1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
9 * Copyright (C) 1992, 1993, 1994, 1995
10 * Remy Card (card@masi.ibp.fr)
11 * Laboratoire MASI - Institut Blaise Pascal
12 * Universite Pierre et Marie Curie (Paris VI)
16 * linux/fs/minix/file.c
18 * Copyright (C) 1991, 1992 Linus Torvalds
20 * ext2 fs regular file handling primitives
22 * 64-bit file support on 64-bit platforms by Jakub Jelinek
23 * (jj@sunsite.ms.mff.cuni.cz)
26 #define DEBUG_SUBSYSTEM S_LLITE
28 #include <linux/lustre_dlm.h>
29 #include <linux/lustre_lite.h>
31 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc);
32 extern int ll_setattr(struct dentry *de, struct iattr *attr);
34 static int ll_file_open(struct inode *inode, struct file *file)
37 struct ptlrpc_request *req = NULL;
38 struct ll_file_data *fd;
39 struct obdo *oa = NULL;
40 struct lov_stripe_md *md = NULL;
41 struct ll_sb_info *sbi = ll_i2sbi(inode);
42 struct ll_inode_info *lli = ll_i2info(inode);
45 if (file->private_data)
48 /* delayed create of object (intent created inode) */
49 /* XXX object needs to be cleaned up if mdc_open fails */
50 /* XXX error handling appropriate here? */
51 if (lli->lli_smd == NULL) {
52 struct client_obd *mdc = sbi2mdc(ll_s2sbi(inode->i_sb));
53 struct inode * inode = file->f_dentry->d_inode;
55 down(&lli->lli_open_sem);
56 /* Check to see if we lost the race */
57 if (lli->lli_smd == NULL) {
60 up(&lli->lli_open_sem);
63 oa->o_mode = S_IFREG | 0600;
64 oa->o_easize = mdc->cl_max_mds_easize;
65 oa->o_id = inode->i_ino;
66 oa->o_valid = OBD_MD_FLMODE | OBD_MD_FLEASIZE |
68 rc = obd_create(ll_i2obdconn(inode), oa, &lli->lli_smd);
72 up(&lli->lli_open_sem);
77 if (lli->lli_smd->lmd_object_id == 0)
79 up(&lli->lli_open_sem);
82 fd = kmem_cache_alloc(ll_file_data_slab, SLAB_KERNEL);
84 GOTO(out, rc = -ENOMEM);
85 memset(fd, 0, sizeof(*fd));
87 rc = mdc_open(&sbi->ll_mdc_conn, inode->i_ino, S_IFREG | inode->i_mode,
88 file->f_flags, md, (__u64)(unsigned long)file,
89 &fd->fd_mdshandle, &req);
91 ptlrpc_req_finished(req);
93 GOTO(out_req, -abs(rc));
94 if (!fd->fd_mdshandle) {
95 CERROR("mdc_open didn't assign fd_mdshandle\n");
96 /* XXX handle this how, abort or is it non-fatal? */
99 if (oa == NULL && (oa = obdo_alloc()) == NULL)
100 GOTO(out_mdc, rc = -EINVAL);
102 oa->o_id = lli->lli_smd->lmd_object_id;
103 oa->o_mode = S_IFREG | inode->i_mode;
104 oa->o_valid = OBD_MD_FLMODE | OBD_MD_FLID | OBD_MD_FLSIZE;
105 rc = obd_open(ll_i2obdconn(inode), oa, lli->lli_smd);
110 GOTO(out_mdc, rc = -abs(rc));
112 file->private_data = fd;
118 mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
119 S_IFREG, fd->fd_mdshandle, &req);
121 ptlrpc_free_req(req);
124 kmem_cache_free(ll_file_data_slab, fd);
125 file->private_data = NULL;
130 int ll_size_lock(struct inode *inode, struct lov_stripe_md *md, __u64 start,
131 int mode, struct lustre_handle **lockhs_p)
133 struct ll_sb_info *sbi = ll_i2sbi(inode);
134 struct ldlm_extent extent;
135 struct lustre_handle *lockhs = NULL;
