1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
7 * Author: Andreas Dilger <adilger@clusterfs.com>
9 * This file is part of Lustre, http://www.lustre.org.
11 * Lustre is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Lustre is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Lustre; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
25 #define DEBUG_SUBSYSTEM S_LLITE
26 #include <lustre_dlm.h>
27 #include <lustre_lite.h>
28 #include <lustre_mdc.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
31 #include "llite_internal.h"
33 /* also used by llite/special.c:ll_special_open() */
34 struct ll_file_data *ll_file_data_get(void)
36 struct ll_file_data *fd;
38 OBD_SLAB_ALLOC_PTR(fd, ll_file_data_slab);
42 static void ll_file_data_put(struct ll_file_data *fd)
45 OBD_SLAB_FREE_PTR(fd, ll_file_data_slab);
48 void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
49 struct lustre_handle *fh)
51 op_data->op_fid1 = ll_i2info(inode)->lli_fid;
52 op_data->op_attr.ia_mode = inode->i_mode;
53 op_data->op_attr.ia_atime = inode->i_atime;
54 op_data->op_attr.ia_mtime = inode->i_mtime;
55 op_data->op_attr.ia_ctime = inode->i_ctime;
56 op_data->op_attr.ia_size = i_size_read(inode);
57 op_data->op_attr_blocks = inode->i_blocks;
58 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = inode->i_flags;
59 op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
60 memcpy(&op_data->op_handle, fh, sizeof(op_data->op_handle));
61 op_data->op_capa1 = ll_mdscapa_get(inode);
64 static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
65 struct obd_client_handle *och)
69 op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
70 ATTR_MTIME_SET | ATTR_CTIME_SET;
72 if (!(och->och_flags & FMODE_WRITE))
75 if (!(ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) ||
76 !S_ISREG(inode->i_mode))
77 op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
79 ll_epoch_close(inode, op_data, &och, 0);
82 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
86 static int ll_close_inode_openhandle(struct obd_export *md_exp,
88 struct obd_client_handle *och)
90 struct obd_export *exp = ll_i2mdexp(inode);
91 struct md_op_data *op_data;
92 struct ptlrpc_request *req = NULL;
93 struct obd_device *obd = class_exp2obd(exp);
100 * XXX: in case of LMV, is this correct to access
103 CERROR("Invalid MDC connection handle "LPX64"\n",
104 ll_i2mdexp(inode)->exp_handle.h_cookie);
109 * here we check if this is forced umount. If so this is called on
110 * canceling "open lock" and we do not call md_close() in this case, as
111 * it will not be successful, as import is already deactivated.
116 OBD_ALLOC_PTR(op_data);
118 GOTO(out, rc = -ENOMEM); // XXX We leak openhandle and request here.
120 ll_prepare_close(inode, op_data, och);
121 epoch_close = (op_data->op_flags & MF_EPOCH_CLOSE);
122 rc = md_close(md_exp, op_data, och->och_mod, &req);
127 /* This close must have the epoch closed. */
128 LASSERT(exp->exp_connect_flags & OBD_CONNECT_SOM);
129 LASSERT(epoch_close);
130 /* MDS has instructed us to obtain Size-on-MDS attribute from
131 * OSTs and send setattr to back to MDS. */
132 rc = ll_sizeonmds_update(inode, och->och_mod,
133 &och->och_fh, op_data->op_ioepoch);
135 CERROR("inode %lu mdc Size-on-MDS update failed: "
136 "rc = %d\n", inode->i_ino, rc);
140 CERROR("inode %lu mdc close failed: rc = %d\n",
143 ll_finish_md_op_data(op_data);
146 rc = ll_objects_destroy(req, inode);
148 CERROR("inode %lu ll_objects destroy: rc = %d\n",
155 if ((exp->exp_connect_flags & OBD_CONNECT_SOM) && !epoch_close &&
156 S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
157 ll_queue_done_writing(inode, LLIF_DONE_WRITING);
160 ptlrpc_close_replay_seq(req);
161 md_clear_open_replay_data(md_exp, och);
162 /* Free @och if it is not waiting for DONE_WRITING. */
163 och->och_fh.cookie = DEAD_HANDLE_MAGIC;
166 if (req) /* This is close request */
167 ptlrpc_req_finished(req);
171 int ll_md_real_close(struct inode *inode, int flags)
173 struct ll_inode_info *lli = ll_i2info(inode);
174 struct obd_client_handle **och_p;
175 struct obd_client_handle *och;
180 if (flags & FMODE_WRITE) {
181 och_p = &lli->lli_mds_write_och;
182 och_usecount = &lli->lli_open_fd_write_count;
183 } else if (flags & FMODE_EXEC) {
184 och_p = &lli->lli_mds_exec_och;
185 och_usecount = &lli->lli_open_fd_exec_count;
187 LASSERT(flags & FMODE_READ);
188 och_p = &lli->lli_mds_read_och;
189 och_usecount = &lli->lli_open_fd_read_count;
192 down(&lli->lli_och_sem);
193 if (*och_usecount) { /* There are still users of this handle, so
195 up(&lli->lli_och_sem);
200 up(&lli->lli_och_sem);
202 if (och) { /* There might be a race and somebody have freed this och
204 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
211 int ll_md_close(struct obd_export *md_exp, struct inode *inode,
214 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
215 struct ll_inode_info *lli = ll_i2info(inode);
219 /* clear group lock, if present */
220 if (unlikely(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
221 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
222 fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
223 rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP,
227 /* Let's see if we have good enough OPEN lock on the file and if
228 we can skip talking to MDS */
229 if (file->f_dentry->d_inode) { /* Can this ever be false? */
231 int flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK;
232 struct lustre_handle lockh;
233 struct inode *inode = file->f_dentry->d_inode;
234 ldlm_policy_data_t policy = {.l_inodebits={MDS_INODELOCK_OPEN}};
236 down(&lli->lli_och_sem);
237 if (fd->fd_omode & FMODE_WRITE) {
239 LASSERT(lli->lli_open_fd_write_count);
240 lli->lli_open_fd_write_count--;
241 } else if (fd->fd_omode & FMODE_EXEC) {
243 LASSERT(lli->lli_open_fd_exec_count);
244 lli->lli_open_fd_exec_count--;
247 LASSERT(lli->lli_open_fd_read_count);
248 lli->lli_open_fd_read_count--;
250 up(&lli->lli_och_sem);
252 if (!md_lock_match(md_exp, flags, ll_inode2fid(inode),
253 LDLM_IBITS, &policy, lockmode,
255 rc = ll_md_real_close(file->f_dentry->d_inode,
259 CERROR("Releasing a file %p with negative dentry %p. Name %s",
260 file, file->f_dentry, file->f_dentry->d_name.name);
263 LUSTRE_FPRIVATE(file) = NULL;
264 ll_file_data_put(fd);
265 ll_capa_close(inode);
270 int lov_test_and_clear_async_rc(struct lov_stripe_md *lsm);
272 /* While this returns an error code, fput() the caller does not, so we need
273 * to make every effort to clean up all of our state here. Also, applications
274 * rarely check close errors and even if an error is returned they will not
275 * re-try the close call.
277 int ll_file_release(struct inode *inode, struct file *file)
279 struct ll_file_data *fd;
280 struct ll_sb_info *sbi = ll_i2sbi(inode);
281 struct ll_inode_info *lli = ll_i2info(inode);
282 struct lov_stripe_md *lsm = lli->lli_smd;
286 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
287 inode->i_generation, inode);
289 /* don't do anything for / */
290 if (inode->i_sb->s_root == file->f_dentry)
293 ll_stats_ops_tally(sbi, LPROC_LL_RELEASE, 1);
294 fd = LUSTRE_FPRIVATE(file);
297 /* don't do anything for / */
298 if (inode->i_sb->s_root == file->f_dentry) {
299 LUSTRE_FPRIVATE(file) = NULL;
300 ll_file_data_put(fd);
305 lov_test_and_clear_async_rc(lsm);
306 lli->lli_async_rc = 0;
308 rc = ll_md_close(sbi->ll_md_exp, inode, file);
312 static int ll_intent_file_open(struct file *file, void *lmm,
313 int lmmsize, struct lookup_intent *itp)
315 struct ll_sb_info *sbi = ll_i2sbi(file->f_dentry->d_inode);
316 struct dentry *parent = file->f_dentry->d_parent;
317 const char *name = file->f_dentry->d_name.name;
318 const int len = file->f_dentry->d_name.len;
319 struct md_op_data *op_data;
320 struct ptlrpc_request *req;
326 /* Usually we come here only for NFSD, and we want open lock.
327 But we can also get here with pre 2.6.15 patchless kernels, and in
328 that case that lock is also ok */
329 /* We can also get here if there was cached open handle in revalidate_it
330 * but it disappeared while we were getting from there to ll_file_open.
331 * But this means this file was closed and immediatelly opened which
332 * makes a good candidate for using OPEN lock */
333 /* If lmmsize & lmm are not 0, we are just setting stripe info
334 * parameters. No need for the open lock */
335 if (!lmm && !lmmsize)
336 itp->it_flags |= MDS_OPEN_LOCK;
338 op_data = ll_prep_md_op_data(NULL, parent->d_inode,
339 file->f_dentry->d_inode, name, len,
340 O_RDWR, LUSTRE_OPC_ANY, NULL);
342 RETURN(PTR_ERR(op_data));
344 rc = md_intent_lock(sbi->ll_md_exp, op_data, lmm, lmmsize, itp,
345 0 /*unused */, &req, ll_md_blocking_ast, 0);
346 ll_finish_md_op_data(op_data);
348 /* reason for keep own exit path - don`t flood log
349 * with messages with -ESTALE errors.
351 if (!it_disposition(itp, DISP_OPEN_OPEN) ||
352 it_open_error(DISP_OPEN_OPEN, itp))
354 ll_release_openhandle(file->f_dentry, itp);
358 if (rc != 0 || it_open_error(DISP_OPEN_OPEN, itp)) {
359 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, itp);
360 CDEBUG(D_VFSTRACE, "lock enqueue: err: %d\n", rc);
364 if (itp->d.lustre.it_lock_mode)
365 md_set_lock_data(sbi->ll_md_exp,
366 &itp->d.lustre.it_lock_handle,
367 file->f_dentry->d_inode);
369 rc = ll_prep_inode(&file->f_dentry->d_inode, req, DLM_REPLY_REC_OFF,
372 ptlrpc_req_finished(itp->d.lustre.it_data);
375 it_clear_disposition(itp, DISP_ENQ_COMPLETE);
376 ll_intent_drop_lock(itp);
381 static int ll_och_fill(struct obd_export *md_exp, struct ll_inode_info *lli,
382 struct lookup_intent *it, struct obd_client_handle *och)
384 struct ptlrpc_request *req = it->d.lustre.it_data;
385 struct mdt_body *body;
389 body = lustre_msg_buf(req->rq_repmsg, DLM_REPLY_REC_OFF, sizeof(*body));
390 /* reply already checked out */
391 LASSERT(body != NULL);
392 /* and swabbed in md_enqueue */
393 LASSERT(lustre_rep_swabbed(req, DLM_REPLY_REC_OFF));
395 memcpy(&och->och_fh, &body->handle, sizeof(body->handle));
396 och->och_magic = OBD_CLIENT_HANDLE_MAGIC;
397 och->och_fid = lli->lli_fid;
398 och->och_flags = it->it_flags;
399 lli->lli_ioepoch = body->ioepoch;
401 return md_set_open_replay_data(md_exp, och, req);
404 int ll_local_open(struct file *file, struct lookup_intent *it,
405 struct ll_file_data *fd, struct obd_client_handle *och)
407 struct inode *inode = file->f_dentry->d_inode;
408 struct ll_inode_info *lli = ll_i2info(inode);
411 LASSERT(!LUSTRE_FPRIVATE(file));
416 struct ptlrpc_request *req = it->d.lustre.it_data;
417 struct mdt_body *body;
420 rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, lli, it, och);
424 body = lustre_msg_buf(req->rq_repmsg,
425 DLM_REPLY_REC_OFF, sizeof(*body));
427 if ((it->it_flags & FMODE_WRITE) &&
428 (body->valid & OBD_MD_FLSIZE))
430 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID"\n",
431 lli->lli_ioepoch, PFID(&lli->lli_fid));
435 LUSTRE_FPRIVATE(file) = fd;
436 ll_readahead_init(inode, &fd->fd_ras);
437 fd->fd_omode = it->it_flags;
441 /* Open a file, and (for the very first open) create objects on the OSTs at
442 * this time. If opened with O_LOV_DELAY_CREATE, then we don't do the object
443 * creation or open until ll_lov_setstripe() ioctl is called. We grab
444 * lli_open_sem to ensure no other process will create objects, send the
445 * stripe MD to the MDS, or try to destroy the objects if that fails.
