1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 * Author: Peter Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
40 * Author: Andreas Dilger <adilger@clusterfs.com>
43 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <lustre_dlm.h>
45 #include <lustre_lite.h>
46 #include <lustre_mdc.h>
47 #include <linux/pagemap.h>
48 #include <linux/file.h>
49 #include "llite_internal.h"
50 #include <lustre/ll_fiemap.h>
52 #include "cl_object.h"
54 struct ll_file_data *ll_file_data_get(void)
56 struct ll_file_data *fd;
58 OBD_SLAB_ALLOC_PTR_GFP(fd, ll_file_data_slab, CFS_ALLOC_IO);
62 static void ll_file_data_put(struct ll_file_data *fd)
65 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
68 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
69 struct lustre_handle *fh)
71 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
72 op_data->op_attr.ia_mode = inode->i_mode;
73 op_data->op_attr.ia_atime = inode->i_atime;
74 op_data->op_attr.ia_mtime = inode->i_mtime;
75 op_data->op_attr.ia_ctime = inode->i_ctime;
76 op_data->op_attr.ia_size = i_size_read(inode);
77 op_data->op_attr_blocks = inode->i_blocks;
78 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags =
79 ll_inode_to_ext_flags(inode->i_flags);
80 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
82 op_data->op_handle = *fh;
83 op_data->op_capa1 = ll_mdscapa_get(inode);
87 * Closes the IO epoch and packs all the attributes into @op_data for
90 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
91 struct obd_client_handle *och)
95 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
96 ATTR_MTIME_SET | ATTR_CTIME_SET;
98 if (!(och->och_flags & FMODE_WRITE))
101 if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
102 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
104 ll_ioepoch_close(inode, op_data, &och, 0);
107 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
108 ll_prep_md_op_data(op_data, inode, NULL, NULL,
109 0, 0, LUSTRE_OPC_ANY, NULL);
113 static int ll_close_inode_openhandle(struct obd_export *md_exp,
115 struct obd_client_handle *och)
117 struct obd_export *exp = ll_i2mdexp(inode);
118 struct md_op_data *op_data;
119 struct ptlrpc_request *req = NULL;
120 struct obd_device *obd = class_exp2obd(exp);
127 * XXX: in case of LMV, is this correct to access
130 CERROR("Invalid MDC connection handle "LPX64"\n",
131 ll_i2mdexp(inode)->exp_handle.h_cookie);
135 OBD_ALLOC_PTR(op_data);
137 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
139 ll_prepare_close(inode, op_data, och);
140 epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE);
141 rc = md_close(md_exp, op_data, och->och_mod, &req);
143 /* This close must have the epoch closed. */
144 LASSERT(epoch_close);
145 /* MDS has instructed us to obtain Size-on-MDS attribute from
146 * OSTs and send setattr to back to MDS. */
147 rc = ll_som_update(inode, op_data);
149 CERROR("inode %lu mdc Size-on-MDS update failed: "
150 "rc = %d\n", inode->i_ino, rc);
154 CERROR("inode %lu mdc close failed: rc = %d\n",
157 ll_finish_md_op_data(op_data);
160 rc = ll_objects_destroy(req, inode);
162 CERROR("inode %lu ll_objects destroy: rc = %d\n",
169 if (exp_connect_som(exp) && !epoch_close &&
170 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
171 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
173 md_clear_open_replay_data(md_exp, och);
174 /* Free @och if it is not waiting for DONE_WRITING. */
175 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
178 if (req) /* This is close request */
179 ptlrpc_req_finished(req);
183 int ll_md_real_close(struct inode *inode, int flags)
185 struct ll_inode_info *lli = ll_i2info(inode);
186 struct obd_client_handle **och_p;
187 struct obd_client_handle *och;
192 if (flags & FMODE_WRITE) {
193 och_p = &lli->lli_mds_write_och;
194 och_usecount = &lli->lli_open_fd_write_count;
195 } else if (flags & FMODE_EXEC) {
196 och_p = &lli->lli_mds_exec_och;
197 och_usecount = &lli->lli_open_fd_exec_count;
199 LASSERT(flags & FMODE_READ);
200 och_p = &lli->lli_mds_read_och;
201 och_usecount = &lli->lli_open_fd_read_count;
204 cfs_down(&lli->lli_och_sem);
205 if (*och_usecount) { /* There are still users of this handle, so
207 cfs_up(&lli->lli_och_sem);
212 cfs_up(&lli->lli_och_sem);
214 if (och) { /* There might be a race and somebody have freed this och
216 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
223 int ll_md_close(struct obd_export *md_exp, struct inode *inode,
226 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
227 struct ll_inode_info *lli = ll_i2info(inode);
231 /* clear group lock, if present */
232 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED))
233 ll_put_grouplock(inode, file, fd->fd_grouplock.cg_gid);
235 /* Let's see if we have good enough OPEN lock on the file and if
236 we can skip talking to MDS */
237 if (file->f_dentry->d_inode) { /* Can this ever be false? */
239 int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
240 struct lustre_handle lockh;
241 struct inode *inode = file->f_dentry->d_inode;
242 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
244 cfs_down(&lli->lli_och_sem);
245 if (fd->fd_omode & FMODE_WRITE) {
247 LASSERT(lli->lli_open_fd_write_count);
248 lli->lli_open_fd_write_count--;
249 } else if (fd->fd_omode & FMODE_EXEC) {
251 LASSERT(lli->lli_open_fd_exec_count);
252 lli->lli_open_fd_exec_count--;
255 LASSERT(lli->lli_open_fd_read_count);
256 lli->lli_open_fd_read_count--;
258 cfs_up(&lli->lli_och_sem);
260 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
261 LDLM_IBITS, &policy, lockmode,
263 rc = ll_md_real_close(file->f_dentry->d_inode,
267 CERROR("Releasing a file %p with negative dentry %p. Name %s",
268 file, file->f_dentry, file->f_dentry->d_name.name);
271 LUSTRE_FPRIVATE(file) = NULL;
272 ll_file_data_put(fd);
273 ll_capa_close(inode);
278 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm);
280 /* While this returns an error code, fput() the caller does not, so we need
281 * to make every effort to clean up all of our state here. Also, applications
282 * rarely check close errors and even if an error is returned they will not
283 * re-try the close call.
285 int ll_file_release(struct inode *inode, struct file *file)
287 struct ll_file_data *fd;
288 struct ll_sb_info *sbi = ll_i2sbi(inode);
289 struct ll_inode_info *lli = ll_i2info(inode);
290 struct lov_stripe_md *lsm = lli->lli_smd;
294 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
295 inode->i_generation, inode);
297 #ifdef CONFIG_FS_POSIX_ACL
298 if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
299 inode == inode->i_sb->s_root->d_inode) {
300 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
303 if (unlikely(fd->fd_flags & LL_FILE_RMTACL)) {
304 fd->fd_flags &= ~LL_FILE_RMTACL;
305 rct_del(&sbi->ll_rct, cfs_curproc_pid());
306 et_search_free(&sbi->ll_et, cfs_curproc_pid());
311 if (inode->i_sb->s_root != file->f_dentry)
312 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
313 fd = LUSTRE_FPRIVATE(file);
316 /* The last ref on @file, maybe not the the owner pid of statahead.
317 * Different processes can open the same dir, "ll_opendir_key" means:
318 * it is me that should stop the statahead thread. */
319 if (lli->lli_opendir_key == fd && lli->lli_opendir_pid != 0)
320 ll_stop_statahead(inode, lli->lli_opendir_key);
322 if (inode->i_sb->s_root == file->f_dentry) {
323 LUSTRE_FPRIVATE(file) = NULL;
324 ll_file_data_put(fd);
329 lov_test_and_clear_async_rc(lsm);
330 lli->lli_async_rc = 0;
332 rc = ll_md_close(sbi->ll_md_exp, inode, file);
334 if (OBD_FAIL_TIMEOUT_MS(OBD_FAIL_PTLRPC_DUMP_LOG, obd_fail_val))
335 libcfs_debug_dumplog();
340 static int ll_intent_file_open(struct file *file, void *lmm,
341 int lmmsize, struct lookup_intent *itp)
343 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
344 struct dentry *parent = file->f_dentry->d_parent;
345 const char *name = file->f_dentry->d_name.name;
346 const int len = file->f_dentry->d_name.len;
347 struct md_op_data *op_data;
348 struct ptlrpc_request *req;
355 /* Usually we come here only for NFSD, and we want open lock.
356 But we can also get here with pre 2.6.15 patchless kernels, and in
357 that case that lock is also ok */
358 /* We can also get here if there was cached open handle in revalidate_it
359 * but it disappeared while we were getting from there to ll_file_open.
