dmake: do not set MAKEFLAGS=k
[unleashed/tickless.git] / kernel / fs / ufs / ufs_filio.c
blob7c94ddf004af58ca92a836f245d584e1b26f8347
1 /*
2 * CDDL HEADER START
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
19 * CDDL HEADER END
22 * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/types.h>
26 #include <sys/t_lock.h>
27 #include <sys/param.h>
28 #include <sys/time.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/resource.h>
32 #include <sys/signal.h>
33 #include <sys/cred.h>
34 #include <sys/user.h>
35 #include <sys/buf.h>
36 #include <sys/vfs.h>
37 #include <sys/vnode.h>
38 #include <sys/proc.h>
39 #include <sys/disp.h>
40 #include <sys/file.h>
41 #include <sys/fcntl.h>
42 #include <sys/flock.h>
43 #include <sys/atomic.h>
44 #include <sys/kmem.h>
45 #include <sys/uio.h>
46 #include <sys/conf.h>
47 #include <sys/mman.h>
48 #include <sys/pathname.h>
49 #include <sys/debug.h>
50 #include <sys/vmsystm.h>
51 #include <sys/cmn_err.h>
52 #include <sys/filio.h>
53 #include <sys/dnlc.h>
55 #include <sys/fs/ufs_filio.h>
56 #include <sys/fs/ufs_lockfs.h>
57 #include <sys/fs/ufs_fs.h>
58 #include <sys/fs/ufs_inode.h>
59 #include <sys/fs/ufs_fsdir.h>
60 #include <sys/fs/ufs_quota.h>
61 #include <sys/fs/ufs_trans.h>
62 #include <sys/fs/ufs_log.h>
63 #include <sys/dirent.h> /* must be AFTER <sys/fs/fsdir.h>! */
64 #include <sys/errno.h>
65 #include <sys/sysinfo.h>
67 #include <vm/hat.h>
68 #include <vm/page.h>
69 #include <vm/pvn.h>
70 #include <vm/as.h>
71 #include <vm/seg.h>
72 #include <vm/seg_map.h>
73 #include <vm/seg_vn.h>
74 #include <vm/rm.h>
75 #include <sys/swap.h>
76 #include <sys/model.h>
77 #include <sys/policy.h>
79 #include "sys/fs_subr.h"
82 * ufs_fioio is the ufs equivalent of NFS_CNVT and is tailored to
83 * metamucil's needs. It may change at any time.
85 /* ARGSUSED */
86 int
87 ufs_fioio(
88 struct vnode *vp, /* any file on the fs */
89 struct fioio *fiou, /* fioio struct in userland */
90 int flag, /* flag from fop_ioctl() */
91 struct cred *cr) /* credentials from ufs_ioctl */
93 int error = 0;
94 struct vnode *vpio = NULL; /* vnode for inode open */
95 struct inode *ipio = NULL; /* inode for inode open */
96 struct file *fpio = NULL; /* file for inode open */
97 struct inode *ip; /* inode for file system */
98 struct fs *fs; /* fs for file system */
99 STRUCT_DECL(fioio, fio); /* copy of user's fioio struct */
102 * must be privileged
104 if (secpolicy_fs_config(cr, vp->v_vfsp) != 0)
105 return (EPERM);
107 STRUCT_INIT(fio, flag & DATAMODEL_MASK);
110 * get user's copy of fioio struct
112 if (copyin(fiou, STRUCT_BUF(fio), STRUCT_SIZE(fio)))
113 return (EFAULT);
115 ip = VTOI(vp);
116 fs = ip->i_fs;
119 * check the inode number against the fs's inode number bounds
121 if (STRUCT_FGET(fio, fio_ino) < UFSROOTINO)
122 return (ESRCH);
123 if (STRUCT_FGET(fio, fio_ino) >= fs->fs_ncg * fs->fs_ipg)
124 return (ESRCH);
126 rw_enter(&ip->i_ufsvfs->vfs_dqrwlock, RW_READER);
129 * get the inode
131 error = ufs_iget(ip->i_vfs, STRUCT_FGET(fio, fio_ino), &ipio, cr);
133 rw_exit(&ip->i_ufsvfs->vfs_dqrwlock);
135 if (error)
136 return (error);
139 * check the generation number
141 rw_enter(&ipio->i_contents, RW_READER);
142 if (ipio->i_gen != STRUCT_FGET(fio, fio_gen)) {
143 error = ESTALE;
144 rw_exit(&ipio->i_contents);
145 goto errout;
149 * check if the inode is free
151 if (ipio->i_mode == 0) {
152 error = ENOENT;
153 rw_exit(&ipio->i_contents);
154 goto errout;
156 rw_exit(&ipio->i_contents);
159 * Adapted from copen: get a file struct
160 * Large Files: We open this file descriptor with FOFFMAX flag
161 * set so that it will be like a large file open.