447 * If we already have the stripe MD locally then we don't request it in
448 * md_open(), by passing a lmm_size = 0.
450 * It is up to the application to ensure no other processes open this file
451 * in the O_LOV_DELAY_CREATE case, or the default striping pattern will be
452 * used. We might be able to avoid races of that sort by getting lli_open_sem
453 * before returning in the O_LOV_DELAY_CREATE case and dropping it here
454 * or in ll_file_release(), but I'm not sure that is desirable/necessary.
456 int ll_file_open(struct inode *inode, struct file *file)
458 struct ll_inode_info *lli = ll_i2info(inode);
459 struct lookup_intent *it, oit = { .it_op = IT_OPEN,
460 .it_flags = file->f_flags };
461 struct lov_stripe_md *lsm;
462 struct ptlrpc_request *req = NULL;
463 struct obd_client_handle **och_p;
465 struct ll_file_data *fd;
469 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), flags %o\n", inode->i_ino,
470 inode->i_generation, inode, file->f_flags);
472 /* don't do anything for / */
473 if (inode->i_sb->s_root == file->f_dentry)
476 #ifdef HAVE_VFS_INTENT_PATCHES
479 it = file->private_data; /* XXX: compat macro */
480 file->private_data = NULL; /* prevent ll_local_open assertion */
483 fd = ll_file_data_get();
487 /* don't do anything for / */
488 if (inode->i_sb->s_root == file->f_dentry) {
489 LUSTRE_FPRIVATE(file) = fd;
493 if (!it || !it->d.lustre.it_disposition) {
494 /* Convert f_flags into access mode. We cannot use file->f_mode,
495 * because everything but O_ACCMODE mask was stripped from
497 if ((oit.it_flags + 1) & O_ACCMODE)
499 if (file->f_flags & O_TRUNC)
500 oit.it_flags |= FMODE_WRITE;
502 /* kernel only call f_op->open in dentry_open. filp_open calls
503 * dentry_open after call to open_namei that checks permissions.
504 * Only nfsd_open call dentry_open directly without checking
505 * permissions and because of that this code below is safe. */
506 if (oit.it_flags & FMODE_WRITE)
507 oit.it_flags |= MDS_OPEN_OWNEROVERRIDE;
509 /* We do not want O_EXCL here, presumably we opened the file
510 * already? XXX - NFS implications? */
511 oit.it_flags &= ~O_EXCL;
516 /* Let's see if we have file open on MDS already. */
517 if (it->it_flags & FMODE_WRITE) {
518 och_p = &lli->lli_mds_write_och;
519 och_usecount = &lli->lli_open_fd_write_count;
520 } else if (it->it_flags & FMODE_EXEC) {
521 och_p = &lli->lli_mds_exec_och;
522 och_usecount = &lli->lli_open_fd_exec_count;
524 och_p = &lli->lli_mds_read_och;
525 och_usecount = &lli->lli_open_fd_read_count;
528 down(&lli->lli_och_sem);
529 if (*och_p) { /* Open handle is present */
530 if (it_disposition(it, DISP_OPEN_OPEN)) {
531 /* Well, there's extra open request that we do not need,
532 let's close it somehow. This will decref request. */
533 rc = it_open_error(DISP_OPEN_OPEN, it);
535 ll_file_data_put(fd);
536 GOTO(out_och_free, rc);
538 ll_release_openhandle(file->f_dentry, it);
539 lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats,
544 rc = ll_local_open(file, it, fd, NULL);
546 up(&lli->lli_och_sem);
547 ll_file_data_put(fd);
551 LASSERT(*och_usecount == 0);
552 OBD_ALLOC(*och_p, sizeof (struct obd_client_handle));
554 ll_file_data_put(fd);
555 GOTO(out_och_free, rc = -ENOMEM);
558 if (!it->d.lustre.it_disposition) {
559 it->it_flags |= O_CHECK_STALE;
560 rc = ll_intent_file_open(file, NULL, 0, it);
561 it->it_flags &= ~O_CHECK_STALE;
563 ll_file_data_put(fd);
564 GOTO(out_och_free, rc);
567 /* Got some error? Release the request */
568 if (it->d.lustre.it_status < 0) {
569 req = it->d.lustre.it_data;
570 ptlrpc_req_finished(req);
572 md_set_lock_data(ll_i2sbi(inode)->ll_md_exp,
573 &it->d.lustre.it_lock_handle,
574 file->f_dentry->d_inode);
576 req = it->d.lustre.it_data;
578 /* md_intent_lock() didn't get a request ref if there was an
579 * open error, so don't do cleanup on the request here
581 /* XXX (green): Should not we bail out on any error here, not
582 * just open error? */
583 rc = it_open_error(DISP_OPEN_OPEN, it);
585 ll_file_data_put(fd);
586 GOTO(out_och_free, rc);
589 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_OPEN, 1);
590 rc = ll_local_open(file, it, fd, *och_p);
592 up(&lli->lli_och_sem);
593 ll_file_data_put(fd);
594 GOTO(out_och_free, rc);
597 up(&lli->lli_och_sem);
599 /* Must do this outside lli_och_sem lock to prevent deadlock where
600 different kind of OPEN lock for this same inode gets cancelled
601 by ldlm_cancel_lru */
602 if (!S_ISREG(inode->i_mode))
609 if (file->f_flags & O_LOV_DELAY_CREATE ||
610 !(file->f_mode & FMODE_WRITE)) {
611 CDEBUG(D_INODE, "object creation was delayed\n");
615 file->f_flags &= ~O_LOV_DELAY_CREATE;
618 ptlrpc_req_finished(req);
620 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
624 OBD_FREE(*och_p, sizeof (struct obd_client_handle));
625 *och_p = NULL; /* OBD_FREE writes some magic there */
628 up(&lli->lli_och_sem);
634 /* Fills the obdo with the attributes for the inode defined by lsm */
635 int ll_inode_getattr(struct inode *inode, struct obdo *obdo)
637 struct ptlrpc_request_set *set;
638 struct ll_inode_info *lli = ll_i2info(inode);
639 struct lov_stripe_md *lsm = lli->lli_smd;
641 struct obd_info oinfo = { { { 0 } } };
645 LASSERT(lsm != NULL);
649 oinfo.oi_oa->o_id = lsm->lsm_object_id;
650 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
651 oinfo.oi_oa->o_mode = S_IFREG;
652 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
653 OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
654 OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
655 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
657 oinfo.oi_capa = ll_mdscapa_get(inode);
659 set = ptlrpc_prep_set();
661 CERROR("can't allocate ptlrpc set\n");
664 rc = obd_getattr_async(ll_i2dtexp(inode), &oinfo, set);
666 rc = ptlrpc_set_wait(set);
667 ptlrpc_set_destroy(set);
669 capa_put(oinfo.oi_capa);
673 oinfo.oi_oa->o_valid &= (OBD_MD_FLBLOCKS | OBD_MD_FLBLKSZ |
674 OBD_MD_FLATIME | OBD_MD_FLMTIME |
675 OBD_MD_FLCTIME | OBD_MD_FLSIZE);
677 obdo_refresh_inode(inode, oinfo.oi_oa, oinfo.oi_oa->o_valid);
678 CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %lu, blksize %lu\n",
679 lli->lli_smd->lsm_object_id, i_size_read(inode),
680 inode->i_blocks, inode->i_blksize);
684 static inline void ll_remove_suid(struct inode *inode)
688 /* set S_IGID if S_IXGRP is set, and always set S_ISUID */
689 mode = (inode->i_mode & S_IXGRP)*(S_ISGID/S_IXGRP) | S_ISUID;
691 /* was any of the uid bits set? */
692 mode &= inode->i_mode;
693 if (mode && !capable(CAP_FSETID)) {
694 inode->i_mode &= ~mode;
695 // XXX careful here - we cannot change the size
699 static int ll_lock_to_stripe_offset(struct inode *inode, struct ldlm_lock *lock)
701 struct ll_inode_info *lli = ll_i2info(inode);
702 struct lov_stripe_md *lsm = lli->lli_smd;
703 struct obd_export *exp = ll_i2dtexp(inode);
706 struct ldlm_lock *lock;
707 struct lov_stripe_md *lsm;
708 } key = { .name = "lock_to_stripe", .lock = lock, .lsm = lsm };
709 __u32 stripe, vallen = sizeof(stripe);
713 if (lsm->lsm_stripe_count == 1)
714 GOTO(check, stripe = 0);
716 /* get our offset in the lov */
717 rc = obd_get_info(exp, sizeof(key), &key, &vallen, &stripe);
719 CERROR("obd_get_info: rc = %d\n", rc);
722 LASSERT(stripe < lsm->lsm_stripe_count);
725 if (lsm->lsm_oinfo[stripe]->loi_id != lock->l_resource->lr_name.name[0]||
726 lsm->lsm_oinfo[stripe]->loi_gr != lock->l_resource->lr_name.name[2]){
727 LDLM_ERROR(lock, "resource doesn't match object "LPU64"/"LPU64,
728 lsm->lsm_oinfo[stripe]->loi_id,
729 lsm->lsm_oinfo[stripe]->loi_gr);
730 RETURN(-ELDLM_NO_LOCK_DATA);
736 /* Flush the page cache for an extent as its canceled. When we're on an LOV,
737 * we get a lock cancellation for each stripe, so we have to map the obd's
738 * region back onto the stripes in the file that it held.
740 * No one can dirty the extent until we've finished our work and they can
741 * enqueue another lock. The DLM protects us from ll_file_read/write here,
742 * but other kernel actors could have pages locked.