138 if (sbi->ll_flags & LL_SBI_NOLCK) {
143 OBD_ALLOC(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
147 extent.start = start;
150 rc = obd_enqueue(&sbi->ll_osc_conn, md, NULL, LDLM_EXTENT, &extent,
151 sizeof(extent), mode, &flags, ll_lock_callback,
152 inode, sizeof(*inode), lockhs);
153 if (rc != ELDLM_OK) {
154 CERROR("lock enqueue: %d\n", rc);
155 OBD_FREE(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
161 int ll_size_unlock(struct inode *inode, struct lov_stripe_md *md, int mode,
162 struct lustre_handle *lockhs)
164 struct ll_sb_info *sbi = ll_i2sbi(inode);
167 if (sbi->ll_flags & LL_SBI_NOLCK)
170 if (lockhs == NULL) {
175 rc = obd_cancel(&sbi->ll_osc_conn, md, mode, lockhs);
176 if (rc != ELDLM_OK) {
177 CERROR("lock cancel: %d\n", rc);
181 OBD_FREE(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
185 int ll_file_size(struct inode *inode, struct lov_stripe_md *md)
187 struct ll_sb_info *sbi = ll_i2sbi(inode);
188 struct lustre_handle *lockhs;
192 rc = ll_size_lock(inode, md, 0, LCK_PR, &lockhs);
193 if (rc != ELDLM_OK) {
194 CERROR("lock enqueue: %d\n", rc);
198 oa.o_id = md->lmd_object_id;
200 oa.o_valid = OBD_MD_FLID|OBD_MD_FLMODE|OBD_MD_FLSIZE|OBD_MD_FLBLOCKS;
201 rc = obd_getattr(&sbi->ll_osc_conn, &oa, md);
203 obdo_to_inode(inode, &oa, oa.o_valid);
205 err = ll_size_unlock(inode, md, LCK_PR, lockhs);
206 if (err != ELDLM_OK) {
207 CERROR("lock cancel: %d\n", err);
213 static int ll_file_release(struct inode *inode, struct file *file)
216 struct ptlrpc_request *req = NULL;
217 struct ll_file_data *fd;
219 struct ll_sb_info *sbi = ll_i2sbi(inode);
220 struct ll_inode_info *lli = ll_i2info(inode);
224 fd = (struct ll_file_data *)file->private_data;
225 if (!fd || !fd->fd_mdshandle) {
227 GOTO(out, rc = -EINVAL);
230 memset(&oa, 0, sizeof(oa));
231 oa.o_id = lli->lli_smd->lmd_object_id;
233 oa.o_valid = (OBD_MD_FLMODE | OBD_MD_FLID);
234 rc = obd_close(ll_i2obdconn(inode), &oa, lli->lli_smd);
236 GOTO(out_fd, abs(rc));
238 /* If this fails and we goto out_fd, the file size on the MDS is out of
239 * date. Is that a big deal? */
240 if (file->f_mode & FMODE_WRITE) {
241 struct lustre_handle *lockhs;
243 rc = ll_size_lock(inode, lli->lli_smd, 0, LCK_PR, &lockhs);
245 GOTO(out_fd, abs(rc));
247 oa.o_id = lli->lli_smd->lmd_object_id;
249 oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLSIZE |
251 rc = obd_getattr(&sbi->ll_osc_conn, &oa, lli->lli_smd);
254 attr.ia_valid = (ATTR_MTIME | ATTR_CTIME | ATTR_ATIME |
256 attr.ia_mtime = inode->i_mtime;
257 attr.ia_ctime = inode->i_ctime;
258 attr.ia_atime = inode->i_atime;
259 attr.ia_size = oa.o_size;
261 inode->i_blocks = oa.o_blocks;
263 /* XXX: this introduces a small race that we should
265 rc = ll_inode_setattr(inode, &attr, 0);
267 CERROR("failed - %d.\n", rc);
268 rc = -EIO; /* XXX - GOTO(out)? -phil */
271 rc = ll_size_unlock(inode, lli->lli_smd, LCK_PR, lockhs);
273 CERROR("lock cancel: %d\n", rc);