163 if (falloc(NULL, (FREAD|FOFFMAX), &fpio, STRUCT_FADDR(fio, fio_fd)))
164 goto errout;
167 * Adapted from vn_open: check access and then open the file
169 vpio = ITOV(ipio);
170 if (error = fop_access(vpio, VREAD, 0, cr, NULL))
171 goto errout;
173 if (error = fop_open(&vpio, FREAD, cr, NULL))
174 goto errout;
177 * Adapted from copen: initialize the file struct
179 fpio->f_vnode = vpio;
182 * return the fd
184 if (copyout(STRUCT_BUF(fio), fiou, STRUCT_SIZE(fio))) {
185 error = EFAULT;
186 goto errout;
188 setf(STRUCT_FGET(fio, fio_fd), fpio);
189 mutex_exit(&fpio->f_tlock);
190 return (0);
191 errout:
193 * free the file struct and fd
195 if (fpio) {
196 setf(STRUCT_FGET(fio, fio_fd), NULL);
197 unfalloc(fpio);
201 * release the hold on the inode
203 if (ipio)
204 VN_RELE(ITOV(ipio));
205 return (error);
209 * ufs_fiosatime
210 * set access time w/o altering change time. This ioctl is tailored
211 * to metamucil's needs and may change at any time.
214 ufs_fiosatime(
215 struct vnode *vp, /* file's vnode */
216 struct timeval *tvu, /* struct timeval in userland */
217 int flag, /* flag from fop_ioctl() */
218 struct cred *cr) /* credentials from ufs_ioctl */
220 struct inode *ip; /* inode for vp */
221 struct timeval32 tv; /* copy of user's timeval */
222 int now = 0;
225 * must have sufficient privileges
227 if (secpolicy_fs_config(cr, vp->v_vfsp) != 0)
228 return (EPERM);
231 * get user's copy of timeval struct and check values
232 * if input is NULL, will set time to now
234 if (tvu == NULL) {
235 now = 1;
236 } else {
237 if ((flag & DATAMODEL_MASK) == DATAMODEL_ILP32) {
238 if (copyin(tvu, &tv, sizeof (tv)))
239 return (EFAULT);
240 } else {
241 struct timeval tv64;
243 if (copyin(tvu, &tv64, sizeof (tv64)))
244 return (EFAULT);
245 if (TIMEVAL_OVERFLOW(&tv64))
246 return (EOVERFLOW);
247 TIMEVAL_TO_TIMEVAL32(&tv, &tv64);
250 if (tv.tv_usec < 0 || tv.tv_usec >= 1000000)
251 return (EINVAL);
255 * update access time
257 ip = VTOI(vp);
258 rw_enter(&ip->i_contents, RW_WRITER);
259 ITIMES_NOLOCK(ip);
260 if (now) {
261 mutex_enter(&ufs_iuniqtime_lock);
262 ip->i_atime = iuniqtime;
263 mutex_exit(&ufs_iuniqtime_lock);
264 } else {
265 ip->i_atime = tv;
267 ip->i_flag |= IMODACC;
268 rw_exit(&ip->i_contents);
270 return (0);
274 * ufs_fiogdio
275 * Get delayed-io state. This ioctl is tailored
276 * to metamucil's needs and may change at any time.
278 /* ARGSUSED */
280 ufs_fiogdio(
281 struct vnode *vp, /* file's vnode */
282 uint_t *diop, /* dio state returned here */
283 int flag, /* flag from ufs_ioctl */
284 struct cred *cr) /* credentials from ufs_ioctl */
286 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
289 * forcibly unmounted
291 if (ufsvfsp == NULL)
292 return (EIO);
294 if (suword32(diop, ufsvfsp->vfs_dio))
295 return (EFAULT);
296 return (0);
300 * ufs_fiosdio
301 * Set delayed-io state. This ioctl is tailored
302 * to metamucil's needs and may change at any time.