744 * Called with the DLM lock held. */
745 void ll_pgcache_remove_extent(struct inode *inode, struct lov_stripe_md *lsm,
746 struct ldlm_lock *lock, __u32 stripe)
748 ldlm_policy_data_t tmpex;
749 unsigned long start, end, count, skip, i, j;
751 int rc, rc2, discard = lock->l_flags & LDLM_FL_DISCARD_DATA;
752 struct lustre_handle lockh;
753 struct address_space *mapping = inode->i_mapping;
756 tmpex = lock->l_policy_data;
757 CDEBUG(D_INODE|D_PAGE, "inode %lu(%p) ["LPU64"->"LPU64"] size: %llu\n",
758 inode->i_ino, inode, tmpex.l_extent.start, tmpex.l_extent.end,
761 /* our locks are page granular thanks to osc_enqueue, we invalidate the
763 if ((tmpex.l_extent.start & ~CFS_PAGE_MASK) != 0 ||
764 ((tmpex.l_extent.end + 1) & ~CFS_PAGE_MASK) != 0)
765 LDLM_ERROR(lock, "lock not aligned on PAGE_SIZE %lu",
767 LASSERT((tmpex.l_extent.start & ~CFS_PAGE_MASK) == 0);
768 LASSERT(((tmpex.l_extent.end + 1) & ~CFS_PAGE_MASK) == 0);
772 start = tmpex.l_extent.start >> CFS_PAGE_SHIFT;
773 end = tmpex.l_extent.end >> CFS_PAGE_SHIFT;
774 if (lsm->lsm_stripe_count > 1) {
775 count = lsm->lsm_stripe_size >> CFS_PAGE_SHIFT;
776 skip = (lsm->lsm_stripe_count - 1) * count;
777 start += start/count * skip + stripe * count;
779 end += end/count * skip + stripe * count;
781 if (end < tmpex.l_extent.end >> CFS_PAGE_SHIFT)
784 i = i_size_read(inode) ? (__u64)(i_size_read(inode) - 1) >>
789 CDEBUG(D_INODE|D_PAGE, "walking page indices start: %lu j: %lu "
790 "count: %lu skip: %lu end: %lu%s\n", start, start % count,
791 count, skip, end, discard ? " (DISCARDING)" : "");
793 /* walk through the vmas on the inode and tear down mmaped pages that
794 * intersect with the lock. this stops immediately if there are no
795 * mmap()ed regions of the file. This is not efficient at all and
796 * should be short lived. We'll associate mmap()ed pages with the lock
797 * and will be able to find them directly */
798 for (i = start; i <= end; i += (j + skip)) {
799 j = min(count - (i % count), end - i + 1);
802 if (ll_teardown_mmaps(mapping,
803 (__u64)i << CFS_PAGE_SHIFT,
804 ((__u64)(i+j) << CFS_PAGE_SHIFT) - 1) )
808 /* this is the simplistic implementation of page eviction at
809 * cancelation. It is careful to get races with other page
810 * lockers handled correctly. fixes from bug 20 will make it
811 * more efficient by associating locks with pages and with
812 * batching writeback under the lock explicitly. */
813 for (i = start, j = start % count; i <= end;
814 j++, i++, tmpex.l_extent.start += CFS_PAGE_SIZE) {
816 CDEBUG(D_PAGE, "skip index %lu to %lu\n", i, i + skip);
822 LASSERTF(tmpex.l_extent.start< lock->l_policy_data.l_extent.end,
823 LPU64" >= "LPU64" start %lu i %lu end %lu\n",
824 tmpex.l_extent.start, lock->l_policy_data.l_extent.end,
827 if (!mapping_has_pages(mapping)) {
828 CDEBUG(D_INODE|D_PAGE, "nothing left\n");
834 page = find_get_page(mapping, i);
837 LL_CDEBUG_PAGE(D_PAGE, page, "lock page idx %lu ext "LPU64"\n",
838 i, tmpex.l_extent.start);
841 /* page->mapping to check with racing against teardown */
842 if (!discard && clear_page_dirty_for_io(page)) {
843 rc = ll_call_writepage(inode, page);
844 /* either waiting for io to complete or reacquiring
845 * the lock that the failed writepage released */
847 wait_on_page_writeback(page);
849 CERROR("writepage inode %lu(%p) of page %p "
850 "failed: %d\n", inode->i_ino, inode,
853 set_bit(AS_ENOSPC, &mapping->flags);
855 set_bit(AS_EIO, &mapping->flags);
859 tmpex.l_extent.end = tmpex.l_extent.start + CFS_PAGE_SIZE - 1;
860 /* check to see if another DLM lock covers this page b=2765 */
861 rc2 = ldlm_lock_match(lock->l_resource->lr_namespace,
862 LDLM_FL_BLOCK_GRANTED|LDLM_FL_CBPENDING |
864 &lock->l_resource->lr_name, LDLM_EXTENT,
865 &tmpex, LCK_PR | LCK_PW, &lockh);
867 if (rc2 <= 0 && page->mapping != NULL) {
868 struct ll_async_page *llap = llap_cast_private(page);
869 /* checking again to account for writeback's
871 LL_CDEBUG_PAGE(D_PAGE, page, "truncating\n");
873 ll_ra_accounting(llap, mapping);
874 ll_truncate_complete_page(page);
877 page_cache_release(page);
879 LASSERTF(tmpex.l_extent.start <=
880 (lock->l_policy_data.l_extent.end == ~0ULL ? ~0ULL :
881 lock->l_policy_data.l_extent.end + 1),
882 "loop too long "LPU64" > "LPU64" start %lu i %lu end %lu\n",
883 tmpex.l_extent.start, lock->l_policy_data.l_extent.end,
888 static int ll_extent_lock_callback(struct ldlm_lock *lock,
889 struct ldlm_lock_desc *new, void *data,
892 struct lustre_handle lockh = { 0 };
896 if ((unsigned long)data > 0 && (unsigned long)data < 0x1000) {
897 LDLM_ERROR(lock, "cancelling lock with bad data %p", data);
902 case LDLM_CB_BLOCKING:
903 ldlm_lock2handle(lock, &lockh);
904 rc = ldlm_cli_cancel(&lockh);
906 CERROR("ldlm_cli_cancel failed: %d\n", rc);
908 case LDLM_CB_CANCELING: {
910 struct ll_inode_info *lli;
911 struct lov_stripe_md *lsm;
915 /* This lock wasn't granted, don't try to evict pages */
916 if (lock->l_req_mode != lock->l_granted_mode)
919 inode = ll_inode_from_lock(lock);
922 lli = ll_i2info(inode);
925 if (lli->lli_smd == NULL)
929 stripe = ll_lock_to_stripe_offset(inode, lock);
933 ll_pgcache_remove_extent(inode, lsm, lock, stripe);
935 lov_stripe_lock(lsm);
936 lock_res_and_lock(lock);
937 kms = ldlm_extent_shift_kms(lock,
938 lsm->lsm_oinfo[stripe]->loi_kms);
940 if (lsm->lsm_oinfo[stripe]->loi_kms != kms)
941 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
942 lsm->lsm_oinfo[stripe]->loi_kms, kms);
943 lsm->lsm_oinfo[stripe]->loi_kms = kms;
944 unlock_res_and_lock(lock);
945 lov_stripe_unlock(lsm);
958 int ll_async_completion_ast(struct ldlm_lock *lock, int flags, void *data)
960 /* XXX ALLOCATE - 160 bytes */
961 struct inode *inode = ll_inode_from_lock(lock);
962 struct ll_inode_info *lli = ll_i2info(inode);
963 struct lustre_handle lockh = { 0 };
968 if (flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
969 LDLM_FL_BLOCK_CONV)) {
970 LBUG(); /* not expecting any blocked async locks yet */
971 LDLM_DEBUG(lock, "client-side async enqueue returned a blocked "
973 ldlm_lock_dump(D_OTHER, lock, 0);
974 ldlm_reprocess_all(lock->l_resource);
978 LDLM_DEBUG(lock, "client-side async enqueue: granted/glimpsed");
980 stripe = ll_lock_to_stripe_offset(inode, lock);
984 if (lock->l_lvb_len) {
985 struct lov_stripe_md *lsm = lli->lli_smd;
987 lvb = lock->l_lvb_data;
988 lsm->lsm_oinfo[stripe].loi_rss = lvb->lvb_size;
990 lock_res_and_lock(lock);
991 ll_inode_size_lock(inode, 1);
992 kms = MAX(lsm->lsm_oinfo[stripe].loi_kms, lvb->lvb_size);
993 kms = ldlm_extent_shift_kms(NULL, kms);
994 if (lsm->lsm_oinfo[stripe].loi_kms != kms)
995 LDLM_DEBUG(lock, "updating kms from "LPU64" to "LPU64,
996 lsm->lsm_oinfo[stripe].loi_kms, kms);
997 lsm->lsm_oinfo[stripe].loi_kms = kms;
998 ll_inode_size_unlock(inode, 1);
999 unlock_res_and_lock(lock);
1004 wake_up(&lock->l_waitq);
1006 ldlm_lock2handle(lock, &lockh);
1007 ldlm_lock_decref(&lockh, LCK_PR);
1012 static int ll_glimpse_callback(struct ldlm_lock *lock, void *reqp)
1014 struct ptlrpc_request *req = reqp;
1015 struct inode *inode = ll_inode_from_lock(lock);
1016 struct ll_inode_info *lli;
1017 struct lov_stripe_md *lsm;
1018 struct ost_lvb *lvb;
1020 int size[2] = { sizeof(struct ptlrpc_body), sizeof(*lvb) };
1024 GOTO(out, rc = -ELDLM_NO_LOCK_DATA);
1025 lli = ll_i2info(inode);
1027 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
1030 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
1032 /* First, find out which stripe index this lock corresponds to. */
1033 stripe = ll_lock_to_stripe_offset(inode, lock);
1035 GOTO(iput, rc = -ELDLM_NO_LOCK_DATA);
1037 rc = lustre_pack_reply(req, 2, size, NULL);
1041 lvb = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*lvb));
1042 lvb->lvb_size = lli->lli_smd->lsm_oinfo[stripe]->loi_kms;
1043 lvb->lvb_mtime = LTIME_S(inode->i_mtime);
1044 lvb->lvb_atime = LTIME_S(inode->i_atime);
1045 lvb->lvb_ctime = LTIME_S(inode->i_ctime);
1047 LDLM_DEBUG(lock, "i_size: %llu -> stripe number %u -> kms "LPU64
1048 " atime "LPU64", mtime "LPU64", ctime "LPU64,
1049 i_size_read(inode), stripe, lvb->lvb_size, lvb->lvb_mtime,
1050 lvb->lvb_atime, lvb->lvb_ctime);
1055 /* These errors are normal races, so we don't want to fill the console
1056 * with messages by calling ptlrpc_error() */
1057 if (rc == -ELDLM_NO_LOCK_DATA)
1058 lustre_pack_reply(req, 1, NULL, NULL);
1060 req->rq_status = rc;
1064 static void ll_merge_lvb(struct inode *inode)
1066 struct ll_inode_info *lli = ll_i2info(inode);
1067 struct ll_sb_info *sbi = ll_i2sbi(inode);
1071 ll_inode_size_lock(inode, 1);
1072 inode_init_lvb(inode, &lvb);
1073 obd_merge_lvb(sbi->ll_dt_exp, lli->lli_smd, &lvb, 0);
1074 i_size_write(inode, lvb.lvb_size);
1075 inode->i_blocks = lvb.lvb_blocks;
1076 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
1077 LTIME_S(inode->i_atime) = lvb.lvb_atime;
1078 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
1079 ll_inode_size_unlock(inode, 1);
1083 int ll_local_size(struct inode *inode)
1085 ldlm_policy_data_t policy = { .l_extent = { 0, OBD_OBJECT_EOF } };
1086 struct ll_inode_info *lli = ll_i2info(inode);
1087 struct ll_sb_info *sbi = ll_i2sbi(inode);
1088 struct lustre_handle lockh = { 0 };
1093 if (lli->lli_smd->lsm_stripe_count == 0)
1096 rc = obd_match(sbi->ll_dt_exp, lli->lli_smd, LDLM_EXTENT,
1097 &policy, LCK_PR, &flags, inode, &lockh);
1103 ll_merge_lvb(inode);
1104 obd_cancel(sbi->ll_dt_exp, lli->lli_smd, LCK_PR, &lockh);
1108 int ll_glimpse_ioctl(struct ll_sb_info *sbi, struct lov_stripe_md *lsm,
1111 struct lustre_handle lockh = { 0 };
1112 struct ldlm_enqueue_info einfo = { 0 };
1113 struct obd_info oinfo = { { { 0 } } };
1119 einfo.ei_type = LDLM_EXTENT;
1120 einfo.ei_mode = LCK_PR;
1121 einfo.ei_cb_bl = ll_extent_lock_callback;