278 rc = mdc_close(&sbi->ll_mdc_conn, inode->i_ino,
279 S_IFREG, fd->fd_mdshandle, &req);
280 ptlrpc_req_finished(req);
286 ptlrpc_req_finished(fd->fd_req);
288 //ldlm_cli_cancel_unused();
293 kmem_cache_free(ll_file_data_slab, fd);
294 file->private_data = NULL;
300 static inline void ll_remove_suid(struct inode *inode)
304 /* set S_IGID if S_IXGRP is set, and always set S_ISUID */
305 mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;
307 /* was any of the uid bits set? */
308 mode &= inode->i_mode;
309 if (mode && !capable(CAP_FSETID)) {
310 inode->i_mode &= ~mode;
311 // XXX careful here - we cannot change the size
315 static void ll_update_atime(struct inode *inode)
319 attr.ia_atime = CURRENT_TIME;
320 attr.ia_valid = ATTR_ATIME;
322 if (inode->i_atime == attr.ia_atime) return;
323 if (IS_RDONLY(inode)) return;
324 if (IS_NOATIME(inode)) return;
326 /* ll_inode_setattr() sets inode->i_atime from attr.ia_atime */
327 ll_inode_setattr(inode, &attr, 0);
330 int ll_lock_callback(struct ldlm_lock *lock, struct ldlm_lock_desc *new,
331 void *data, __u32 data_len)
333 struct inode *inode = data;
334 struct lustre_handle lockh;
338 if (data_len != sizeof(struct inode))
344 CDEBUG(D_INODE, "invalidating obdo/inode %ld\n", inode->i_ino);
345 /* FIXME: do something better than throwing away everything */
346 invalidate_inode_pages(inode);
349 ldlm_lock2handle(lock, &lockh);
350 rc = ldlm_cli_cancel(&lockh);
352 CERROR("ldlm_cli_cancel failed: %d\n", rc);
356 static ssize_t ll_file_read(struct file *filp, char *buf, size_t count,
359 struct ll_file_data *fd = (struct ll_file_data *)filp->private_data;
360 struct inode *inode = filp->f_dentry->d_inode;
361 struct ll_sb_info *sbi = ll_i2sbi(inode);
362 struct ldlm_extent extent;
363 struct lustre_handle *lockhs = NULL;
364 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
370 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
371 sbi->ll_flags & LL_SBI_NOLCK) {
372 OBD_ALLOC(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
376 extent.start = *ppos;
377 extent.end = *ppos + count;
378 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
379 inode->i_ino, extent.start, extent.end);
381 err = obd_enqueue(&sbi->ll_osc_conn, md, NULL, LDLM_EXTENT,
382 &extent, sizeof(extent), LCK_PR, &flags,
383 ll_lock_callback, inode, sizeof(*inode),
385 if (err != ELDLM_OK) {
386 OBD_FREE(lockhs, md->lmd_stripe_count *sizeof(*lockhs));
387 CERROR("lock enqueue: err: %d\n", err);
392 CDEBUG(D_INFO, "Reading inode %ld, %d bytes, offset %Ld\n",
393 inode->i_ino, count, *ppos);
394 retval = generic_file_read(filp, buf, count, ppos);
397 ll_update_atime(inode);
399 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
400 sbi->ll_flags & LL_SBI_NOLCK) {
401 err = obd_cancel(&sbi->ll_osc_conn, md, LCK_PR, lockhs);
402 if (err != ELDLM_OK) {
403 CERROR("lock cancel: err: %d\n", err);
409 OBD_FREE(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
414 * Write to a file (through the page cache).