360 * But this means this file was closed and immediatelly opened which
361 * makes a good candidate for using OPEN lock */
362 /* If lmmsize & lmm are not 0, we are just setting stripe info
363 * parameters. No need for the open lock */
364 if (!lmm && !lmmsize)
365 itp->it_flags |= MDS_OPEN_LOCK;
367 op_data = ll_prep_md_op_data(NULL, parent->d_inode,
368 file->f_dentry->d_inode, name, len,
369 O_RDWR, LUSTRE_OPC_ANY, NULL);
371 RETURN(PTR_ERR(op_data));
373 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
374 0 /*unused */, &req, ll_md_blocking_ast, 0);
375 ll_finish_md_op_data(op_data);
377 /* reason for keep own exit path - don`t flood log
378 * with messages with -ESTALE errors.
380 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
381 it_open_error(DISP_OPEN_OPEN, itp))
383 ll_release_openhandle(file->f_dentry, itp);
387 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
388 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
389 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
393 rc = ll_prep_inode(&file->f_dentry->d_inode, req, NULL);
394 if (!rc && itp->d.lustre.it_lock_mode)
395 md_set_lock_data(sbi->ll_md_exp,
396 &itp->d.lustre.it_lock_handle,
397 file->f_dentry->d_inode, NULL);
400 ptlrpc_req_finished(itp->d.lustre.it_data);
401 it_clear_disposition(itp, DISP_ENQ_COMPLETE);
402 ll_intent_drop_lock(itp);
408 * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
409 * not believe attributes if a few ioepoch holders exist. Attributes for
410 * previous ioepoch if new one is opened are also skipped by MDS.
412 void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
414 if (ioepoch && lli->lli_ioepoch != ioepoch) {
415 lli->lli_ioepoch = ioepoch;
416 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
417 ioepoch, PFID(&lli->lli_fid));
421 static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli,
422 struct lookup_intent *it, struct obd_client_handle *och)
424 struct ptlrpc_request *req = it->d.lustre.it_data;
425 struct mdt_body *body;
429 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
430 LASSERT(body != NULL); /* reply already checked out */
432 memcpy(&och->och_fh, &body->handle, sizeof(body->handle));
433 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
434 och->och_fid = lli->lli_fid;
435 och->och_flags = it->it_flags;
436 ll_ioepoch_open(lli, body->ioepoch);
438 return md_set_open_replay_data(md_exp, och, req);
441 int ll_local_open(struct file *file, struct lookup_intent *it,
442 struct ll_file_data *fd, struct obd_client_handle *och)
444 struct inode *inode = file->f_dentry->d_inode;
445 struct ll_inode_info *lli = ll_i2info(inode);
448 LASSERT(!LUSTRE_FPRIVATE(file));
453 struct ptlrpc_request *req = it->d.lustre.it_data;
454 struct mdt_body *body;
457 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, lli, it, och);
461 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
462 if ((it->it_flags & FMODE_WRITE) &&
463 (body->valid & OBD_MD_FLSIZE))
464 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
465 lli->lli_ioepoch, PFID(&lli->lli_fid));
468 LUSTRE_FPRIVATE(file) = fd;
469 ll_readahead_init(inode, &fd->fd_ras);
470 fd->fd_omode = it->it_flags;
474 /* Open a file, and (for the very first open) create objects on the OSTs at
475 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
476 * creation or open until ll_lov_setstripe() ioctl is called. We grab
477 * lli_open_sem to ensure no other process will create objects, send the
478 * stripe MD to the MDS, or try to destroy the objects if that fails.
480 * If we already have the stripe MD locally then we don't request it in
481 * md_open(), by passing a lmm_size = 0.
483 * It is up to the application to ensure no other processes open this file
484 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
485 * used. We might be able to avoid races of that sort by getting lli_open_sem
486 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
487 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
489 int ll_file_open(struct inode *inode, struct file *file)
491 struct ll_inode_info *lli = ll_i2info(inode);
492 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
493 .it_flags = file->f_flags };
494 struct lov_stripe_md *lsm;
495 struct ptlrpc_request *req = NULL;
496 struct obd_client_handle **och_p;
498 struct ll_file_data *fd;
499 int rc = 0, opendir_set = 0;
502 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino,
503 inode->i_generation, inode, file->f_flags);
505 #ifdef HAVE_VFS_INTENT_PATCHES
508 it = file->private_data; /* XXX: compat macro */
509 file->private_data = NULL; /* prevent ll_local_open assertion */
512 fd = ll_file_data_get();
517 if (S_ISDIR(inode->i_mode)) {
518 cfs_spin_lock(&lli->lli_sa_lock);
519 if (lli->lli_opendir_key == NULL && lli->lli_opendir_pid == 0) {
520 LASSERT(lli->lli_sai == NULL);
521 lli->lli_opendir_key = fd;
522 lli->lli_opendir_pid = cfs_curproc_pid();
525 cfs_spin_unlock(&lli->lli_sa_lock);
528 if (inode->i_sb->s_root == file->f_dentry) {
529 LUSTRE_FPRIVATE(file) = fd;
533 if (!it || !it->d.lustre.it_disposition) {
534 /* Convert f_flags into access mode. We cannot use file->f_mode,
535 * because everything but O_ACCMODE mask was stripped from
537 if ((oit.it_flags + 1) & O_ACCMODE)
539 if (file->f_flags & O_TRUNC)
540 oit.it_flags |= FMODE_WRITE;
542 /* kernel only call f_op->open in dentry_open. filp_open calls
543 * dentry_open after call to open_namei that checks permissions.
544 * Only nfsd_open call dentry_open directly without checking
545 * permissions and because of that this code below is safe. */
546 if (oit.it_flags & FMODE_WRITE)
547 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
549 /* We do not want O_EXCL here, presumably we opened the file
550 * already? XXX - NFS implications? */
551 oit.it_flags &= ~O_EXCL;
553 /* bug20584, if "it_flags" contains O_CREAT, the file will be
554 * created if necessary, then "IT_CREAT" should be set to keep
555 * consistent with it */
556 if (oit.it_flags & O_CREAT)
557 oit.it_op |= IT_CREAT;
563 /* Let's see if we have file open on MDS already. */
564 if (it->it_flags & FMODE_WRITE) {
565 och_p = &lli->lli_mds_write_och;
566 och_usecount = &lli->lli_open_fd_write_count;
567 } else if (it->it_flags & FMODE_EXEC) {
568 och_p = &lli->lli_mds_exec_och;
569 och_usecount = &lli->lli_open_fd_exec_count;
571 och_p = &lli->lli_mds_read_och;
572 och_usecount = &lli->lli_open_fd_read_count;
575 cfs_down(&lli->lli_och_sem);
576 if (*och_p) { /* Open handle is present */
577 if (it_disposition(it, DISP_OPEN_OPEN)) {
578 /* Well, there's extra open request that we do not need,
579 let's close it somehow. This will decref request. */
580 rc = it_open_error(DISP_OPEN_OPEN, it);
582 cfs_up(&lli->lli_och_sem);
583 ll_file_data_put(fd);
584 GOTO(out_openerr, rc);
586 ll_release_openhandle(file->f_dentry, it);
587 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
592 rc = ll_local_open(file, it, fd, NULL);
595 cfs_up(&lli->lli_och_sem);
596 ll_file_data_put(fd);
597 GOTO(out_openerr, rc);
600 LASSERT(*och_usecount == 0);
601 if (!it->d.lustre.it_disposition) {
602 /* We cannot just request lock handle now, new ELC code
603 means that one of other OPEN locks for this file
604 could be cancelled, and since blocking ast handler
605 would attempt to grab och_sem as well, that would
606 result in a deadlock */
607 cfs_up(&lli->lli_och_sem);
608 it->it_create_mode |= M_CHECK_STALE;
609 rc = ll_intent_file_open(file, NULL, 0, it);
610 it->it_create_mode &= ~M_CHECK_STALE;
612 ll_file_data_put(fd);
613 GOTO(out_openerr, rc);
616 /* Got some error? Release the request */
617 if (it->d.lustre.it_status < 0) {
618 req = it->d.lustre.it_data;
619 ptlrpc_req_finished(req);
623 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
625 ll_file_data_put(fd);
626 GOTO(out_och_free, rc = -ENOMEM);
629 req = it->d.lustre.it_data;
631 /* md_intent_lock() didn't get a request ref if there was an
632 * open error, so don't do cleanup on the request here
634 /* XXX (green): Should not we bail out on any error here, not
635 * just open error? */
636 rc = it_open_error(DISP_OPEN_OPEN, it);
638 ll_file_data_put(fd);
639 GOTO(out_och_free, rc);
642 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
643 rc = ll_local_open(file, it, fd, *och_p);
645 ll_file_data_put(fd);
646 GOTO(out_och_free, rc);
649 cfs_up(&lli->lli_och_sem);
651 /* Must do this outside lli_och_sem lock to prevent deadlock where
652 different kind of OPEN lock for this same inode gets cancelled
653 by ldlm_cancel_lru */
654 if (!S_ISREG(inode->i_mode))
661 if (file->f_flags & O_LOV_DELAY_CREATE ||
662 !(file->f_mode & FMODE_WRITE)) {
663 CDEBUG(D_INODE, "object creation was delayed\n");
667 file->f_flags &= ~O_LOV_DELAY_CREATE;
670 ptlrpc_req_finished(req);
672 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
676 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
677 *och_p = NULL; /* OBD_FREE writes some magic there */
680 cfs_up(&lli->lli_och_sem);
682 if (opendir_set != 0)
683 ll_stop_statahead(inode, lli->lli_opendir_key);
689 /* Fills the obdo with the attributes for the lsm */
690 static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
691 struct obd_capa *capa, struct obdo *obdo,
692 __u64 ioepoch, int sync)
694 struct ptlrpc_request_set *set;
695 struct obd_info oinfo = { { { 0 } } };
700 LASSERT(lsm != NULL);
704 oinfo.oi_oa->o_id = lsm->lsm_object_id;
705 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
706 oinfo.oi_oa->o_mode = S_IFREG;
707 oinfo.oi_oa->o_ioepoch = ioepoch;
708 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
709 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
710 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
711 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
712 OBD_MD_FLGROUP | OBD_MD_FLEPOCH;
713 oinfo.oi_capa = capa;
715 oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
716 oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
719 set = ptlrpc_prep_set();
721 CERROR("can't allocate ptlrpc set\n");
724 rc = obd_getattr_async(exp, &oinfo, set);
726 rc = ptlrpc_set_wait(set);
727 ptlrpc_set_destroy(set);
730 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
731 OBD_MD_FLATIME | OBD_MD_FLMTIME |
732 OBD_MD_FLCTIME | OBD_MD_FLSIZE);
737 * Performs the getattr on the inode and updates its fields.