1122 einfo.ei_cb_cp = ldlm_completion_ast;
1123 einfo.ei_cb_gl = ll_glimpse_callback;
1124 einfo.ei_cbdata = NULL;
1126 oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
1127 oinfo.oi_lockh = &lockh;
1129 oinfo.oi_flags = LDLM_FL_HAS_INTENT;
1131 rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo);
1135 CERROR("obd_enqueue returned rc %d, "
1136 "returning -EIO\n", rc);
1137 RETURN(rc > 0 ? -EIO : rc);
1140 lov_stripe_lock(lsm);
1141 memset(&lvb, 0, sizeof(lvb));
1142 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 0);
1143 st->st_size = lvb.lvb_size;
1144 st->st_blocks = lvb.lvb_blocks;
1145 st->st_mtime = lvb.lvb_mtime;
1146 st->st_atime = lvb.lvb_atime;
1147 st->st_ctime = lvb.lvb_ctime;
1148 lov_stripe_unlock(lsm);
1153 /* NB: obd_merge_lvb will prefer locally cached writes if they extend the
1154 * file (because it prefers KMS over RSS when larger) */
1155 int ll_glimpse_size(struct inode *inode, int ast_flags)
1157 struct ll_inode_info *lli = ll_i2info(inode);
1158 struct ll_sb_info *sbi = ll_i2sbi(inode);
1159 struct lustre_handle lockh = { 0 };
1160 struct ldlm_enqueue_info einfo = { 0 };
1161 struct obd_info oinfo = { { { 0 } } };
1165 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
1168 CDEBUG(D_DLMTRACE, "Glimpsing inode %lu\n", inode->i_ino);
1170 if (!lli->lli_smd) {
1171 CDEBUG(D_DLMTRACE, "No objects for inode %lu\n", inode->i_ino);
1175 /* NOTE: this looks like DLM lock request, but it may not be one. Due
1176 * to LDLM_FL_HAS_INTENT flag, this is glimpse request, that
1177 * won't revoke any conflicting DLM locks held. Instead,
1178 * ll_glimpse_callback() will be called on each client
1179 * holding a DLM lock against this file, and resulting size
1180 * will be returned for each stripe. DLM lock on [0, EOF] is
1181 * acquired only if there were no conflicting locks. */
1182 einfo.ei_type = LDLM_EXTENT;
1183 einfo.ei_mode = LCK_PR;
1184 einfo.ei_cb_bl = ll_extent_lock_callback;
1185 einfo.ei_cb_cp = ldlm_completion_ast;
1186 einfo.ei_cb_gl = ll_glimpse_callback;
1187 einfo.ei_cbdata = inode;
1189 oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
1190 oinfo.oi_lockh = &lockh;
1191 oinfo.oi_md = lli->lli_smd;
1192 oinfo.oi_flags = ast_flags | LDLM_FL_HAS_INTENT;
1194 rc = obd_enqueue_rqset(sbi->ll_dt_exp, &oinfo, &einfo);
1198 CERROR("obd_enqueue returned rc %d, returning -EIO\n", rc);
1199 RETURN(rc > 0 ? -EIO : rc);
1202 ll_merge_lvb(inode);
1204 CDEBUG(D_DLMTRACE, "glimpse: size: %llu, blocks: %lu\n",
1205 i_size_read(inode), inode->i_blocks);
1210 int ll_extent_lock(struct ll_file_data *fd, struct inode *inode,
1211 struct lov_stripe_md *lsm, int mode,
1212 ldlm_policy_data_t *policy, struct lustre_handle *lockh,
1215 struct ll_sb_info *sbi = ll_i2sbi(inode);
1217 struct ldlm_enqueue_info einfo = { 0 };
1218 struct obd_info oinfo = { { { 0 } } };
1222 LASSERT(!lustre_handle_is_used(lockh));
1223 LASSERT(lsm != NULL);
1225 /* don't drop the mmapped file to LRU */
1226 if (mapping_mapped(inode->i_mapping))
1227 ast_flags |= LDLM_FL_NO_LRU;
1229 /* XXX phil: can we do this? won't it screw the file size up? */
1230 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
1231 (sbi->ll_flags & LL_SBI_NOLCK))
1234 CDEBUG(D_DLMTRACE, "Locking inode %lu, start "LPU64" end "LPU64"\n",
1235 inode->i_ino, policy->l_extent.start, policy->l_extent.end);
1237 einfo.ei_type = LDLM_EXTENT;
1238 einfo.ei_mode = mode;
1239 einfo.ei_cb_bl = ll_extent_lock_callback;
1240 einfo.ei_cb_cp = ldlm_completion_ast;
1241 einfo.ei_cb_gl = ll_glimpse_callback;
1242 einfo.ei_cbdata = inode;
1244 oinfo.oi_policy = *policy;
1245 oinfo.oi_lockh = lockh;
1247 oinfo.oi_flags = ast_flags;
1249 rc = obd_enqueue(sbi->ll_dt_exp, &oinfo, &einfo, NULL);
1250 *policy = oinfo.oi_policy;
1254 ll_inode_size_lock(inode, 1);
1255 inode_init_lvb(inode, &lvb);
1256 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
1258 if (policy->l_extent.start == 0 &&
1259 policy->l_extent.end == OBD_OBJECT_EOF) {
1260 /* vmtruncate()->ll_truncate() first sets the i_size and then
1261 * the kms under both a DLM lock and the
1262 * ll_inode_size_lock(). If we don't get the
1263 * ll_inode_size_lock() here we can match the DLM lock and
1264 * reset i_size from the kms before the truncating path has
1265 * updated the kms. generic_file_write can then trust the
1266 * stale i_size when doing appending writes and effectively
1267 * cancel the result of the truncate. Getting the
1268 * ll_inode_size_lock() after the enqueue maintains the DLM
1269 * -> ll_inode_size_lock() acquiring order. */
1270 i_size_write(inode, lvb.lvb_size);
1271 CDEBUG(D_INODE, "inode=%lu, updating i_size %llu\n",
1272 inode->i_ino, i_size_read(inode));
1276 LTIME_S(inode->i_mtime) = lvb.lvb_mtime;
1277 LTIME_S(inode->i_atime) = lvb.lvb_atime;
1278 LTIME_S(inode->i_ctime) = lvb.lvb_ctime;
1280 ll_inode_size_unlock(inode, 1);
1285 int ll_extent_unlock(struct ll_file_data *fd, struct inode *inode,
1286 struct lov_stripe_md *lsm, int mode,
1287 struct lustre_handle *lockh)
1289 struct ll_sb_info *sbi = ll_i2sbi(inode);
1293 /* XXX phil: can we do this? won't it screw the file size up? */
1294 if ((fd && (fd->fd_flags & LL_FILE_IGNORE_LOCK)) ||
1295 (sbi->ll_flags & LL_SBI_NOLCK))
1298 rc = obd_cancel(sbi->ll_dt_exp, lsm, mode, lockh);
1303 static ssize_t ll_file_read(struct file *file, char *buf, size_t count,
1306 struct inode *inode = file->f_dentry->d_inode;
1307 struct ll_inode_info *lli = ll_i2info(inode);
1308 struct lov_stripe_md *lsm = lli->lli_smd;
1309 struct ll_sb_info *sbi = ll_i2sbi(inode);
1310 struct ll_lock_tree tree;
1311 struct ll_lock_tree_node *node;
1313 struct ll_ra_read bead;
1316 ssize_t retval, chunk, sum = 0;
1320 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n",
1321 inode->i_ino, inode->i_generation, inode, count, *ppos);
1322 /* "If nbyte is 0, read() will return 0 and have no other results."
1323 * -- Single Unix Spec */
1327 ll_stats_ops_tally(sbi, LPROC_LL_READ_BYTES, count);
1330 /* Read on file with no objects should return zero-filled
1331 * buffers up to file size (we can get non-zero sizes with
1332 * mknod + truncate, then opening file for read. This is a
1333 * common pattern in NFS case, it seems). Bug 6243 */
1335 /* Since there are no objects on OSTs, we have nothing to get
1336 * lock on and so we are forced to access inode->i_size
1339 /* Read beyond end of file */
1340 if (*ppos >= i_size_read(inode))
1343 if (count > i_size_read(inode) - *ppos)
1344 count = i_size_read(inode) - *ppos;
1345 /* Make sure to correctly adjust the file pos pointer for
1347 notzeroed = clear_user(buf, count);
1356 if (sbi->ll_max_rw_chunk != 0) {
1357 /* first, let's know the end of the current stripe */
1359 obd_extent_calc(sbi->ll_dt_exp, lsm, OBD_CALC_STRIPE_END,
1362 /* correct, the end is beyond the request */
1363 if (end > *ppos + count - 1)
1364 end = *ppos + count - 1;
1366 /* and chunk shouldn't be too large even if striping is wide */
1367 if (end - *ppos > sbi->ll_max_rw_chunk)
1368 end = *ppos + sbi->ll_max_rw_chunk - 1;
1370 end = *ppos + count - 1;
1373 node = ll_node_from_inode(inode, *ppos, end, LCK_PR);
1375 GOTO(out, retval = PTR_ERR(node));
1378 tree.lt_fd = LUSTRE_FPRIVATE(file);
1379 rc = ll_tree_lock(&tree, node, buf, count,
1380 file->f_flags & O_NONBLOCK ? LDLM_FL_BLOCK_NOWAIT :0);
1382 GOTO(out, retval = rc);
1384 ll_inode_size_lock(inode, 1);
1386 * Consistency guarantees: following possibilities exist for the
1387 * relation between region being read and real file size at this
1390 * (A): the region is completely inside of the file;
1392 * (B-x): x bytes of region are inside of the file, the rest is
1395 * (C): the region is completely outside of the file.
1397 * This classification is stable under DLM lock acquired by
1398 * ll_tree_lock() above, because to change class, other client has to
1399 * take DLM lock conflicting with our lock. Also, any updates to
1400 * ->i_size by other threads on this client are serialized by
1401 * ll_inode_size_lock(). This guarantees that short reads are handled
1402 * correctly in the face of concurrent writes and truncates.
1404 inode_init_lvb(inode, &lvb);
1405 obd_merge_lvb(sbi->ll_dt_exp, lsm, &lvb, 1);
1407 if (*ppos + count - 1 > kms) {
1408 /* A glimpse is necessary to determine whether we return a
1409 * short read (B) or some zeroes at the end of the buffer (C) */
1410 ll_inode_size_unlock(inode, 1);
1411 retval = ll_glimpse_size(inode, LDLM_FL_BLOCK_GRANTED);
1413 ll_tree_unlock(&tree);
1417 /* region is within kms and, hence, within real file size (A).
1418 * We need to increase i_size to cover the read region so that
1419 * generic_file_read() will do its job, but that doesn't mean
1420 * the kms size is _correct_, it is only the _minimum_ size.
1421 * If someone does a stat they will get the correct size which
1422 * will always be >= the kms value here. b=11081 */
1423 if (i_size_read(inode) < kms)
1424 i_size_write(inode, kms);
1425 ll_inode_size_unlock(inode, 1);
1428 chunk = end - *ppos + 1;
1429 CDEBUG(D_INODE, "Read ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n",
1430 inode->i_ino, chunk, *ppos, i_size_read(inode));
1432 /* turn off the kernel's read-ahead */
1433 file->f_ra.ra_pages = 0;
1435 /* initialize read-ahead window once per syscall */
1438 bead.lrr_start = *ppos >> CFS_PAGE_SHIFT;
1439 bead.lrr_count = (count + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
1440 ll_ra_read_in(file, &bead);
1444 file_accessed(file);
1445 retval = generic_file_read(file, buf, chunk, ppos);
1446 ll_rw_stats_tally(sbi, current->pid, file, count, 0);
1448 ll_tree_unlock(&tree);
1454 if (retval == chunk && count > 0)
1460 ll_ra_read_ex(file, &bead);
1461 retval = (sum > 0) ? sum : retval;
1466 * Write to a file (through the page cache).