305 ufs_fiosdio(
306 struct vnode *vp, /* file's vnode */
307 uint_t *diop, /* dio flag */
308 int flag, /* flag from ufs_ioctl */
309 struct cred *cr) /* credentials from ufs_ioctl */
311 uint_t dio; /* copy of user's dio */
312 struct inode *ip; /* inode for vp */
313 struct ufsvfs *ufsvfsp;
314 struct fs *fs;
315 struct ulockfs *ulp;
316 int error = 0;
319 /* check input conditions */
320 if (secpolicy_fs_config(cr, vp->v_vfsp) != 0)
321 return (EPERM);
323 if (copyin(diop, &dio, sizeof (dio)))
324 return (EFAULT);
326 if (dio > 1)
327 return (EINVAL);
329 /* file system has been forcibly unmounted */
330 if (VTOI(vp)->i_ufsvfs == NULL)
331 return (EIO);
333 ip = VTOI(vp);
334 ufsvfsp = ip->i_ufsvfs;
335 ulp = &ufsvfsp->vfs_ulockfs;
337 /* logging file system; dio ignored */
338 if (TRANS_ISTRANS(ufsvfsp))
339 return (error);
341 /* hold the mutex to prevent race with a lockfs request */
342 vfs_lock_wait(vp->v_vfsp);
343 mutex_enter(&ulp->ul_lock);
344 atomic_inc_ulong(&ufs_quiesce_pend);
346 if (ULOCKFS_IS_HLOCK(ulp)) {
347 error = EIO;
348 goto out;
351 if (ULOCKFS_IS_ELOCK(ulp)) {
352 error = EBUSY;
353 goto out;
355 /* wait for outstanding accesses to finish */
356 if (error = ufs_quiesce(ulp))
357 goto out;
359 /* flush w/invalidate */
360 if (error = ufs_flush(vp->v_vfsp))
361 goto out;
364 * update dio
366 mutex_enter(&ufsvfsp->vfs_lock);
367 ufsvfsp->vfs_dio = dio;
370 * enable/disable clean flag processing
372 fs = ip->i_fs;
373 if (fs->fs_ronly == 0 &&
374 fs->fs_clean != FSBAD &&
375 fs->fs_clean != FSLOG) {
376 if (dio)
377 fs->fs_clean = FSSUSPEND;
378 else
379 fs->fs_clean = FSACTIVE;
380 ufs_sbwrite(ufsvfsp);
381 mutex_exit(&ufsvfsp->vfs_lock);
382 } else
383 mutex_exit(&ufsvfsp->vfs_lock);
384 out:
386 * we need this broadcast because of the ufs_quiesce call above
388 atomic_dec_ulong(&ufs_quiesce_pend);
389 cv_broadcast(&ulp->ul_cv);
390 mutex_exit(&ulp->ul_lock);
391 vfs_unlock(vp->v_vfsp);
392 return (error);
396 * ufs_fioffs - ioctl handler for flushing file system
398 /* ARGSUSED */
400 ufs_fioffs(
401 struct vnode *vp,
402 char *vap, /* must be NULL - reserved */
403 struct cred *cr) /* credentials from ufs_ioctl */
405 int error;
406 struct ufsvfs *ufsvfsp;
407 struct ulockfs *ulp;
409 /* file system has been forcibly unmounted */
410 ufsvfsp = VTOI(vp)->i_ufsvfs;
411 if (ufsvfsp == NULL)
412 return (EIO);
414 ulp = &ufsvfsp->vfs_ulockfs;
417 * suspend the delete thread
418 * this must be done outside the lockfs locking protocol
420 vfs_lock_wait(vp->v_vfsp);
421 ufs_thread_suspend(&ufsvfsp->vfs_delete);
423 /* hold the mutex to prevent race with a lockfs request */
424 mutex_enter(&ulp->ul_lock);
425 atomic_inc_ulong(&ufs_quiesce_pend);
427 if (ULOCKFS_IS_HLOCK(ulp)) {
428 error = EIO;
429 goto out;
431 if (ULOCKFS_IS_ELOCK(ulp)) {
432 error = EBUSY;
433 goto out;
435 /* wait for outstanding accesses to finish */
436 if (error = ufs_quiesce(ulp))
437 goto out;
440 * If logging, and the logmap was marked as not rollable,
441 * make it rollable now, and start the trans_roll thread and
442 * the reclaim thread. The log at this point is safe to write to.
