truncate_inode_pages(file->f_path.dentry->d_inode->i_mapping,
604 0);
605 return 0;
606}
607
608/*
609 * Push all data for a specific file onto permanent storage.
610 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500611static int orangefs_fsync(struct file *file,
Mike Marshall84d02152015-07-28 13:27:51 -0400612 loff_t start,
613 loff_t end,
614 int datasync)
Mike Marshall5db11c22015-07-17 10:38:12 -0400615{
616 int ret = -EINVAL;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500617 struct orangefs_inode_s *orangefs_inode =
618 ORANGEFS_I(file->f_path.dentry->d_inode);
619 struct orangefs_kernel_op_s *new_op = NULL;
Mike Marshall5db11c22015-07-17 10:38:12 -0400620
621 /* required call */
622 filemap_write_and_wait_range(file->f_mapping, start, end);
623
Yi Liu8bb8aef2015-11-24 15:12:14 -0500624 new_op = op_alloc(ORANGEFS_VFS_OP_FSYNC);
Mike Marshall5db11c22015-07-17 10:38:12 -0400625 if (!new_op)
626 return -ENOMEM;
Yi Liu8bb8aef2015-11-24 15:12:14 -0500627 new_op->upcall.req.fsync.refn = orangefs_inode->refn;
Mike Marshall5db11c22015-07-17 10:38:12 -0400628
629 ret = service_operation(new_op,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500630 "orangefs_fsync",
Mike Marshall5db11c22015-07-17 10:38:12 -0400631 get_interruptible_flag(file->f_path.dentry->d_inode));
632
633 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500634 "orangefs_fsync got return value of %d\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400635 ret);
636
637 op_release(new_op);
638
Yi Liu8bb8aef2015-11-24 15:12:14 -0500639 orangefs_flush_inode(file->f_path.dentry->d_inode);
Mike Marshall5db11c22015-07-17 10:38:12 -0400640 return ret;
641}
642
643/*
644 * Change the file pointer position for an instance of an open file.
645 *
646 * \note If .llseek is overriden, we must acquire lock as described in
647 * Documentation/filesystems/Locking.
648 *
649 * Future upgrade could support SEEK_DATA and SEEK_HOLE but would
650 * require much changes to the FS
651 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500652static loff_t orangefs_file_llseek(struct file *file, loff_t offset, int origin)
Mike Marshall5db11c22015-07-17 10:38:12 -0400653{
654 int ret = -EINVAL;
Al Viro177f8fc2016-02-16 20:25:19 -0500655 struct inode *inode = file_inode(file);
Mike Marshall5db11c22015-07-17 10:38:12 -0400656
Al Viro177f8fc2016-02-16 20:25:19 -0500657 if (origin == SEEK_END) {
Mike Marshall5db11c22015-07-17 10:38:12 -0400658 /*
659 * revalidate the inode's file size.
660 * NOTE: We are only interested in file size here,
661 * so we set mask accordingly.
662 */
Martin Brandenburge2f7f0d2016-03-15 12:33:20 -0400663 ret = orangefs_inode_getattr(file->f_mapping->host, 0, 1);
664 if (ret == -ESTALE)
665 ret = -EIO;
Mike Marshall5db11c22015-07-17 10:38:12 -0400666 if (ret) {
667 gossip_debug(GOSSIP_FILE_DEBUG,
668 "%s:%s:%d calling make bad inode\n",
669 __FILE__,
670 __func__,
671 __LINE__);
Mike Marshall5db11c22015-07-17 10:38:12 -0400672 return ret;
673 }
674 }
675
676 gossip_debug(GOSSIP_FILE_DEBUG,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500677 "orangefs_file_llseek: offset is %ld | origin is %d"
Mike Marshall54804942015-10-05 13:44:24 -0400678 " | inode size is %lu\n",
Mike Marshall5db11c22015-07-17 10:38:12 -0400679 (long)offset,
680 origin,
Al Viro177f8fc2016-02-16 20:25:19 -0500681 (unsigned long)i_size_read(inode));
Mike Marshall5db11c22015-07-17 10:38:12 -0400682
683 return generic_file_llseek(file, offset, origin);
684}
685
686/*
687 * Support local locks (locks that only this kernel knows about)
688 * if Orangefs was mounted -o local_lock.
689 */
Yi Liu8bb8aef2015-11-24 15:12:14 -0500690static int orangefs_lock(struct file *filp, int cmd, struct file_lock *fl)
Mike Marshall5db11c22015-07-17 10:38:12 -0400691{
Mike Marshallf957ae22015-09-24 12:53:05 -0400692 int rc = -EINVAL;
Mike Marshall5db11c22015-07-17 10:38:12 -0400693
Yi Liu8bb8aef2015-11-24 15:12:14 -0500694 if (ORANGEFS_SB(filp->f_inode->i_sb)->flags & ORANGEFS_OPT_LOCAL_LOCK) {
Mike Marshall5db11c22015-07-17 10:38:12 -0400695 if (cmd == F_GETLK) {
696 rc = 0;
697 posix_test_lock(filp, fl);
698 } else {
699 rc = posix_lock_file(filp, fl, NULL);
700 }
701 }
702
703 return rc;
704}
705
Yi Liu8bb8aef2015-11-24 15:12:14 -0500706/** ORANGEFS implementation of VFS file operations */
707const 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,
Mike Marshall5db11c22015-07-17 10:38:12 -0400714 .open = generic_file_open,
Yi Liu8bb8aef2015-11-24 15:12:14 -0500715 .release = orangefs_file_release,
716 .fsync = orangefs_fsync,
Mike Marshall5db11c22015-07-17 10:38:12 -0400717};