738 * If @sync != 0, perform the getattr under the server-side lock.
740 int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
741 __u64 ioepoch, int sync)
743 struct ll_inode_info *lli = ll_i2info(inode);
744 struct obd_capa *capa = ll_mdscapa_get(inode);
748 rc = ll_lsm_getattr(lli->lli_smd, ll_i2dtexp(inode),
749 capa, obdo, ioepoch, sync);
752 obdo_refresh_inode(inode, obdo, obdo->o_valid);
754 "objid "LPX64" size %Lu, blocks %llu, blksize %lu\n",
755 lli->lli_smd->lsm_object_id, i_size_read(inode),
756 (unsigned long long)inode->i_blocks,
757 (unsigned long)ll_inode_blksize(inode));
762 int ll_merge_lvb(struct inode *inode)
764 struct ll_inode_info *lli = ll_i2info(inode);
765 struct ll_sb_info *sbi = ll_i2sbi(inode);
771 ll_inode_size_lock(inode, 1);
772 inode_init_lvb(inode, &lvb);
774 /* merge timestamps the most resently obtained from mds with
775 timestamps obtained from osts */
776 lvb.lvb_atime = lli->lli_lvb.lvb_atime;
777 lvb.lvb_mtime = lli->lli_lvb.lvb_mtime;
778 lvb.lvb_ctime = lli->lli_lvb.lvb_ctime;
779 rc = obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
780 cl_isize_write_nolock(inode, lvb.lvb_size);
782 CDEBUG(D_VFSTRACE, DFID" updating i_size "LPU64"\n",
783 PFID(&lli->lli_fid), lvb.lvb_size);
784 inode->i_blocks = lvb.lvb_blocks;
786 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
787 LTIME_S(inode->i_atime) = lvb.lvb_atime;
788 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
789 ll_inode_size_unlock(inode, 1);
794 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
797 struct obdo obdo = { 0 };
800 rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, NULL, &obdo, 0, 0);
802 st->st_size = obdo.o_size;
803 st->st_blocks = obdo.o_blocks;
804 st->st_mtime = obdo.o_mtime;
805 st->st_atime = obdo.o_atime;
806 st->st_ctime = obdo.o_ctime;
811 void ll_io_init(struct cl_io *io, const struct file *file, int write)
813 struct inode *inode = file->f_dentry->d_inode;
815 memset(io, 0, sizeof *io);
816 io->u.ci_rw.crw_nonblock = file->f_flags & O_NONBLOCK;
818 io->u.ci_wr.wr_append = !!(file->f_flags & O_APPEND);
819 io->ci_obj = ll_i2info(inode)->lli_clob;
820 io->ci_lockreq = CILR_MAYBE;
821 if (ll_file_nolock(file)) {
822 io->ci_lockreq = CILR_NEVER;
823 io->ci_no_srvlock = 1;
824 } else if (file->f_flags & O_APPEND) {
825 io->ci_lockreq = CILR_MANDATORY;
829 static ssize_t ll_file_io_generic(const struct lu_env *env,
830 struct vvp_io_args *args, struct file *file,
831 enum cl_io_type iot, loff_t *ppos, size_t count)
833 struct ll_inode_info *lli = ll_i2info(file->f_dentry->d_inode);
838 io = &ccc_env_info(env)->cti_io;
839 ll_io_init(io, file, iot == CIT_WRITE);
841 if (cl_io_rw_init(env, io, iot, *ppos, count) == 0) {
842 struct vvp_io *vio = vvp_env_io(env);
843 struct ccc_io *cio = ccc_env_io(env);
844 int write_sem_locked = 0;
846 cio->cui_fd = LUSTRE_FPRIVATE(file);
847 vio->cui_io_subtype = args->via_io_subtype;
849 switch (vio->cui_io_subtype) {
851 cio->cui_iov = args->u.normal.via_iov;
852 cio->cui_nrsegs = args->u.normal.via_nrsegs;
853 cio->cui_tot_nrsegs = cio->cui_nrsegs;
854 #ifndef HAVE_FILE_WRITEV
855 cio->cui_iocb = args->u.normal.via_iocb;
857 if ((iot == CIT_WRITE) &&
858 !(cio->cui_fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
859 if(cfs_down_interruptible(&lli->lli_write_sem))
860 GOTO(out, result = -ERESTARTSYS);
861 write_sem_locked = 1;
862 } else if (iot == CIT_READ) {
863 cfs_down_read(&lli->lli_trunc_sem);
867 vio->u.sendfile.cui_actor = args->u.sendfile.via_actor;
868 vio->u.sendfile.cui_target = args->u.sendfile.via_target;
871 vio->u.splice.cui_pipe = args->u.splice.via_pipe;
872 vio->u.splice.cui_flags = args->u.splice.via_flags;
875 CERROR("Unknow IO type - %u\n", vio->cui_io_subtype);
878 result = cl_io_loop(env, io);
879 if (write_sem_locked)
880 cfs_up(&lli->lli_write_sem);
881 else if (args->via_io_subtype == IO_NORMAL && iot == CIT_READ)
882 cfs_up_read(&lli->lli_trunc_sem);
884 /* cl_io_rw_init() handled IO */
885 result = io->ci_result;
888 if (io->ci_nob > 0) {
890 *ppos = io->u.ci_wr.wr.crw_pos;
895 if (iot == CIT_WRITE)
896 lli->lli_write_rc = result < 0 ? : 0;
902 * XXX: exact copy from kernel code (__generic_file_aio_write_nolock)
904 static int ll_file_get_iov_count(const struct iovec *iov,
905 unsigned long *nr_segs, size_t *count)
910 for (seg = 0; seg < *nr_segs; seg++) {
911 const struct iovec *iv = &iov[seg];
914 * If any segment has a negative length, or the cumulative
915 * length ever wraps negative then return -EINVAL.
918 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
920 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
925 cnt -= iv->iov_len; /* This segment is no good */
932 #ifdef HAVE_FILE_READV
933 static ssize_t ll_file_readv(struct file *file, const struct iovec *iov,
934 unsigned long nr_segs, loff_t *ppos)
937 struct vvp_io_args *args;
943 result = ll_file_get_iov_count(iov, &nr_segs, &count);
947 env = cl_env_get(&refcheck);
949 RETURN(PTR_ERR(env));
951 args = vvp_env_args(env, IO_NORMAL);
952 args->u.normal.via_iov = (struct iovec *)iov;
953 args->u.normal.via_nrsegs = nr_segs;
955 result = ll_file_io_generic(env, args, file, CIT_READ, ppos, count);
956 cl_env_put(env, &refcheck);
960 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
964 struct iovec *local_iov;
969 env = cl_env_get(&refcheck);
971 RETURN(PTR_ERR(env));
973 local_iov = &vvp_env_info(env)->vti_local_iov;
974 local_iov->iov_base = (void __user *)buf;
975 local_iov->iov_len = count;
976 result = ll_file_readv(file, local_iov, 1, ppos);
977 cl_env_put(env, &refcheck);
982 static ssize_t ll_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
983 unsigned long nr_segs, loff_t pos)
986 struct vvp_io_args *args;
992 result = ll_file_get_iov_count(iov, &nr_segs, &count);
996 env = cl_env_get(&refcheck);
998 RETURN(PTR_ERR(env));
1000 args = vvp_env_args(env, IO_NORMAL);
1001 args->u.normal.via_iov = (struct iovec *)iov;
1002 args->u.normal.via_nrsegs = nr_segs;
1003 args->u.normal.via_iocb = iocb;
1005 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_READ,
1006 &iocb->ki_pos, count);
1007 cl_env_put(env, &refcheck);
1011 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1015 struct iovec *local_iov;
1016 struct kiocb *kiocb;
1021 env = cl_env_get(&refcheck);
1023 RETURN(PTR_ERR(env));
1025 local_iov = &vvp_env_info(env)->vti_local_iov;
1026 kiocb = &vvp_env_info(env)->vti_kiocb;
1027 local_iov->iov_base = (void __user *)buf;
1028 local_iov->iov_len = count;
1029 init_sync_kiocb(kiocb, file);
1030 kiocb->ki_pos = *ppos;
1031 kiocb->ki_left = count;
1033 result = ll_file_aio_read(kiocb, local_iov, 1, kiocb->ki_pos);
1034 *ppos = kiocb->ki_pos;
1036 cl_env_put(env, &refcheck);
1042 * Write to a file (through the page cache).