1468 static ssize_t ll_file_write(struct file *file, const char *buf, size_t count,
1471 struct inode *inode = file->f_dentry->d_inode;
1472 struct ll_sb_info *sbi = ll_i2sbi(inode);
1473 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1474 struct ll_lock_tree tree;
1475 struct ll_lock_tree_node *node;
1476 loff_t maxbytes = ll_file_maxbytes(inode);
1477 loff_t lock_start, lock_end, end;
1478 ssize_t retval, chunk, sum = 0;
1482 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n",
1483 inode->i_ino, inode->i_generation, inode, count, *ppos);
1485 SIGNAL_MASK_ASSERT(); /* XXX BUG 1511 */
1487 /* POSIX, but surprised the VFS doesn't check this already */
1491 /* If file was opened for LL_IOC_LOV_SETSTRIPE but the ioctl wasn't
1492 * called on the file, don't fail the below assertion (bug 2388). */
1493 if (file->f_flags & O_LOV_DELAY_CREATE &&
1494 ll_i2info(inode)->lli_smd == NULL)
1497 LASSERT(ll_i2info(inode)->lli_smd != NULL);
1499 down(&ll_i2info(inode)->lli_write_sem);
1502 chunk = 0; /* just to fix gcc's warning */
1503 end = *ppos + count - 1;
1505 if (file->f_flags & O_APPEND) {
1507 lock_end = OBD_OBJECT_EOF;
1508 } else if (sbi->ll_max_rw_chunk != 0) {
1509 /* first, let's know the end of the current stripe */
1511 obd_extent_calc(sbi->ll_dt_exp, lsm, OBD_CALC_STRIPE_END,
1514 /* correct, the end is beyond the request */
1515 if (end > *ppos + count - 1)
1516 end = *ppos + count - 1;
1518 /* and chunk shouldn't be too large even if striping is wide */
1519 if (end - *ppos > sbi->ll_max_rw_chunk)
1520 end = *ppos + sbi->ll_max_rw_chunk - 1;
1525 lock_end = *ppos + count - 1;
1527 node = ll_node_from_inode(inode, lock_start, lock_end, LCK_PW);
1530 GOTO(out, retval = PTR_ERR(node));
1532 tree.lt_fd = LUSTRE_FPRIVATE(file);
1533 rc = ll_tree_lock(&tree, node, buf, count,
1534 file->f_flags & O_NONBLOCK ? LDLM_FL_BLOCK_NOWAIT :0);
1536 GOTO(out, retval = rc);
1538 /* This is ok, g_f_w will overwrite this under i_sem if it races
1539 * with a local truncate, it just makes our maxbyte checking easier.
1540 * The i_size value gets updated in ll_extent_lock() as a consequence
1541 * of the [0,EOF] extent lock we requested above. */
1542 if (file->f_flags & O_APPEND) {
1543 *ppos = i_size_read(inode);
1544 end = *ppos + count - 1;
1547 if (*ppos >= maxbytes) {
1548 send_sig(SIGXFSZ, current, 0);
1549 GOTO(out_unlock, retval = -EFBIG);
1551 if (*ppos + count > maxbytes)
1552 count = maxbytes - *ppos;
1554 /* generic_file_write handles O_APPEND after getting i_mutex */
1555 chunk = end - *ppos + 1;
1556 CDEBUG(D_INFO, "Writing inode %lu, "LPSZ" bytes, offset %Lu\n",
1557 inode->i_ino, chunk, *ppos);
1558 retval = generic_file_write(file, buf, chunk, ppos);
1559 ll_rw_stats_tally(ll_i2sbi(inode), current->pid, file, count, 1);
1562 ll_tree_unlock(&tree);
1569 if (retval == chunk && count > 0)
1573 up(&ll_i2info(inode)->lli_write_sem);
1575 retval = (sum > 0) ? sum : retval;
1576 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_WRITE_BYTES,
1577 retval > 0 ? retval : 0);
1582 * Send file content (through pagecache) somewhere with helper
1584 static ssize_t ll_file_sendfile(struct file *in_file, loff_t *ppos,size_t count,
1585 read_actor_t actor, void *target)
1587 struct inode *inode = in_file->f_dentry->d_inode;
1588 struct ll_inode_info *lli = ll_i2info(inode);
1589 struct lov_stripe_md *lsm = lli->lli_smd;
1590 struct ll_lock_tree tree;
1591 struct ll_lock_tree_node *node;
1593 struct ll_ra_read bead;
1598 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),size="LPSZ",offset=%Ld\n",
1599 inode->i_ino, inode->i_generation, inode, count, *ppos);
1601 /* "If nbyte is 0, read() will return 0 and have no other results."
1602 * -- Single Unix Spec */
1606 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_READ_BYTES, count);
1607 /* turn off the kernel's read-ahead */
1608 in_file->f_ra.ra_pages = 0;
1610 /* File with no objects, nothing to lock */
1612 RETURN(generic_file_sendfile(in_file, ppos, count, actor, target));
1614 node = ll_node_from_inode(inode, *ppos, *ppos + count - 1, LCK_PR);
1616 RETURN(PTR_ERR(node));
1618 tree.lt_fd = LUSTRE_FPRIVATE(in_file);
1619 rc = ll_tree_lock(&tree, node, NULL, count,
1620 in_file->f_flags & O_NONBLOCK?LDLM_FL_BLOCK_NOWAIT:0);
1624 ll_inode_size_lock(inode, 1);
1626 * Consistency guarantees: following possibilities exist for the
1627 * relation between region being read and real file size at this
1630 * (A): the region is completely inside of the file;
1632 * (B-x): x bytes of region are inside of the file, the rest is
1635 * (C): the region is completely outside of the file.
1637 * This classification is stable under DLM lock acquired by
1638 * ll_tree_lock() above, because to change class, other client has to
1639 * take DLM lock conflicting with our lock. Also, any updates to
1640 * ->i_size by other threads on this client are serialized by
1641 * ll_inode_size_lock(). This guarantees that short reads are handled
1642 * correctly in the face of concurrent writes and truncates.
1644 inode_init_lvb(inode, &lvb);
1645 obd_merge_lvb(ll_i2sbi(inode)->ll_dt_exp, lsm, &lvb, 1);
1647 if (*ppos + count - 1 > kms) {
1648 /* A glimpse is necessary to determine whether we return a
1649 * short read (B) or some zeroes at the end of the buffer (C) */
1650 ll_inode_size_unlock(inode, 1);
1651 retval = ll_glimpse_size(inode, LDLM_FL_BLOCK_GRANTED);
1655 /* region is within kms and, hence, within real file size (A) */
1656 i_size_write(inode, kms);
1657 ll_inode_size_unlock(inode, 1);
1660 CDEBUG(D_INFO, "Send ino %lu, "LPSZ" bytes, offset %lld, i_size %llu\n",
1661 inode->i_ino, count, *ppos, i_size_read(inode));
1663 bead.lrr_start = *ppos >> CFS_PAGE_SHIFT;
1664 bead.lrr_count = (count + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
1665 ll_ra_read_in(in_file, &bead);
1667 file_accessed(in_file);
1668 retval = generic_file_sendfile(in_file, ppos, count, actor, target);
1669 ll_ra_read_ex(in_file, &bead);
1672 ll_tree_unlock(&tree);
1676 static int ll_lov_recreate_obj(struct inode *inode, struct file *file,
1679 struct ll_inode_info *lli = ll_i2info(inode);
1680 struct obd_export *exp = ll_i2dtexp(inode);
1681 struct ll_recreate_obj ucreatp;
1682 struct obd_trans_info oti = { 0 };
1683 struct obdo *oa = NULL;
1686 struct lov_stripe_md *lsm, *lsm2;
1689 if (!capable (CAP_SYS_ADMIN))
1692 rc = copy_from_user(&ucreatp, (struct ll_recreate_obj *)arg,
1693 sizeof(struct ll_recreate_obj));
1701 down(&lli->lli_size_sem);
1704 GOTO(out, rc = -ENOENT);
1705 lsm_size = sizeof(*lsm) + (sizeof(struct lov_oinfo) *
1706 (lsm->lsm_stripe_count));
1708 OBD_ALLOC(lsm2, lsm_size);
1710 GOTO(out, rc = -ENOMEM);
1712 oa->o_id = ucreatp.lrc_id;
1713 oa->o_gr = ucreatp.lrc_group;
1714 oa->o_nlink = ucreatp.lrc_ost_idx;
1715 oa->o_flags |= OBD_FL_RECREATE_OBJS;
1716 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1717 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1718 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1720 memcpy(lsm2, lsm, lsm_size);
1721 rc = obd_create(exp, oa, &lsm2, &oti);
1723 OBD_FREE(lsm2, lsm_size);
1726 up(&lli->lli_size_sem);
1731 int ll_lov_setstripe_ea_info(struct inode *inode, struct file *file,
1732 int flags, struct lov_user_md *lum, int lum_size)
1734 struct ll_inode_info *lli = ll_i2info(inode);
1735 struct lov_stripe_md *lsm;
1736 struct lookup_intent oit = {.it_op = IT_OPEN, .it_flags = flags};
1740 down(&lli->lli_size_sem);
1743 up(&lli->lli_size_sem);
1744 CDEBUG(D_IOCTL, "stripe already exists for ino %lu\n",
1749 rc = ll_intent_file_open(file, lum, lum_size, &oit);
1752 if (it_disposition(&oit, DISP_LOOKUP_NEG))
1753 GOTO(out_req_free, rc = -ENOENT);
1754 rc = oit.d.lustre.it_status;
1756 GOTO(out_req_free, rc);
1758 ll_release_openhandle(file->f_dentry, &oit);
1761 up(&lli->lli_size_sem);
1762 ll_intent_release(&oit);
1765 ptlrpc_req_finished((struct ptlrpc_request *) oit.d.lustre.it_data);
1769 int ll_lov_getstripe_ea_info(struct inode *inode, const char *filename,
1770 struct lov_mds_md **lmmp, int *lmm_size,
1771 struct ptlrpc_request **request)
1773 struct ll_sb_info *sbi = ll_i2sbi(inode);
1774 struct mdt_body *body;
1775 struct lov_mds_md *lmm = NULL;
1776 struct ptlrpc_request *req = NULL;
1777 struct obd_capa *oc;
1780 rc = ll_get_max_mdsize(sbi, &lmmsize);
1784 oc = ll_mdscapa_get(inode);
1785 rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode),
1786 oc, filename, strlen(filename) + 1,
1787 OBD_MD_FLEASIZE | OBD_MD_FLDIREA, lmmsize, &req);
1790 CDEBUG(D_INFO, "md_getattr_name failed "
1791 "on %s: rc %d\n", filename, rc);
1795 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1796 LASSERT(body != NULL); /* checked by mdc_getattr_name */
1797 /* swabbed by mdc_getattr_name */
1798 LASSERT(lustre_rep_swabbed(req, REPLY_REC_OFF));
1800 lmmsize = body->eadatasize;
1802 if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
1804 GOTO(out, rc = -ENODATA);
1807 lmm = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1, lmmsize);
1808 LASSERT(lmm != NULL);
1809 LASSERT(lustre_rep_swabbed(req, REPLY_REC_OFF + 1));
1812 * This is coming from the MDS, so is probably in
1813 * little endian. We convert it to host endian before
1814 * passing it to userspace.