417 ll_file_write(struct file *file, const char *buf, size_t count, loff_t *ppos)
419 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
420 struct inode *inode = file->f_dentry->d_inode;
421 struct ll_sb_info *sbi = ll_i2sbi(inode);
422 struct ldlm_extent extent;
423 struct lustre_handle *lockhs = NULL, *eof_lockhs = NULL;
424 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
430 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
432 err = ll_size_lock(inode, md, 0, LCK_PW, &eof_lockhs);
436 oa.o_id = md->lmd_object_id;
437 oa.o_mode = inode->i_mode;
438 oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLSIZE |
440 retval = obd_getattr(&sbi->ll_osc_conn, &oa, md);
442 GOTO(out_eof, retval);
445 obdo_to_inode(inode, &oa, oa.o_valid);
448 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
449 sbi->ll_flags & LL_SBI_NOLCK) {
450 OBD_ALLOC(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
452 GOTO(out_eof, retval = -ENOMEM);
453 extent.start = *ppos;
454 extent.end = *ppos + count;
455 CDEBUG(D_INFO, "Locking inode %ld, start "LPU64" end "LPU64"\n",
456 inode->i_ino, extent.start, extent.end);
458 err = obd_enqueue(&sbi->ll_osc_conn, md, NULL, LDLM_EXTENT,
459 &extent, sizeof(extent), LCK_PW, &flags,
460 ll_lock_callback, inode, sizeof(*inode),
462 if (err != ELDLM_OK) {
463 CERROR("lock enqueue: err: %d\n", err);
464 GOTO(out_free, retval = err);
468 CDEBUG(D_INFO, "Writing inode %ld, %ld bytes, offset %Ld\n",
469 inode->i_ino, (long)count, *ppos);
471 retval = generic_file_write(file, buf, count, ppos);
473 if (!(fd->fd_flags & LL_FILE_IGNORE_LOCK) ||
474 sbi->ll_flags & LL_SBI_NOLCK) {
475 err = obd_cancel(&sbi->ll_osc_conn, md, LCK_PW, lockhs);
476 if (err != ELDLM_OK) {
477 CERROR("lock cancel: err: %d\n", err);
478 GOTO(out_free, retval = err);
485 OBD_FREE(lockhs, md->lmd_stripe_count * sizeof(*lockhs));
488 if (!S_ISBLK(inode->i_mode) && file->f_flags & O_APPEND) {
489 err = ll_size_unlock(inode, md, LCK_PW, eof_lockhs);
497 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
500 struct ll_file_data *fd = (struct ll_file_data *)file->private_data;
504 case LL_IOC_GETFLAGS:
505 /* Get the current value of the file flags */
506 return put_user(fd->fd_flags, (int *)arg);
507 case LL_IOC_SETFLAGS:
508 case LL_IOC_CLRFLAGS:
509 /* Set or clear specific file flags */
510 /* XXX This probably needs checks to ensure the flags are
511 * not abused, and to handle any flag side effects.
513 if (get_user(flags, (int *) arg))
516 if (cmd == LL_IOC_SETFLAGS)
517 fd->fd_flags |= flags;
519 fd->fd_flags &= ~flags;
522 /* We need to special case any other ioctls we want to handle,
523 * to send them to the MDS/OST as appropriate and to properly
524 * network encode the arg field.
525 case EXT2_IOC_GETFLAGS:
526 case EXT2_IOC_SETFLAGS:
527 case EXT2_IOC_GETVERSION_OLD:
528 case EXT2_IOC_GETVERSION_NEW:
529 case EXT2_IOC_SETVERSION_OLD:
530 case EXT2_IOC_SETVERSION_NEW:
537 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
539 struct inode *inode = file->f_dentry->d_inode;
545 struct ll_inode_info *lli = ll_i2info(inode);
547 retval = ll_file_size(inode, lli->lli_smd);
551 offset += inode->i_size;
555 offset += file->f_pos;
558 if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
559 if (offset != file->f_pos) {
560 file->f_pos = offset;
562 file->f_version = ++event;
569 /* XXX this does not need to do anything for data, it _does_ need to
571 int ll_fsync(struct file *file, struct dentry *dentry, int data)
576 struct file_operations ll_file_operations = {
578 write: ll_file_write,
579 ioctl: ll_file_ioctl,
581 release: ll_file_release,
582 mmap: generic_file_mmap,
583 llseek: ll_file_seek,
587 struct inode_operations ll_file_inode_operations = {
588 truncate: ll_truncate,