1044 #ifdef HAVE_FILE_WRITEV
1045 static ssize_t ll_file_writev(struct file *file, const struct iovec *iov,
1046 unsigned long nr_segs, loff_t *ppos)
1049 struct vvp_io_args *args;
1055 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1059 env = cl_env_get(&refcheck);
1061 RETURN(PTR_ERR(env));
1063 args = vvp_env_args(env, IO_NORMAL);
1064 args->u.normal.via_iov = (struct iovec *)iov;
1065 args->u.normal.via_nrsegs = nr_segs;
1067 result = ll_file_io_generic(env, args, file, CIT_WRITE, ppos, count);
1068 cl_env_put(env, &refcheck);
1072 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1076 struct iovec *local_iov;
1081 env = cl_env_get(&refcheck);
1083 RETURN(PTR_ERR(env));
1085 local_iov = &vvp_env_info(env)->vti_local_iov;
1086 local_iov->iov_base = (void __user *)buf;
1087 local_iov->iov_len = count;
1089 result = ll_file_writev(file, local_iov, 1, ppos);
1090 cl_env_put(env, &refcheck);
1094 #else /* AIO stuff */
1095 static ssize_t ll_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
1096 unsigned long nr_segs, loff_t pos)
1099 struct vvp_io_args *args;
1105 result = ll_file_get_iov_count(iov, &nr_segs, &count);
1109 env = cl_env_get(&refcheck);
1111 RETURN(PTR_ERR(env));
1113 args = vvp_env_args(env, IO_NORMAL);
1114 args->u.normal.via_iov = (struct iovec *)iov;
1115 args->u.normal.via_nrsegs = nr_segs;
1116 args->u.normal.via_iocb = iocb;
1118 result = ll_file_io_generic(env, args, iocb->ki_filp, CIT_WRITE,
1119 &iocb->ki_pos, count);
1120 cl_env_put(env, &refcheck);
1124 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1128 struct iovec *local_iov;
1129 struct kiocb *kiocb;
1134 env = cl_env_get(&refcheck);
1136 RETURN(PTR_ERR(env));
1138 local_iov = &vvp_env_info(env)->vti_local_iov;
1139 kiocb = &vvp_env_info(env)->vti_kiocb;
1140 local_iov->iov_base = (void __user *)buf;
1141 local_iov->iov_len = count;
1142 init_sync_kiocb(kiocb, file);
1143 kiocb->ki_pos = *ppos;
1144 kiocb->ki_left = count;
1146 result = ll_file_aio_write(kiocb, local_iov, 1, kiocb->ki_pos);
1147 *ppos = kiocb->ki_pos;
1149 cl_env_put(env, &refcheck);
1155 #ifdef HAVE_KERNEL_SENDFILE
1157 * Send file content (through pagecache) somewhere with helper
1159 static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
1160 read_actor_t actor, void *target)
1163 struct vvp_io_args *args;
1168 env = cl_env_get(&refcheck);
1170 RETURN(PTR_ERR(env));
1172 args = vvp_env_args(env, IO_SENDFILE);
1173 args->u.sendfile.via_target = target;
1174 args->u.sendfile.via_actor = actor;
1176 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1177 cl_env_put(env, &refcheck);
1182 #ifdef HAVE_KERNEL_SPLICE_READ
1184 * Send file content (through pagecache) somewhere with helper
1186 static ssize_t ll_file_splice_read(struct file *in_file, loff_t *ppos,
1187 struct pipe_inode_info *pipe, size_t count,
1191 struct vvp_io_args *args;
1196 env = cl_env_get(&refcheck);
1198 RETURN(PTR_ERR(env));
1200 args = vvp_env_args(env, IO_SPLICE);
1201 args->u.splice.via_pipe = pipe;
1202 args->u.splice.via_flags = flags;
1204 result = ll_file_io_generic(env, args, in_file, CIT_READ, ppos, count);
1205 cl_env_put(env, &refcheck);
1210 static int ll_lov_recreate(struct inode *inode, obd_id id, obd_seq seq,
1213 struct obd_export *exp = ll_i2dtexp(inode);
1214 struct obd_trans_info oti = { 0 };
1215 struct obdo *oa = NULL;
1218 struct lov_stripe_md *lsm, *lsm2;
1225 ll_inode_size_lock(inode, 0);
1226 lsm = ll_i2info(inode)->lli_smd;
1228 GOTO(out, rc = -ENOENT);
1229 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1230 (lsm->lsm_stripe_count));
1232 OBD_ALLOC_LARGE(lsm2, lsm_size);
1234 GOTO(out, rc = -ENOMEM);
1238 oa->o_nlink = ost_idx;
1239 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1240 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1241 obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid, OBD_MD_FLTYPE |
1242 OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1243 memcpy(lsm2, lsm, lsm_size);
1244 rc = obd_create(exp, oa, &lsm2, &oti);
1246 OBD_FREE_LARGE(lsm2, lsm_size);
1249 ll_inode_size_unlock(inode, 0);
1254 static int ll_lov_recreate_obj(struct inode *inode, unsigned long arg)
1256 struct ll_recreate_obj ucreat;
1259 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1262 if (cfs_copy_from_user(&ucreat, (struct ll_recreate_obj *)arg,
1263 sizeof(struct ll_recreate_obj)))
1266 RETURN(ll_lov_recreate(inode, ucreat.lrc_id, 0,
1267 ucreat.lrc_ost_idx));
1270 static int ll_lov_recreate_fid(struct inode *inode, unsigned long arg)
1277 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1280 if (cfs_copy_from_user(&fid, (struct lu_fid *)arg,
1281 sizeof(struct lu_fid)))
1284 id = fid_oid(&fid) | ((fid_seq(&fid) & 0xffff) << 32);
1285 ost_idx = (fid_seq(&fid) >> 16) & 0xffff;
1286 RETURN(ll_lov_recreate(inode, id, 0, ost_idx));
1289 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1290 int flags, struct lov_user_md *lum, int lum_size)
1292 struct lov_stripe_md *lsm;
1293 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1297 ll_inode_size_lock(inode, 0);
1298 lsm = ll_i2info(inode)->lli_smd;
1300 ll_inode_size_unlock(inode, 0);
1301 CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n",
1306 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1309 if (it_disposition(&oit, DISP_LOOKUP_NEG))
1310 GOTO(out_req_free, rc = -ENOENT);
1311 rc = oit.d.lustre.it_status;
1313 GOTO(out_req_free, rc);
1315 ll_release_openhandle(file->f_dentry, &oit);
1318 ll_inode_size_unlock(inode, 0);
1319 ll_intent_release(&oit);
1322 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1326 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1327 struct lov_mds_md **lmmp, int *lmm_size,
1328 struct ptlrpc_request **request)
1330 struct ll_sb_info *sbi = ll_i2sbi(inode);
1331 struct mdt_body *body;
1332 struct lov_mds_md *lmm = NULL;
1333 struct ptlrpc_request *req = NULL;
1334 struct md_op_data *op_data;
1337 rc = ll_get_max_mdsize(sbi, &lmmsize);
1341 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename,
1342 strlen(filename), lmmsize,
1343 LUSTRE_OPC_ANY, NULL);
1344 if (op_data == NULL)
1347 op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
1348 rc = md_getattr_name(sbi->ll_md_exp, op_data, &req);
1349 ll_finish_md_op_data(op_data);
1351 CDEBUG(D_INFO, "md_getattr_name failed "
1352 "on %s: rc %d\n", filename, rc);
1356 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1357 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1359 lmmsize = body->eadatasize;
1361 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1363 GOTO(out, rc = -ENODATA);
1366 lmm = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD, lmmsize);
1367 LASSERT(lmm != NULL);
1369 if ((lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V1)) &&
1370 (lmm->lmm_magic != cpu_to_le32(LOV_MAGIC_V3))) {
1371 GOTO(out, rc = -EPROTO);
1375 * This is coming from the MDS, so is probably in
1376 * little endian. We convert it to host endian before
1377 * passing it to userspace.