1816 if (lmm->lmm_magic == __swab32(LOV_MAGIC)) {
1817 lustre_swab_lov_user_md((struct lov_user_md *)lmm);
1818 lustre_swab_lov_user_md_objects((struct lov_user_md *)lmm);
1819 } else if (lmm->lmm_magic == __swab32(LOV_MAGIC_JOIN)) {
1820 lustre_swab_lov_user_md_join((struct lov_user_md_join *)lmm);
1823 if (lmm->lmm_magic == LOV_MAGIC_JOIN) {
1824 struct lov_stripe_md *lsm;
1825 struct lov_user_md_join *lmj;
1826 int lmj_size, i, aindex = 0;
1828 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1830 GOTO(out, rc = -ENOMEM);
1831 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
1833 GOTO(out_free_memmd, rc);
1835 lmj_size = sizeof(struct lov_user_md_join) +
1836 lsm->lsm_stripe_count *
1837 sizeof(struct lov_user_ost_data_join);
1838 OBD_ALLOC(lmj, lmj_size);
1840 GOTO(out_free_memmd, rc = -ENOMEM);
1842 memcpy(lmj, lmm, sizeof(struct lov_user_md_join));
1843 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1844 struct lov_extent *lex =
1845 &lsm->lsm_array->lai_ext_array[aindex];
1847 if (lex->le_loi_idx + lex->le_stripe_count <= i)
1849 CDEBUG(D_INFO, "aindex %d i %d l_extent_start "
1850 LPU64" len %d\n", aindex, i,
1851 lex->le_start, (int)lex->le_len);
1852 lmj->lmm_objects[i].l_extent_start =
1855 if ((int)lex->le_len == -1)
1856 lmj->lmm_objects[i].l_extent_end = -1;
1858 lmj->lmm_objects[i].l_extent_end =
1859 lex->le_start + lex->le_len;
1860 lmj->lmm_objects[i].l_object_id =
1861 lsm->lsm_oinfo[i]->loi_id;
1862 lmj->lmm_objects[i].l_object_gr =
1863 lsm->lsm_oinfo[i]->loi_gr;
1864 lmj->lmm_objects[i].l_ost_gen =
1865 lsm->lsm_oinfo[i]->loi_ost_gen;
1866 lmj->lmm_objects[i].l_ost_idx =
1867 lsm->lsm_oinfo[i]->loi_ost_idx;
1869 lmm = (struct lov_mds_md *)lmj;
1872 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1876 *lmm_size = lmmsize;
1881 static int ll_lov_setea(struct inode *inode, struct file *file,
1884 int flags = MDS_OPEN_HAS_OBJS | FMODE_WRITE;
1885 struct lov_user_md *lump;
1886 int lum_size = sizeof(struct lov_user_md) +
1887 sizeof(struct lov_user_ost_data);
1891 if (!capable (CAP_SYS_ADMIN))
1894 OBD_ALLOC(lump, lum_size);
1898 rc = copy_from_user(lump, (struct lov_user_md *)arg, lum_size);
1900 OBD_FREE(lump, lum_size);
1904 rc = ll_lov_setstripe_ea_info(inode, file, flags, lump, lum_size);
1906 OBD_FREE(lump, lum_size);
1910 static int ll_lov_setstripe(struct inode *inode, struct file *file,
1913 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
1915 int flags = FMODE_WRITE;
1918 /* Bug 1152: copy properly when this is no longer true */
1919 LASSERT(sizeof(lum) == sizeof(*lump));
1920 LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lump->lmm_objects[0]));
1921 rc = copy_from_user(&lum, lump, sizeof(lum));
1925 rc = ll_lov_setstripe_ea_info(inode, file, flags, &lum, sizeof(lum));
1927 put_user(0, &lump->lmm_stripe_count);
1928 rc = obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode),
1929 0, ll_i2info(inode)->lli_smd, lump);
1934 static int ll_lov_getstripe(struct inode *inode, unsigned long arg)
1936 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1941 return obd_iocontrol(LL_IOC_LOV_GETSTRIPE, ll_i2dtexp(inode), 0, lsm,
1945 static int ll_get_grouplock(struct inode *inode, struct file *file,
1948 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1949 ldlm_policy_data_t policy = { .l_extent = { .start = 0,
1950 .end = OBD_OBJECT_EOF}};
1951 struct lustre_handle lockh = { 0 };
1952 struct ll_inode_info *lli = ll_i2info(inode);
1953 struct lov_stripe_md *lsm = lli->lli_smd;
1957 if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
1961 policy.l_extent.gid = arg;
1962 if (file->f_flags & O_NONBLOCK)
1963 flags = LDLM_FL_BLOCK_NOWAIT;
1965 rc = ll_extent_lock(fd, inode, lsm, LCK_GROUP, &policy, &lockh, flags);
1969 fd->fd_flags |= LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK;
1971 memcpy(&fd->fd_cwlockh, &lockh, sizeof(lockh));
1976 static int ll_put_grouplock(struct inode *inode, struct file *file,
1979 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1980 struct ll_inode_info *lli = ll_i2info(inode);
1981 struct lov_stripe_md *lsm = lli->lli_smd;
1985 if (!(fd->fd_flags & LL_FILE_GROUP_LOCKED)) {
1986 /* Ugh, it's already unlocked. */
1990 if (fd->fd_gid != arg) /* Ugh? Unlocking with different gid? */
1993 fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
1995 rc = ll_extent_unlock(fd, inode, lsm, LCK_GROUP, &fd->fd_cwlockh);
2000 memset(&fd->fd_cwlockh, 0, sizeof(fd->fd_cwlockh));
2005 static int join_sanity_check(struct inode *head, struct inode *tail)
2008 if ((ll_i2sbi(head)->ll_flags & LL_SBI_JOIN) == 0) {
2009 CERROR("server do not support join \n");
2012 if (!S_ISREG(tail->i_mode) || !S_ISREG(head->i_mode)) {
2013 CERROR("tail ino %lu and ino head %lu must be regular\n",
2014 head->i_ino, tail->i_ino);
2017 if (head->i_ino == tail->i_ino) {
2018 CERROR("file %lu can not be joined to itself \n", head->i_ino);
2021 if (i_size_read(head) % JOIN_FILE_ALIGN) {
2022 CERROR("hsize %llu must be times of 64K\n", i_size_read(head));
2028 static int join_file(struct inode *head_inode, struct file *head_filp,
2029 struct file *tail_filp)
2031 struct dentry *tail_dentry = tail_filp->f_dentry;
2032 struct lookup_intent oit = {.it_op = IT_OPEN,
2033 .it_flags = head_filp->f_flags|O_JOIN_FILE};
2034 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_CW,
2035 ll_md_blocking_ast, ldlm_completion_ast, NULL, NULL };
2037 struct lustre_handle lockh;
2038 struct md_op_data *op_data;
2043 tail_dentry = tail_filp->f_dentry;
2045 data = i_size_read(head_inode);
2046 op_data = ll_prep_md_op_data(NULL, head_inode,
2047 tail_dentry->d_parent->d_inode,
2048 tail_dentry->d_name.name,
2049 tail_dentry->d_name.len, 0,
2050 LUSTRE_OPC_ANY, &data);
2051 if (IS_ERR(op_data))
2052 RETURN(PTR_ERR(op_data));
2054 rc = md_enqueue(ll_i2mdexp(head_inode), &einfo, &oit,
2055 op_data, &lockh, NULL, 0, 0);
2057 ll_finish_md_op_data(op_data);
2061 rc = oit.d.lustre.it_status;
2063 if (rc < 0 || it_open_error(DISP_OPEN_OPEN, &oit)) {
2064 rc = rc ? rc : it_open_error(DISP_OPEN_OPEN, &oit);
2065 ptlrpc_req_finished((struct ptlrpc_request *)
2066 oit.d.lustre.it_data);
2070 if (oit.d.lustre.it_lock_mode) { /* If we got lock - release it right
2072 ldlm_lock_decref(&lockh, oit.d.lustre.it_lock_mode);
2073 oit.d.lustre.it_lock_mode = 0;
2075 ll_release_openhandle(head_filp->f_dentry, &oit);
2077 ll_intent_release(&oit);
2081 static int ll_file_join(struct inode *head, struct file *filp,
2082 char *filename_tail)
2084 struct inode *tail = NULL, *first = NULL, *second = NULL;
2085 struct dentry *tail_dentry;
2086 struct file *tail_filp, *first_filp, *second_filp;
2087 struct ll_lock_tree first_tree, second_tree;
2088 struct ll_lock_tree_node *first_node, *second_node;
2089 struct ll_inode_info *hlli = ll_i2info(head), *tlli;
2090 int rc = 0, cleanup_phase = 0;
2093 CDEBUG(D_VFSTRACE, "VFS Op:head=%lu/%u(%p) tail %s\n",
2094 head->i_ino, head->i_generation, head, filename_tail);
2096 tail_filp = filp_open(filename_tail, O_WRONLY, 0644);
2097 if (IS_ERR(tail_filp)) {
2098 CERROR("Can not open tail file %s", filename_tail);
2099 rc = PTR_ERR(tail_filp);
2102 tail = igrab(tail_filp->f_dentry->d_inode);
2104 tlli = ll_i2info(tail);
2105 tail_dentry = tail_filp->f_dentry;
2106 LASSERT(tail_dentry);
2109 /*reorder the inode for lock sequence*/
2110 first = head->i_ino > tail->i_ino ? head : tail;
2111 second = head->i_ino > tail->i_ino ? tail : head;
2112 first_filp = head->i_ino > tail->i_ino ? filp : tail_filp;
2113 second_filp = head->i_ino > tail->i_ino ? tail_filp : filp;
2115 CDEBUG(D_INFO, "reorder object from %lu:%lu to %lu:%lu \n",
2116 head->i_ino, tail->i_ino, first->i_ino, second->i_ino);
2117 first_node = ll_node_from_inode(first, 0, OBD_OBJECT_EOF, LCK_EX);
2118 if (IS_ERR(first_node)){
2119 rc = PTR_ERR(first_node);
2122 first_tree.lt_fd = first_filp->private_data;
2123 rc = ll_tree_lock(&first_tree, first_node, NULL, 0, 0);
2128 second_node = ll_node_from_inode(second, 0, OBD_OBJECT_EOF, LCK_EX);
2129 if (IS_ERR(second_node)){
2130 rc = PTR_ERR(second_node);
2133 second_tree.lt_fd = second_filp->private_data;
2134 rc = ll_tree_lock(&second_tree, second_node, NULL, 0, 0);
2139 rc = join_sanity_check(head, tail);
2143 rc = join_file(head, filp, tail_filp);
2147 switch (cleanup_phase) {
2149 ll_tree_unlock(&second_tree);
2150 obd_cancel_unused(ll_i2dtexp(second),
2151 ll_i2info(second)->lli_smd, 0, NULL);
2153 ll_tree_unlock(&first_tree);
2154 obd_cancel_unused(ll_i2dtexp(first),
2155 ll_i2info(first)->lli_smd, 0, NULL);
2157 filp_close(tail_filp, 0);
2160 if (head && rc == 0) {
2161 obd_free_memmd(ll_i2sbi(head)->ll_dt_exp,
2163 hlli->lli_smd = NULL;
2168 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
2174 int ll_release_openhandle(struct dentry *dentry, struct lookup_intent *it)
2176 struct inode *inode = dentry->d_inode;
2177 struct obd_client_handle *och;
2183 /* Root ? Do nothing. */
2184 if (dentry->d_inode->i_sb->s_root == dentry)
2187 /* No open handle to close? Move away */
2188 if (!it_disposition(it, DISP_OPEN_OPEN))
2191 LASSERT(it_open_error(DISP_OPEN_OPEN, it) == 0);
2193 OBD_ALLOC(och, sizeof(*och));
2195 GOTO(out, rc = -ENOMEM);
2197 ll_och_fill(ll_i2sbi(inode)->ll_md_exp,
2198 ll_i2info(inode), it, och);
2200 rc = ll_close_inode_openhandle(ll_i2sbi(inode)->ll_md_exp,
2203 /* this one is in place of ll_file_open */
2204 ptlrpc_req_finished(it->d.lustre.it_data);
2205 it_clear_disposition(it, DISP_ENQ_OPEN_REF);
2209 int ll_file_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
2212 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
2216 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),cmd=%x\n", inode->i_ino,
2217 inode->i_generation, inode, cmd);
2218 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
2220 /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
2221 if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
2225 case LL_IOC_GETFLAGS:
2226 /* Get the current value of the file flags */
2227 return put_user(fd->fd_flags, (int *)arg);
2228 case LL_IOC_SETFLAGS:
2229 case LL_IOC_CLRFLAGS:
2230 /* Set or clear specific file flags */
2231 /* XXX This probably needs checks to ensure the flags are
2232 * not abused, and to handle any flag side effects.