1379 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC)) {
1380 /* if function called for directory - we should
1381 * avoid swab not existent lsm objects */
1382 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V1)) {
1383 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1384 if (S_ISREG(body->mode))
1385 lustre_swab_lov_user_md_objects(
1386 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1387 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1388 } else if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_V3)) {
1389 lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1390 if (S_ISREG(body->mode))
1391 lustre_swab_lov_user_md_objects(
1392 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1393 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1399 *lmm_size = lmmsize;
1404 static int ll_lov_setea(struct inode *inode, struct file *file,
1407 int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1408 struct lov_user_md *lump;
1409 int lum_size = sizeof(struct lov_user_md) +
1410 sizeof(struct lov_user_ost_data);
1414 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1417 OBD_ALLOC_LARGE(lump, lum_size);
1421 if (cfs_copy_from_user(lump, (struct lov_user_md *)arg, lum_size)) {
1422 OBD_FREE_LARGE(lump, lum_size);
1426 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1428 OBD_FREE_LARGE(lump, lum_size);
1432 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1435 struct lov_user_md_v3 lumv3;
1436 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1437 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1438 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1441 int flags = FMODE_WRITE;
1444 /* first try with v1 which is smaller than v3 */
1445 lum_size = sizeof(struct lov_user_md_v1);
1446 if (cfs_copy_from_user(lumv1, lumv1p, lum_size))
1449 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1450 lum_size = sizeof(struct lov_user_md_v3);
1451 if (cfs_copy_from_user(&lumv3, lumv3p, lum_size))
1455 rc = ll_lov_setstripe_ea_info(inode, file, flags, lumv1, lum_size);
1457 put_user(0, &lumv1p->lmm_stripe_count);
1458 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1459 0, ll_i2info(inode)->lli_smd,
1465 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1467 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1472 return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm,
1476 int ll_get_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1478 struct ll_inode_info *lli = ll_i2info(inode);
1479 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1480 struct ccc_grouplock grouplock;
1484 if (ll_file_nolock(file))
1485 RETURN(-EOPNOTSUPP);
1487 cfs_spin_lock(&lli->lli_lock);
1488 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1489 CWARN("group lock already existed with gid %lu\n",
1490 fd->fd_grouplock.cg_gid);
1491 cfs_spin_unlock(&lli->lli_lock);
1494 LASSERT(fd->fd_grouplock.cg_lock == NULL);
1495 cfs_spin_unlock(&lli->lli_lock);
1497 rc = cl_get_grouplock(cl_i2info(inode)->lli_clob,
1498 arg, (file->f_flags & O_NONBLOCK), &grouplock);
1502 cfs_spin_lock(&lli->lli_lock);
1503 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1504 cfs_spin_unlock(&lli->lli_lock);
1505 CERROR("another thread just won the race\n");
1506 cl_put_grouplock(&grouplock);
1510 fd->fd_flags |= LL_FILE_GROUP_LOCKED;
1511 fd->fd_grouplock = grouplock;
1512 cfs_spin_unlock(&lli->lli_lock);
1514 CDEBUG(D_INFO, "group lock %lu obtained\n", arg);
1518 int ll_put_grouplock(struct inode *inode, struct file *file, unsigned long arg)
1520 struct ll_inode_info *lli = ll_i2info(inode);
1521 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1522 struct ccc_grouplock grouplock;
1525 cfs_spin_lock(&lli->lli_lock);
1526 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1527 cfs_spin_unlock(&lli->lli_lock);
1528 CWARN("no group lock held\n");
1531 LASSERT(fd->fd_grouplock.cg_lock != NULL);
1533 if (fd->fd_grouplock.cg_gid != arg) {
1534 CWARN("group lock %lu doesn't match current id %lu\n",
1535 arg, fd->fd_grouplock.cg_gid);
1536 cfs_spin_unlock(&lli->lli_lock);
1540 grouplock = fd->fd_grouplock;
1541 memset(&fd->fd_grouplock, 0, sizeof(fd->fd_grouplock));
1542 fd->fd_flags &= ~LL_FILE_GROUP_LOCKED;
1543 cfs_spin_unlock(&lli->lli_lock);
1545 cl_put_grouplock(&grouplock);
1546 CDEBUG(D_INFO, "group lock %lu released\n", arg);
1551 * Close inode open handle
1553 * \param dentry [in] dentry which contains the inode
1554 * \param it [in,out] intent which contains open info and result
1557 * \retval <0 failure
1559 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
1561 struct inode *inode = dentry->d_inode;
1562 struct obd_client_handle *och;
1568 /* Root ? Do nothing. */
1569 if (dentry->d_inode->i_sb->s_root == dentry)
1572 /* No open handle to close? Move away */
1573 if (!it_disposition(it, DISP_OPEN_OPEN))
1576 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
1578 OBD_ALLOC(och, sizeof(*och));
1580 GOTO(out, rc = -ENOMEM);
1582 ll_och_fill(ll_i2sbi(inode)->ll_md_exp,
1583 ll_i2info(inode), it, och);
1585 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
1588 /* this one is in place of ll_file_open */
1589 if (it_disposition(it, DISP_ENQ_OPEN_REF))
1590 ptlrpc_req_finished(it->d.lustre.it_data);
1591 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
1596 * Get size for inode for which FIEMAP mapping is requested.
1597 * Make the FIEMAP get_info call and returns the result.
1599 int ll_do_fiemap(struct inode *inode, struct ll_user_fiemap *fiemap,
1602 struct obd_export *exp = ll_i2dtexp(inode);
1603 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1604 struct ll_fiemap_info_key fm_key = { .name = KEY_FIEMAP, };
1605 int vallen = num_bytes;
1609 /* Checks for fiemap flags */
1610 if (fiemap->fm_flags & ~LUSTRE_FIEMAP_FLAGS_COMPAT) {
1611 fiemap->fm_flags &= ~LUSTRE_FIEMAP_FLAGS_COMPAT;
1615 /* Check for FIEMAP_FLAG_SYNC */
1616 if (fiemap->fm_flags & FIEMAP_FLAG_SYNC) {
1617 rc = filemap_fdatawrite(inode->i_mapping);
1622 /* If the stripe_count > 1 and the application does not understand
1623 * DEVICE_ORDER flag, then it cannot interpret the extents correctly.
1625 if (lsm->lsm_stripe_count > 1 &&
1626 !(fiemap->fm_flags & FIEMAP_FLAG_DEVICE_ORDER))
1629 fm_key.oa.o_id = lsm->lsm_object_id;
1630 fm_key.oa.o_seq = lsm->lsm_object_seq;
1631 fm_key.oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1633 obdo_from_inode(&fm_key.oa, inode, &ll_i2info(inode)->lli_fid,
1635 /* If filesize is 0, then there would be no objects for mapping */
1636 if (fm_key.oa.o_size == 0) {
1637 fiemap->fm_mapped_extents = 0;
1641 memcpy(&fm_key.fiemap, fiemap, sizeof(*fiemap));
1643 rc = obd_get_info(exp, sizeof(fm_key), &fm_key, &vallen, fiemap, lsm);
1645 CERROR("obd_get_info failed: rc = %d\n", rc);
1650 int ll_fid2path(struct obd_export *exp, void *arg)
1652 struct getinfo_fid2path *gfout, *gfin;
1656 /* Need to get the buflen */
1657 OBD_ALLOC_PTR(gfin);
1660 if (cfs_copy_from_user(gfin, arg, sizeof(*gfin))) {
1665 outsize = sizeof(*gfout) + gfin->gf_pathlen;
1666 OBD_ALLOC(gfout, outsize);
1667 if (gfout == NULL) {
1671 memcpy(gfout, gfin, sizeof(*gfout));
1674 /* Call mdc_iocontrol */
1675 rc = obd_iocontrol(OBD_IOC_FID2PATH, exp, outsize, gfout, NULL);
1678 if (cfs_copy_to_user(arg, gfout, outsize))
1682 OBD_FREE(gfout, outsize);
1686 static int ll_ioctl_fiemap(struct inode *inode, unsigned long arg)
1688 struct ll_user_fiemap *fiemap_s;
1689 size_t num_bytes, ret_bytes;
1690 unsigned int extent_count;
1693 /* Get the extent count so we can calculate the size of
1694 * required fiemap buffer */
1695 if (get_user(extent_count,
1696 &((struct ll_user_fiemap __user *)arg)->fm_extent_count))
1698 num_bytes = sizeof(*fiemap_s) + (extent_count *
1699 sizeof(struct ll_fiemap_extent));
1701 OBD_ALLOC_LARGE(fiemap_s, num_bytes);
1702 if (fiemap_s == NULL)
1705 /* get the fiemap value */
1706 if (copy_from_user(fiemap_s,(struct ll_user_fiemap __user *)arg,
1708 GOTO(error, rc = -EFAULT);
1710 /* If fm_extent_count is non-zero, read the first extent since
1711 * it is used to calculate end_offset and device from previous
1714 if (copy_from_user(&fiemap_s->fm_extents[0],
1715 (char __user *)arg + sizeof(*fiemap_s),
1716 sizeof(struct ll_fiemap_extent)))
1717 GOTO(error, rc = -EFAULT);
1720 rc = ll_do_fiemap(inode, fiemap_s, num_bytes);
1724 ret_bytes = sizeof(struct ll_user_fiemap);
1726 if (extent_count != 0)
1727 ret_bytes += (fiemap_s->fm_mapped_extents *
1728 sizeof(struct ll_fiemap_extent));
1730 if (copy_to_user((void *)arg, fiemap_s, ret_bytes))
1734 OBD_FREE_LARGE(fiemap_s, num_bytes);
1738 #ifdef HAVE_UNLOCKED_IOCTL
1739 long ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1741 struct inode *inode = file->f_dentry->d_inode;
1743 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
1747 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1751 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
1752 inode->i_generation, inode, cmd);
1753 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1755 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1756 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1760 case LL_IOC_GETFLAGS:
1761 /* Get the current value of the file flags */
1762 return put_user(fd->fd_flags, (int *)arg);
1763 case LL_IOC_SETFLAGS:
1764 case LL_IOC_CLRFLAGS:
1765 /* Set or clear specific file flags */
1766 /* XXX This probably needs checks to ensure the flags are
1767 * not abused, and to handle any flag side effects.