2234 if (get_user(flags, (int *) arg))
2237 if (cmd == LL_IOC_SETFLAGS) {
2238 if ((flags & LL_FILE_IGNORE_LOCK) &&
2239 !(file->f_flags & O_DIRECT)) {
2240 CERROR("%s: unable to disable locking on "
2241 "non-O_DIRECT file\n", current->comm);
2245 fd->fd_flags |= flags;
2247 fd->fd_flags &= ~flags;
2250 case LL_IOC_LOV_SETSTRIPE:
2251 RETURN(ll_lov_setstripe(inode, file, arg));
2252 case LL_IOC_LOV_SETEA:
2253 RETURN(ll_lov_setea(inode, file, arg));
2254 case LL_IOC_LOV_GETSTRIPE:
2255 RETURN(ll_lov_getstripe(inode, arg));
2256 case LL_IOC_RECREATE_OBJ:
2257 RETURN(ll_lov_recreate_obj(inode, file, arg));
2258 case EXT3_IOC_GETFLAGS:
2259 case EXT3_IOC_SETFLAGS:
2260 RETURN(ll_iocontrol(inode, file, cmd, arg));
2261 case EXT3_IOC_GETVERSION_OLD:
2262 case EXT3_IOC_GETVERSION:
2263 RETURN(put_user(inode->i_generation, (int *)arg));
2268 ftail = getname((const char *)arg);
2270 RETURN(PTR_ERR(ftail));
2271 rc = ll_file_join(inode, file, ftail);
2275 case LL_IOC_GROUP_LOCK:
2276 RETURN(ll_get_grouplock(inode, file, arg));
2277 case LL_IOC_GROUP_UNLOCK:
2278 RETURN(ll_put_grouplock(inode, file, arg));
2279 case IOC_OBD_STATFS:
2280 RETURN(ll_obd_statfs(inode, (void *)arg));
2282 /* We need to special case any other ioctls we want to handle,
2283 * to send them to the MDS/OST as appropriate and to properly
2284 * network encode the arg field.
2285 case EXT3_IOC_SETVERSION_OLD:
2286 case EXT3_IOC_SETVERSION:
2288 case LL_IOC_FLUSHCTX:
2289 RETURN(ll_flush_ctx(inode));
2290 case LL_IOC_GETFACL: {
2291 struct rmtacl_ioctl_data ioc;
2293 if (copy_from_user(&ioc, (void *)arg, sizeof(ioc)))
2296 RETURN(ll_ioctl_getfacl(inode, &ioc));
2298 case LL_IOC_SETFACL: {
2299 struct rmtacl_ioctl_data ioc;
2301 if (copy_from_user(&ioc, (void *)arg, sizeof(ioc)))
2304 RETURN(ll_ioctl_setfacl(inode, &ioc));
2310 ll_iocontrol_call(inode, file, cmd, arg, &err))
2313 RETURN(obd_iocontrol(cmd, ll_i2dtexp(inode), 0, NULL,
2319 loff_t ll_file_seek(struct file *file, loff_t offset, int origin)
2321 struct inode *inode = file->f_dentry->d_inode;
2322 struct ll_inode_info *lli = ll_i2info(inode);
2323 struct lov_stripe_md *lsm = lli->lli_smd;
2326 retval = offset + ((origin == 2) ? i_size_read(inode) :
2327 (origin == 1) ? file->f_pos : 0);
2328 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), to=%Lu=%#Lx(%s)\n",
2329 inode->i_ino, inode->i_generation, inode, retval, retval,
2330 origin == 2 ? "SEEK_END": origin == 1 ? "SEEK_CUR" : "SEEK_SET");
2331 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_LLSEEK, 1);
2333 if (origin == 2) { /* SEEK_END */
2334 int nonblock = 0, rc;
2336 if (file->f_flags & O_NONBLOCK)
2337 nonblock = LDLM_FL_BLOCK_NOWAIT;
2340 rc = ll_glimpse_size(inode, nonblock);
2345 ll_inode_size_lock(inode, 0);
2346 offset += i_size_read(inode);
2347 ll_inode_size_unlock(inode, 0);
2348 } else if (origin == 1) { /* SEEK_CUR */
2349 offset += file->f_pos;
2353 if (offset >= 0 && offset <= ll_file_maxbytes(inode)) {
2354 if (offset != file->f_pos) {
2355 file->f_pos = offset;
2356 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2358 file->f_version = ++event;
2367 int ll_fsync(struct file *file, struct dentry *dentry, int data)
2369 struct inode *inode = dentry->d_inode;
2370 struct ll_inode_info *lli = ll_i2info(inode);
2371 struct lov_stripe_md *lsm = lli->lli_smd;
2372 struct ptlrpc_request *req;
2373 struct obd_capa *oc;
2376 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2377 inode->i_generation, inode);
2378 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FSYNC, 1);
2380 /* fsync's caller has already called _fdata{sync,write}, we want
2381 * that IO to finish before calling the osc and mdc sync methods */
2382 rc = filemap_fdatawait(inode->i_mapping);
2384 /* catch async errors that were recorded back when async writeback
2385 * failed for pages in this mapping. */
2386 err = lli->lli_async_rc;
2387 lli->lli_async_rc = 0;
2391 err = lov_test_and_clear_async_rc(lsm);
2396 oc = ll_mdscapa_get(inode);
2397 err = md_sync(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2403 ptlrpc_req_finished(req);
2410 RETURN(rc ? rc : -ENOMEM);
2412 oa->o_id = lsm->lsm_object_id;
2413 oa->o_gr = lsm->lsm_object_gr;
2414 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2415 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
2416 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
2419 oc = ll_osscapa_get(inode, CAPA_OPC_OSS_WRITE);
2420 err = obd_sync(ll_i2sbi(inode)->ll_dt_exp, oa, lsm,
2421 0, OBD_OBJECT_EOF, oc);
2431 int ll_file_flock(struct file *file, int cmd, struct file_lock *file_lock)
2433 struct inode *inode = file->f_dentry->d_inode;
2434 struct ll_sb_info *sbi = ll_i2sbi(inode);
2435 struct ldlm_res_id res_id =
2436 { .name = { fid_seq(ll_inode2fid(inode)),
2437 fid_oid(ll_inode2fid(inode)),
2438 fid_ver(ll_inode2fid(inode)),
2440 struct ldlm_enqueue_info einfo = { LDLM_FLOCK, 0, NULL,
2441 ldlm_flock_completion_ast, NULL, file_lock };
2442 struct lustre_handle lockh = {0};
2443 ldlm_policy_data_t flock;
2448 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu file_lock=%p\n",
2449 inode->i_ino, file_lock);
2451 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_FLOCK, 1);
2453 if (file_lock->fl_flags & FL_FLOCK) {
2454 LASSERT((cmd == F_SETLKW) || (cmd == F_SETLK));
2455 /* set missing params for flock() calls */
2456 file_lock->fl_end = OFFSET_MAX;
2457 file_lock->fl_pid = current->tgid;
2459 flock.l_flock.pid = file_lock->fl_pid;
2460 flock.l_flock.start = file_lock->fl_start;
2461 flock.l_flock.end = file_lock->fl_end;
2463 switch (file_lock->fl_type) {
2465 einfo.ei_mode = LCK_PR;
2468 /* An unlock request may or may not have any relation to
2469 * existing locks so we may not be able to pass a lock handle
2470 * via a normal ldlm_lock_cancel() request. The request may even
2471 * unlock a byte range in the middle of an existing lock. In
2472 * order to process an unlock request we need all of the same
2473 * information that is given with a normal read or write record
2474 * lock request. To avoid creating another ldlm unlock (cancel)
2475 * message we'll treat a LCK_NL flock request as an unlock. */
2476 einfo.ei_mode = LCK_NL;
2479 einfo.ei_mode = LCK_PW;
2482 CERROR("unknown fcntl lock type: %d\n", file_lock->fl_type);
2497 flags = LDLM_FL_BLOCK_NOWAIT;
2503 flags = LDLM_FL_TEST_LOCK;
2504 /* Save the old mode so that if the mode in the lock changes we
2505 * can decrement the appropriate reader or writer refcount. */
2506 file_lock->fl_type = einfo.ei_mode;
2509 CERROR("unknown fcntl lock command: %d\n", cmd);
2513 CDEBUG(D_DLMTRACE, "inode=%lu, pid=%u, flags=%#x, mode=%u, "
2514 "start="LPU64", end="LPU64"\n", inode->i_ino, flock.l_flock.pid,
2515 flags, einfo.ei_mode, flock.l_flock.start, flock.l_flock.end);
2517 rc = ldlm_cli_enqueue(sbi->ll_md_exp, NULL, &einfo, &res_id,
2518 &flock, &flags, NULL, 0, NULL, &lockh, 0);
2519 if ((file_lock->fl_flags & FL_FLOCK) && (rc == 0))
2520 ll_flock_lock_file_wait(file, file_lock, (cmd == F_SETLKW));
2521 #ifdef HAVE_F_OP_FLOCK
2522 if ((file_lock->fl_flags & FL_POSIX) && (rc == 0) &&
2523 !(flags & LDLM_FL_TEST_LOCK))
2524 posix_lock_file_wait(file, file_lock);
2530 int ll_file_noflock(struct file *file, int cmd, struct file_lock *file_lock)
2537 int ll_have_md_lock(struct inode *inode, __u64 bits)
2539 struct lustre_handle lockh;
2540 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2548 fid = &ll_i2info(inode)->lli_fid;
2549 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2551 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING | LDLM_FL_TEST_LOCK;
2552 if (md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
2553 LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh)) {
2559 ldlm_mode_t ll_take_md_lock(struct inode *inode, __u64 bits,
2560 struct lustre_handle *lockh)
2562 ldlm_policy_data_t policy = { .l_inodebits = {bits}};
2568 fid = &ll_i2info(inode)->lli_fid;
2569 CDEBUG(D_INFO, "trying to match res "DFID"\n", PFID(fid));
2571 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
2572 rc = md_lock_match(ll_i2mdexp(inode), flags, fid, LDLM_IBITS, &policy,
2573 LCK_CR|LCK_CW|LCK_PR|LCK_PW, lockh);
2577 static int ll_inode_revalidate_fini(struct inode *inode, int rc) {
2578 if (rc == -ENOENT) { /* Already unlinked. Just update nlink
2579 * and return success */
2581 /* This path cannot be hit for regular files unless in
2582 * case of obscure races, so no need to to validate
2584 if (!S_ISREG(inode->i_mode) &&
2585 !S_ISDIR(inode->i_mode))
2590 CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2598 int ll_inode_revalidate_it(struct dentry *dentry, struct lookup_intent *it)
2600 struct inode *inode = dentry->d_inode;
2601 struct ptlrpc_request *req = NULL;
2602 struct ll_sb_info *sbi;
2603 struct obd_export *exp;
2608 CERROR("REPORT THIS LINE TO PETER\n");
2611 sbi = ll_i2sbi(inode);
2613 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p),name=%s\n",
2614 inode->i_ino, inode->i_generation, inode, dentry->d_name.name);
2616 exp = ll_i2mdexp(inode);
2618 if (exp->exp_connect_flags & OBD_CONNECT_ATTRFID) {
2619 struct lookup_intent oit = { .it_op = IT_GETATTR };
2620 struct md_op_data *op_data;
2622 /* Call getattr by fid, so do not provide name at all. */
2623 op_data = ll_prep_md_op_data(NULL, dentry->d_parent->d_inode,
2624 dentry->d_inode, NULL, 0, 0,
2625 LUSTRE_OPC_ANY, NULL);
2626 if (IS_ERR(op_data))
2627 RETURN(PTR_ERR(op_data));
2629 oit.it_flags |= O_CHECK_STALE;
2630 rc = md_intent_lock(exp, op_data, NULL, 0,
2631 /* we are not interested in name
2634 ll_md_blocking_ast, 0);
2635 ll_finish_md_op_data(op_data);
2636 oit.it_flags &= ~O_CHECK_STALE;
2638 rc = ll_inode_revalidate_fini(inode, rc);
2642 rc = ll_revalidate_it_finish(req, DLM_REPLY_REC_OFF, &oit, dentry);
2644 ll_intent_release(&oit);
2648 /* Unlinked? Unhash dentry, so it is not picked up later by
2649 do_lookup() -> ll_revalidate_it(). We cannot use d_drop
2650 here to preserve get_cwd functionality on 2.6.
2652 if (!dentry->d_inode->i_nlink) {
2653 spin_lock(&dcache_lock);
2654 ll_drop_dentry(dentry);
2655 spin_unlock(&dcache_lock);
2658 ll_lookup_finish_locks(&oit, dentry);
2659 } else if (!ll_have_md_lock(dentry->d_inode, MDS_INODELOCK_UPDATE)) {
2660 struct ll_sb_info *sbi = ll_i2sbi(dentry->d_inode);
2661 obd_valid valid = OBD_MD_FLGETATTR;
2662 struct obd_capa *oc;
2665 if (S_ISREG(inode->i_mode)) {
2666 rc = ll_get_max_mdsize(sbi, &ealen);
2669 valid |= OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE;
2671 /* Once OBD_CONNECT_ATTRFID is not supported, we can't find one
2672 * capa for this inode. Because we only keep capas of dirs
2674 oc = ll_mdscapa_get(inode);
2675 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc, valid,
2679 rc = ll_inode_revalidate_fini(inode, rc);
2683 rc = ll_prep_inode(&inode, req, REPLY_REC_OFF,
2689 /* if object not yet allocated, don't validate size */
2690 if (ll_i2info(inode)->lli_smd == NULL)
2693 /* ll_glimpse_size will prefer locally cached writes if they extend
2695 rc = ll_glimpse_size(inode, 0);
2698 ptlrpc_req_finished(req);
2702 int ll_getattr_it(struct vfsmount *mnt, struct dentry *de,
2703 struct lookup_intent *it, struct kstat *stat)
2705 struct inode *inode = de->d_inode;
2708 res = ll_inode_revalidate_it(de, it);
2709 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_GETATTR, 1);
2714 stat->dev = inode->i_sb->s_dev;
2715 stat->ino = inode->i_ino;
2716 stat->mode = inode->i_mode;
2717 stat->nlink = inode->i_nlink;
2718 stat->uid = inode->i_uid;
2719 stat->gid = inode->i_gid;
2720 stat->rdev = kdev_t_to_nr(inode->i_rdev);
2721 stat->atime = inode->i_atime;
2722 stat->mtime = inode->i_mtime;
2723 stat->ctime = inode->i_ctime;
2724 #ifdef HAVE_INODE_BLKSIZE
2725 stat->blksize = inode->i_blksize;
2727 stat->blksize = 1 << inode->i_blkbits;
2730 ll_inode_size_lock(inode, 0);
2731 stat->size = i_size_read(inode);
2732 stat->blocks = inode->i_blocks;
2733 ll_inode_size_unlock(inode, 0);
2737 int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat)
2739 struct lookup_intent it = { .it_op = IT_GETATTR };
2741 return ll_getattr_it(mnt, de, &it, stat);
2745 int lustre_check_acl(struct inode *inode, int mask)
2747 #ifdef CONFIG_FS_POSIX_ACL
2748 struct ll_inode_info *lli = ll_i2info(inode);
2749 struct posix_acl *acl;
2753 spin_lock(&lli->lli_lock);
2754 acl = posix_acl_dup(lli->lli_posix_acl);
2755 spin_unlock(&lli->lli_lock);
2760 rc = posix_acl_permission(inode, acl, mask);
2761 posix_acl_release(acl);
2769 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,10))
2770 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2772 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n",
2773 inode->i_ino, inode->i_generation, inode, mask);
2774 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2775 return lustre_check_remote_perm(inode, mask);
2777 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2778 return generic_permission(inode, mask, lustre_check_acl);
2781 int ll_inode_permission(struct inode *inode, int mask, struct nameidata *nd)
2783 int mode = inode->i_mode;
2786 CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), mask %o\n",
2787 inode->i_ino, inode->i_generation, inode, mask);
2789 if (ll_i2sbi(inode)->ll_flags & LL_SBI_RMT_CLIENT)
2790 return lustre_check_remote_perm(inode, mask);
2792 ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_INODE_PERM, 1);
2794 if ((mask & MAY_WRITE) && IS_RDONLY(inode) &&
2795 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
2797 if ((mask & MAY_WRITE) && IS_IMMUTABLE(inode))
2799 if (current->fsuid == inode->i_uid) {
2802 if (((mode >> 3) & mask & S_IRWXO) != mask)
2804 rc = lustre_check_acl(inode, mask);
2808 goto check_capabilities;
2812 if (in_group_p(inode->i_gid))
2815 if ((mode & mask & S_IRWXO) == mask)
2819 if (!(mask & MAY_EXEC) ||
2820 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
2821 if (capable(CAP_DAC_OVERRIDE))
2824 if (capable(CAP_DAC_READ_SEARCH) && ((mask == MAY_READ) ||
2825 (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))))
2832 /* -o localflock - only provides locally consistent flock locks */
2833 struct file_operations ll_file_operations = {
2834 .read = ll_file_read,
2835 .write = ll_file_write,
2836 .ioctl = ll_file_ioctl,
2837 .open = ll_file_open,
2838 .release = ll_file_release,
2839 .mmap = ll_file_mmap,
2840 .llseek = ll_file_seek,
2841 .sendfile = ll_file_sendfile,
2845 struct file_operations ll_file_operations_flock = {
2846 .read = ll_file_read,
2847 .write = ll_file_write,
2848 .ioctl = ll_file_ioctl,
2849 .open = ll_file_open,
2850 .release = ll_file_release,
2851 .mmap = ll_file_mmap,
2852 .llseek = ll_file_seek,
2853 .sendfile = ll_file_sendfile,
2855 #ifdef HAVE_F_OP_FLOCK
2856 .flock = ll_file_flock,
2858 .lock = ll_file_flock
2861 /* These are for -o noflock - to return ENOSYS on flock calls */
2862 struct file_operations ll_file_operations_noflock = {
2863 .read = ll_file_read,
2864 .write = ll_file_write,
2865 .ioctl = ll_file_ioctl,
2866 .open = ll_file_open,
2867 .release = ll_file_release,
2868 .mmap = ll_file_mmap,
2869 .llseek = ll_file_seek,
2870 .sendfile = ll_file_sendfile,
2872 #ifdef HAVE_F_OP_FLOCK
2873 .flock = ll_file_noflock,
2875 .lock = ll_file_noflock
2878 struct inode_operations ll_file_inode_operations = {
2879 #ifdef HAVE_VFS_INTENT_PATCHES
2880 .setattr_raw = ll_setattr_raw,
2882 .setattr = ll_setattr,
2883 .truncate = ll_truncate,
2884 .getattr = ll_getattr,
2885 .permission = ll_inode_permission,
2886 .setxattr = ll_setxattr,
2887 .getxattr = ll_getxattr,
2888 .listxattr = ll_listxattr,
2889 .removexattr = ll_removexattr,
2892 /* dynamic ioctl number support routins */
2893 static struct llioc_ctl_data {
2894 struct rw_semaphore ioc_sem;
2895 struct list_head ioc_head;
2897 __RWSEM_INITIALIZER(llioc.ioc_sem),
2898 CFS_LIST_HEAD_INIT(llioc.ioc_head)
2903 struct list_head iocd_list;
2904 unsigned int iocd_size;
2905 llioc_callback_t iocd_cb;
2906 unsigned int iocd_count;
2907 unsigned int iocd_cmd[0];
2910 void *ll_iocontrol_register(llioc_callback_t cb, int count, unsigned int *cmd)
2913 struct llioc_data *in_data = NULL;
2916 if (cb == NULL || cmd == NULL ||
2917 count > LLIOC_MAX_CMD || count < 0)
2920 size = sizeof(*in_data) + count * sizeof(unsigned int);
2921 OBD_ALLOC(in_data, size);
2922 if (in_data == NULL)
2925 memset(in_data, 0, sizeof(*in_data));
2926 in_data->iocd_size = size;
2927 in_data->iocd_cb = cb;
2928 in_data->iocd_count = count;
2929 memcpy(in_data->iocd_cmd, cmd, sizeof(unsigned int) * count);
2931 down_write(&llioc.ioc_sem);
2932 list_add_tail(&in_data->iocd_list, &llioc.ioc_head);
2933 up_write(&llioc.ioc_sem);
2938 void ll_iocontrol_unregister(void *magic)
2940 struct llioc_data *tmp;
2945 down_write(&llioc.ioc_sem);
2946 list_for_each_entry(tmp, &llioc.ioc_head, iocd_list) {
2948 unsigned int size = tmp->iocd_size;
2950 list_del(&tmp->iocd_list);
2951 up_write(&llioc.ioc_sem);
2953 OBD_FREE(tmp, size);
2957 up_write(&llioc.ioc_sem);
2959 CWARN("didn't find iocontrol register block with magic: %p\n", magic);
2962 EXPORT_SYMBOL(ll_iocontrol_register);
2963 EXPORT_SYMBOL(ll_iocontrol_unregister);
2965 enum llioc_iter ll_iocontrol_call(struct inode *inode, struct file *file,
2966 unsigned int cmd, unsigned long arg, int *rcp)
2968 enum llioc_iter ret = LLIOC_CONT;
2969 struct llioc_data *data;
2970 int rc = -EINVAL, i;
2972 down_read(&llioc.ioc_sem);
2973 list_for_each_entry(data, &llioc.ioc_head, iocd_list) {
2974 for (i = 0; i < data->iocd_count; i++) {
2975 if (cmd != data->iocd_cmd[i])
2978 ret = data->iocd_cb(inode, file, cmd, arg, data, &rc);
2982 if (ret == LLIOC_STOP)
2985 up_read(&llioc.ioc_sem);