1769 if (get_user(flags, (int *) arg))
1772 if (cmd == LL_IOC_SETFLAGS) {
1773 if ((flags & LL_FILE_IGNORE_LOCK) &&
1774 !(file->f_flags & O_DIRECT)) {
1775 CERROR("%s: unable to disable locking on "
1776 "non-O_DIRECT file\n", current->comm);
1780 fd->fd_flags |= flags;
1782 fd->fd_flags &= ~flags;
1785 case LL_IOC_LOV_SETSTRIPE:
1786 RETURN(ll_lov_setstripe(inode, file, arg));
1787 case LL_IOC_LOV_SETEA:
1788 RETURN(ll_lov_setea(inode, file, arg));
1789 case LL_IOC_LOV_GETSTRIPE:
1790 RETURN(ll_lov_getstripe(inode, arg));
1791 case LL_IOC_RECREATE_OBJ:
1792 RETURN(ll_lov_recreate_obj(inode, arg));
1793 case LL_IOC_RECREATE_FID:
1794 RETURN(ll_lov_recreate_fid(inode, arg));
1795 case FSFILT_IOC_FIEMAP:
1796 RETURN(ll_ioctl_fiemap(inode, arg));
1797 case FSFILT_IOC_GETFLAGS:
1798 case FSFILT_IOC_SETFLAGS:
1799 RETURN(ll_iocontrol(inode, file, cmd, arg));
1800 case FSFILT_IOC_GETVERSION_OLD:
1801 case FSFILT_IOC_GETVERSION:
1802 RETURN(put_user(inode->i_generation, (int *)arg));
1803 case LL_IOC_GROUP_LOCK:
1804 RETURN(ll_get_grouplock(inode, file, arg));
1805 case LL_IOC_GROUP_UNLOCK:
1806 RETURN(ll_put_grouplock(inode, file, arg));
1807 case IOC_OBD_STATFS:
1808 RETURN(ll_obd_statfs(inode, (void *)arg));
1810 /* We need to special case any other ioctls we want to handle,
1811 * to send them to the MDS/OST as appropriate and to properly
1812 * network encode the arg field.
1813 case FSFILT_IOC_SETVERSION_OLD:
1814 case FSFILT_IOC_SETVERSION:
1816 case LL_IOC_FLUSHCTX:
1817 RETURN(ll_flush_ctx(inode));
1818 case LL_IOC_PATH2FID: {
1819 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1820 sizeof(struct lu_fid)))
1825 case OBD_IOC_FID2PATH:
1826 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1828 case LL_IOC_GET_MDTIDX: {
1831 mdtidx = ll_get_mdt_idx(inode);
1835 if (put_user((int)mdtidx, (int*)arg))
1845 ll_iocontrol_call(inode, file, cmd, arg, &err))
1848 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
1854 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
1856 struct inode *inode = file->f_dentry->d_inode;
1859 retval = offset + ((origin == 2) ? i_size_read(inode) :
1860 (origin == 1) ? file->f_pos : 0);
1861 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%Lu=%#Lx(%s)\n",
1862 inode->i_ino, inode->i_generation, inode, retval, retval,
1863 origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET");
1864 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
1866 if (origin == 2) { /* SEEK_END */
1867 int nonblock = 0, rc;
1869 if (file->f_flags & O_NONBLOCK)
1870 nonblock = LDLM_FL_BLOCK_NOWAIT;
1872 rc = cl_glimpse_size(inode);
1876 offset += i_size_read(inode);
1877 } else if (origin == 1) { /* SEEK_CUR */
1878 offset += file->f_pos;
1882 if (offset >= 0 && offset <= ll_file_maxbytes(inode)) {
1883 if (offset != file->f_pos) {
1884 file->f_pos = offset;
1892 #ifdef HAVE_FLUSH_OWNER_ID
1893 int ll_flush(struct file *file, fl_owner_t id)
1895 int ll_flush(struct file *file)
1898 struct inode *inode = file->f_dentry->d_inode;
1899 struct ll_inode_info *lli = ll_i2info(inode);
1900 struct lov_stripe_md *lsm = lli->lli_smd;
1903 /* the application should know write failure already. */
1904 if (lli->lli_write_rc)
1907 /* catch async errors that were recorded back when async writeback
1908 * failed for pages in this mapping. */
1909 rc = lli->lli_async_rc;
1910 lli->lli_async_rc = 0;
1912 err = lov_test_and_clear_async_rc(lsm);
1917 return rc ? -EIO : 0;
1920 int ll_fsync(struct file *file, struct dentry *dentry, int data)
1922 struct inode *inode = dentry->d_inode;
1923 struct ll_inode_info *lli = ll_i2info(inode);
1924 struct lov_stripe_md *lsm = lli->lli_smd;
1925 struct ptlrpc_request *req;
1926 struct obd_capa *oc;
1929 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1930 inode->i_generation, inode);
1931 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
1933 /* fsync's caller has already called _fdata{sync,write}, we want
1934 * that IO to finish before calling the osc and mdc sync methods */
1935 rc = filemap_fdatawait(inode->i_mapping);
1937 /* catch async errors that were recorded back when async writeback
1938 * failed for pages in this mapping. */
1939 err = lli->lli_async_rc;
1940 lli->lli_async_rc = 0;
1944 err = lov_test_and_clear_async_rc(lsm);
1949 oc = ll_mdscapa_get(inode);
1950 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
1956 ptlrpc_req_finished(req);
1963 RETURN(rc ? rc : -ENOMEM);
1965 oa->o_id = lsm->lsm_object_id;
1966 oa->o_seq = lsm->lsm_object_seq;
1967 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1968 obdo_from_inode(oa, inode, &ll_i2info(inode)->lli_fid,
1969 OBD_MD_FLTYPE | OBD_MD_FLATIME |
1970 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1973 oc = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
1974 err = obd_sync(ll_i2sbi(inode)->ll_dt_exp, oa, lsm,
1975 0, OBD_OBJECT_EOF, oc);
1980 lli->lli_write_rc = err < 0 ? : 0;
1986 int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
1988 struct inode *inode = file->f_dentry->d_inode;
1989 struct ll_sb_info *sbi = ll_i2sbi(inode);
1990 struct ldlm_enqueue_info einfo = { .ei_type = LDLM_FLOCK,
1991 .ei_cb_cp =ldlm_flock_completion_ast,
1992 .ei_cbdata = file_lock };
1993 struct md_op_data *op_data;
1994 struct lustre_handle lockh = {0};
1995 ldlm_policy_data_t flock;
2000 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n",
2001 inode->i_ino, file_lock);
2003 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2005 if (file_lock->fl_flags & FL_FLOCK) {
2006 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2007 /* set missing params for flock() calls */
2008 file_lock->fl_end = OFFSET_MAX;
2009 file_lock->fl_pid = current->tgid;
2011 flock.l_flock.pid = file_lock->fl_pid;
2012 flock.l_flock.start = file_lock->fl_start;
2013 flock.l_flock.end = file_lock->fl_end;
2015 switch (file_lock->fl_type) {
2017 einfo.ei_mode = LCK_PR;
2020 /* An unlock request may or may not have any relation to
2021 * existing locks so we may not be able to pass a lock handle
2022 * via a normal ldlm_lock_cancel() request. The request may even
2023 * unlock a byte range in the middle of an existing lock. In
2024 * order to process an unlock request we need all of the same
2025 * information that is given with a normal read or write record
2026 * lock request. To avoid creating another ldlm unlock (cancel)
2027 * message we'll treat a LCK_NL flock request as an unlock. */
2028 einfo.ei_mode = LCK_NL;
2031 einfo.ei_mode = LCK_PW;
2034 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
2049 flags = LDLM_FL_BLOCK_NOWAIT;
2055 flags = LDLM_FL_TEST_LOCK;
2056 /* Save the old mode so that if the mode in the lock changes we
2057 * can decrement the appropriate reader or writer refcount. */
2058 file_lock->fl_type = einfo.ei_mode;
2061 CERROR("unknown fcntl lock command: %d\n", cmd);
2065 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2066 LUSTRE_OPC_ANY, NULL);
2067 if (IS_ERR(op_data))
2068 RETURN(PTR_ERR(op_data));
2070 CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, "
2071 "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
2072 flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2074 rc = md_enqueue(sbi->ll_md_exp, &einfo, NULL,
2075 op_data, &lockh, &flock, 0, NULL /* req */, flags);
2077 ll_finish_md_op_data(op_data);
2079 if ((file_lock->fl_flags & FL_FLOCK) &&
2080 (rc == 0 || file_lock->fl_type == F_UNLCK))
2081 ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW));
2082 #ifdef HAVE_F_OP_FLOCK
2083 if ((file_lock->fl_flags & FL_POSIX) &&
2084 (rc == 0 || file_lock->fl_type == F_UNLCK) &&
2085 !(flags & LDLM_FL_TEST_LOCK))
2086 posix_lock_file_wait(file, file_lock);
2092 int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2099 int ll_have_md_lock(struct inode *inode, __u64 bits, ldlm_mode_t l_req_mode)
2101 struct lustre_handle lockh;
2102 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2103 ldlm_mode_t mode = (l_req_mode == LCK_MINMODE) ?
2104 (LCK_CR|LCK_CW|LCK_PR|LCK_PW) : l_req_mode;
2112 fid = &ll_i2info(inode)->lli_fid;
2113 CDEBUG(D_INFO, "trying to match res "DFID" mode %s\n", PFID(fid),
2114 ldlm_lockname[mode]);
2116 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2117 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
2124 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
2125 struct lustre_handle *lockh)
2127 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2133 fid = &ll_i2info(inode)->lli_fid;
2134 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2136 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
2137 rc = md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
2138 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
2142 static int ll_inode_revalidate_fini(struct inode *inode, int rc) {
2143 if (rc == -ENOENT) { /* Already unlinked. Just update nlink
2144 * and return success */
2146 /* This path cannot be hit for regular files unless in
2147 * case of obscure races, so no need to to validate
2149 if (!S_ISREG(inode->i_mode) &&
2150 !S_ISDIR(inode->i_mode))
2155 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2163 int __ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it,
2166 struct inode *inode = dentry->d_inode;
2167 struct ptlrpc_request *req = NULL;
2168 struct ll_sb_info *sbi;
2169 struct obd_export *exp;
2174 CERROR("REPORT THIS LINE TO PETER\n");
2177 sbi = ll_i2sbi(inode);
2179 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n",
2180 inode->i_ino, inode->i_generation, inode, dentry->d_name.name);
2182 exp = ll_i2mdexp(inode);
2184 /* XXX: Enable OBD_CONNECT_ATTRFID to reduce unnecessary getattr RPC.
2185 * But under CMD case, it caused some lock issues, should be fixed
2186 * with new CMD ibits lock. See bug 12718 */
2187 if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) {
2188 struct lookup_intent oit = { .it_op = IT_GETATTR };
2189 struct md_op_data *op_data;
2191 /* Call getattr by fid, so do not provide name at all. */
2192 op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode,
2193 dentry->d_inode, NULL, 0, 0,
2194 LUSTRE_OPC_ANY, NULL);
2195 if (IS_ERR(op_data))
2196 RETURN(PTR_ERR(op_data));
2198 oit.it_create_mode |= M_CHECK_STALE;
2199 rc = md_intent_lock(exp, op_data, NULL, 0,
2200 /* we are not interested in name
2203 ll_md_blocking_ast, 0);
2204 ll_finish_md_op_data(op_data);
2205 oit.it_create_mode &= ~M_CHECK_STALE;
2207 rc = ll_inode_revalidate_fini(inode, rc);
2211 rc = ll_revalidate_it_finish(req, &oit, dentry);
2213 ll_intent_release(&oit);
2217 /* Unlinked? Unhash dentry, so it is not picked up later by
2218 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2219 here to preserve get_cwd functionality on 2.6.
2221 if (!dentry->d_inode->i_nlink) {
2222 cfs_spin_lock(&ll_lookup_lock);
2223 spin_lock(&dcache_lock);
2224 ll_drop_dentry(dentry);
2225 spin_unlock(&dcache_lock);
2226 cfs_spin_unlock(&ll_lookup_lock);
2229 ll_lookup_finish_locks(&oit, dentry);
2230 } else if (!ll_have_md_lock(dentry->d_inode, ibits, LCK_MINMODE)) {
2231 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
2232 obd_valid valid = OBD_MD_FLGETATTR;
2233 struct md_op_data *op_data;
2236 if (S_ISREG(inode->i_mode)) {
2237 rc = ll_get_max_mdsize(sbi, &ealen);
2240 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2243 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2244 0, ealen, LUSTRE_OPC_ANY,
2246 if (op_data == NULL)
2249 op_data->op_valid = valid;
2250 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
2251 * capa for this inode. Because we only keep capas of dirs
2253 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2254 ll_finish_md_op_data(op_data);
2256 rc = ll_inode_revalidate_fini(inode, rc);
2260 rc = ll_prep_inode(&inode, req, NULL);
2263 ptlrpc_req_finished(req);
2267 int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it)
2269 struct inode *inode = dentry->d_inode;
2273 rc = __ll_inode_revalidate_it(dentry, it, MDS_INODELOCK_UPDATE |
2274 MDS_INODELOCK_LOOKUP);
2276 /* if object not yet allocated, don't validate size */
2277 if (rc == 0 && ll_i2info(dentry->d_inode)->lli_smd == NULL) {
2278 LTIME_S(inode->i_atime) = ll_i2info(inode)->lli_lvb.lvb_atime;
2279 LTIME_S(inode->i_mtime) = ll_i2info(inode)->lli_lvb.lvb_mtime;
2280 LTIME_S(inode->i_ctime) = ll_i2info(inode)->lli_lvb.lvb_ctime;
2284 /* cl_glimpse_size will prefer locally cached writes if they extend
2288 rc = cl_glimpse_size(inode);
2293 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
2294 struct lookup_intent *it, struct kstat *stat)
2296 struct inode *inode = de->d_inode;
2297 struct ll_inode_info *lli = ll_i2info(inode);
2300 res = ll_inode_revalidate_it(de, it);
2301 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETATTR, 1);
2306 stat->dev = inode->i_sb->s_dev;
2307 if (ll_need_32bit_api(ll_i2sbi(inode)))
2308 stat->ino = cl_fid_build_ino32(&lli->lli_fid);
2310 stat->ino = inode->i_ino;
2312 stat->mode = inode->i_mode;
2313 stat->nlink = inode->i_nlink;
2314 stat->uid = inode->i_uid;
2315 stat->gid = inode->i_gid;
2316 stat->rdev = kdev_t_to_nr(inode->i_rdev);
2317 stat->atime = inode->i_atime;
2318 stat->mtime = inode->i_mtime;
2319 stat->ctime = inode->i_ctime;
2320 #ifdef HAVE_INODE_BLKSIZE
2321 stat->blksize = inode->i_blksize;
2323 stat->blksize = 1 << inode->i_blkbits;
2326 stat->size = i_size_read(inode);
2327 stat->blocks = inode->i_blocks;
2331 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
2333 struct lookup_intent it = { .it_op = IT_GETATTR };
2335 return ll_getattr_it(mnt, de, &it, stat);
2338 #ifdef HAVE_LINUX_FIEMAP_H
2339 int ll_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
2340 __u64 start, __u64 len)
2344 struct ll_user_fiemap *fiemap;
2345 unsigned int extent_count = fieinfo->fi_extents_max;
2347 num_bytes = sizeof(*fiemap) + (extent_count *
2348 sizeof(struct ll_fiemap_extent));
2349 OBD_ALLOC_LARGE(fiemap, num_bytes);
2354 fiemap->fm_flags = fieinfo->fi_flags;
2355 fiemap->fm_extent_count = fieinfo->fi_extents_max;
2356 fiemap->fm_start = start;
2357 fiemap->fm_length = len;
2358 memcpy(&fiemap->fm_extents[0], fieinfo->fi_extents_start,
2359 sizeof(struct ll_fiemap_extent));
2361 rc = ll_do_fiemap(inode, fiemap, num_bytes);
2363 fieinfo->fi_flags = fiemap->fm_flags;
2364 fieinfo->fi_extents_mapped = fiemap->fm_mapped_extents;
2365 memcpy(fieinfo->fi_extents_start, &fiemap->fm_extents[0],
2366 fiemap->fm_mapped_extents * sizeof(struct ll_fiemap_extent));
2368 OBD_FREE_LARGE(fiemap, num_bytes);
2375 int lustre_check_acl(struct inode *inode, int mask)
2377 #ifdef CONFIG_FS_POSIX_ACL
2378 struct ll_inode_info *lli = ll_i2info(inode);
2379 struct posix_acl *acl;
2383 cfs_spin_lock(&lli->lli_lock);
2384 acl = posix_acl_dup(lli->lli_posix_acl);
2385 cfs_spin_unlock(&lli->lli_lock);
2390 rc = posix_acl_permission(inode, acl, mask);
2391 posix_acl_release(acl);
2399 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10))
2400 #ifndef HAVE_INODE_PERMISION_2ARGS
2401 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2403 int ll_inode_permission(struct inode *inode, int mask)
2409 /* as root inode are NOT getting validated in lookup operation,
2410 * need to do it before permission check. */
2412 if (inode == inode->i_sb->s_root->d_inode) {
2413 struct lookup_intent it = { .it_op = IT_LOOKUP };
2415 rc = __ll_inode_revalidate_it(inode->i_sb->s_root, &it,
2416 MDS_INODELOCK_LOOKUP);
2421 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), inode mode %x mask %o\n",
2422 inode->i_ino, inode->i_generation, inode, inode->i_mode, mask);
2424 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2425 return lustre_check_remote_perm(inode, mask);
2427 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2428 rc = generic_permission(inode, mask, lustre_check_acl);
2433 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2435 int mode = inode->i_mode;
2438 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n",
2439 inode->i_ino, inode->i_generation, inode, mask);
2441 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2442 return lustre_check_remote_perm(inode, mask);
2444 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2446 if ((mask & MAY_WRITE) && IS_RDONLY(inode) &&
2447 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
2449 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
2451 if (cfs_curproc_fsuid() == inode->i_uid) {
2454 if (((mode >> 3) & mask & S_IRWXO) != mask)
2456 rc = lustre_check_acl(inode, mask);
2460 goto check_capabilities;
2464 if (cfs_curproc_is_in_groups(inode->i_gid))
2467 if ((mode & mask & S_IRWXO) == mask)
2471 if (!(mask & MAY_EXEC) ||
2472 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
2473 if (cfs_capable(CFS_CAP_DAC_OVERRIDE))
2476 if (cfs_capable(CFS_CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) ||
2477 (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))))
2484 #ifdef HAVE_FILE_READV
2485 #define READ_METHOD readv
2486 #define READ_FUNCTION ll_file_readv
2487 #define WRITE_METHOD writev
2488 #define WRITE_FUNCTION ll_file_writev
2490 #define READ_METHOD aio_read
2491 #define READ_FUNCTION ll_file_aio_read
2492 #define WRITE_METHOD aio_write
2493 #define WRITE_FUNCTION ll_file_aio_write
2496 /* -o localflock - only provides locally consistent flock locks */
2497 struct file_operations ll_file_operations = {
2498 .read = ll_file_read,
2499 .READ_METHOD = READ_FUNCTION,
2500 .write = ll_file_write,
2501 .WRITE_METHOD = WRITE_FUNCTION,
2502 #ifdef HAVE_UNLOCKED_IOCTL
2503 .unlocked_ioctl = ll_file_ioctl,
2505 .ioctl = ll_file_ioctl,
2507 .open = ll_file_open,
2508 .release = ll_file_release,
2509 .mmap = ll_file_mmap,
2510 .llseek = ll_file_seek,
2511 #ifdef HAVE_KERNEL_SENDFILE
2512 .sendfile = ll_file_sendfile,
2514 #ifdef HAVE_KERNEL_SPLICE_READ
2515 .splice_read = ll_file_splice_read,
2521 struct file_operations ll_file_operations_flock = {
2522 .read = ll_file_read,
2523 .READ_METHOD = READ_FUNCTION,
2524 .write = ll_file_write,
2525 .WRITE_METHOD = WRITE_FUNCTION,
2526 #ifdef HAVE_UNLOCKED_IOCTL
2527 .unlocked_ioctl = ll_file_ioctl,
2529 .ioctl = ll_file_ioctl,
2531 .open = ll_file_open,
2532 .release = ll_file_release,
2533 .mmap = ll_file_mmap,
2534 .llseek = ll_file_seek,
2535 #ifdef HAVE_KERNEL_SENDFILE
2536 .sendfile = ll_file_sendfile,
2538 #ifdef HAVE_KERNEL_SPLICE_READ
2539 .splice_read = ll_file_splice_read,
2543 #ifdef HAVE_F_OP_FLOCK
2544 .flock = ll_file_flock,
2546 .lock = ll_file_flock
2549 /* These are for -o noflock - to return ENOSYS on flock calls */
2550 struct file_operations ll_file_operations_noflock = {
2551 .read = ll_file_read,
2552 .READ_METHOD = READ_FUNCTION,
2553 .write = ll_file_write,
2554 .WRITE_METHOD = WRITE_FUNCTION,
2555 #ifdef HAVE_UNLOCKED_IOCTL
2556 .unlocked_ioctl = ll_file_ioctl,
2558 .ioctl = ll_file_ioctl,
2560 .open = ll_file_open,
2561 .release = ll_file_release,
2562 .mmap = ll_file_mmap,
2563 .llseek = ll_file_seek,
2564 #ifdef HAVE_KERNEL_SENDFILE
2565 .sendfile = ll_file_sendfile,
2567 #ifdef HAVE_KERNEL_SPLICE_READ
2568 .splice_read = ll_file_splice_read,
2572 #ifdef HAVE_F_OP_FLOCK
2573 .flock = ll_file_noflock,
2575 .lock = ll_file_noflock
2578 struct inode_operations ll_file_inode_operations = {
2579 #ifdef HAVE_VFS_INTENT_PATCHES
2580 .setattr_raw = ll_setattr_raw,
2582 .setattr = ll_setattr,
2583 .truncate = ll_truncate,
2584 .getattr = ll_getattr,
2585 .permission = ll_inode_permission,
2586 .setxattr = ll_setxattr,
2587 .getxattr = ll_getxattr,
2588 .listxattr = ll_listxattr,
2589 .removexattr = ll_removexattr,
2590 #ifdef HAVE_LINUX_FIEMAP_H
2591 .fiemap = ll_fiemap,
2595 /* dynamic ioctl number support routins */
2596 static struct llioc_ctl_data {
2597 cfs_rw_semaphore_t ioc_sem;
2598 cfs_list_t ioc_head;
2600 __RWSEM_INITIALIZER(llioc.ioc_sem),
2601 CFS_LIST_HEAD_INIT(llioc.ioc_head)
2606 cfs_list_t iocd_list;
2607 unsigned int iocd_size;
2608 llioc_callback_t iocd_cb;
2609 unsigned int iocd_count;
2610 unsigned int iocd_cmd[0];
2613 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
2616 struct llioc_data *in_data = NULL;
2619 if (cb == NULL || cmd == NULL ||
2620 count > LLIOC_MAX_CMD || count < 0)
2623 size = sizeof(*in_data) + count * sizeof(unsigned int);
2624 OBD_ALLOC(in_data, size);
2625 if (in_data == NULL)
2628 memset(in_data, 0, sizeof(*in_data));
2629 in_data->iocd_size = size;
2630 in_data->iocd_cb = cb;
2631 in_data->iocd_count = count;
2632 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
2634 cfs_down_write(&llioc.ioc_sem);
2635 cfs_list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
2636 cfs_up_write(&llioc.ioc_sem);
2641 void ll_iocontrol_unregister(void *magic)
2643 struct llioc_data *tmp;
2648 cfs_down_write(&llioc.ioc_sem);
2649 cfs_list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
2651 unsigned int size = tmp->iocd_size;
2653 cfs_list_del(&tmp->iocd_list);
2654 cfs_up_write(&llioc.ioc_sem);
2656 OBD_FREE(tmp, size);
2660 cfs_up_write(&llioc.ioc_sem);
2662 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
2665 EXPORT_SYMBOL(ll_iocontrol_register);
2666 EXPORT_SYMBOL(ll_iocontrol_unregister);
2668 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
2669 unsigned int cmd, unsigned long arg, int *rcp)
2671 enum llioc_iter ret = LLIOC_CONT;
2672 struct llioc_data *data;
2673 int rc = -EINVAL, i;
2675 cfs_down_read(&llioc.ioc_sem);
2676 cfs_list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
2677 for (i = 0; i < data->iocd_count; i++) {
2678 if (cmd != data->iocd_cmd[i])
2681 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
2685 if (ret == LLIOC_STOP)
2688 cfs_up_read(&llioc.ioc_sem);