444 if (ufsvfsp->vfs_log) {
445 ml_unit_t *ul = ufsvfsp->vfs_log;
446 struct fs *fsp = ufsvfsp->vfs_fs;
447 int err;
449 if (ul->un_flags & LDL_NOROLL) {
450 ul->un_flags &= ~LDL_NOROLL;
451 logmap_start_roll(ul);
452 if (!fsp->fs_ronly && (fsp->fs_reclaim &
453 (FS_RECLAIM|FS_RECLAIMING))) {
454 fsp->fs_reclaim &= ~FS_RECLAIM;
455 fsp->fs_reclaim |= FS_RECLAIMING;
456 ufs_thread_start(&ufsvfsp->vfs_reclaim,
457 ufs_thread_reclaim, vp->v_vfsp);
458 if (!fsp->fs_ronly) {
459 TRANS_SBWRITE(ufsvfsp,
460 TOP_SBUPDATE_UPDATE);
461 if (err =
462 geterror(ufsvfsp->vfs_bufp)) {
463 refstr_t *mntpt;
464 mntpt = vfs_getmntpoint(
465 vp->v_vfsp);
466 cmn_err(CE_NOTE,
467 "Filesystem Flush "
468 "Failed to update "
469 "Reclaim Status for "
470 " %s, Write failed to "
471 "update superblock, "
472 "error %d",
473 refstr_value(mntpt),
474 err);
475 refstr_rele(mntpt);
482 /* synchronously flush dirty data and metadata */
483 error = ufs_flush(vp->v_vfsp);
485 out:
486 atomic_dec_ulong(&ufs_quiesce_pend);
487 cv_broadcast(&ulp->ul_cv);
488 mutex_exit(&ulp->ul_lock);
489 vfs_unlock(vp->v_vfsp);
492 * allow the delete thread to continue
494 ufs_thread_continue(&ufsvfsp->vfs_delete);
495 return (error);
499 * ufs_fioisbusy
500 * Get number of references on this vnode.
501 * Contract-private interface for Legato's NetWorker product.
503 /* ARGSUSED */
505 ufs_fioisbusy(struct vnode *vp, int *isbusy, struct cred *cr)
507 int is_it_busy;
510 * The caller holds one reference, there may be one in the dnlc
511 * so we need to flush it.
513 if (vp->v_count > 1)
514 dnlc_purge_vp(vp);
516 * Since we've just flushed the dnlc and we hold a reference
517 * to this vnode, then anything but 1 means busy (this had
518 * BETTER not be zero!). Also, it's possible for someone to
519 * have this file mmap'ed with no additional reference count.
521 ASSERT(vp->v_count > 0);
522 if ((vp->v_count == 1) && (VTOI(vp)->i_mapcnt == 0))
523 is_it_busy = 0;
524 else
525 is_it_busy = 1;
527 if (suword32(isbusy, is_it_busy))
528 return (EFAULT);
529 return (0);
532 /* ARGSUSED */
534 ufs_fiodirectio(struct vnode *vp, int cmd, struct cred *cr)
536 int error = 0;
537 struct inode *ip = VTOI(vp);
540 * Acquire reader lock and set/reset direct mode
542 rw_enter(&ip->i_contents, RW_READER);
543 mutex_enter(&ip->i_tlock);
544 if (cmd == DIRECTIO_ON)
545 ip->i_flag |= IDIRECTIO; /* enable direct mode */
546 else if (cmd == DIRECTIO_OFF)
547 ip->i_flag &= ~IDIRECTIO; /* disable direct mode */
548 else
549 error = EINVAL;
550 mutex_exit(&ip->i_tlock);
551 rw_exit(&ip->i_contents);
552 return (error);
556 * ufs_fiotune
557 * Allow some tunables to be set on a mounted fs
560 ufs_fiotune(struct vnode *vp, struct fiotune *uftp, struct cred *cr)
562 struct fiotune ftp;
563 struct fs *fs;
564 struct ufsvfs *ufsvfsp;
567 * must have sufficient privileges
569 if (secpolicy_fs_config(cr, vp->v_vfsp) != 0)
570 return (EPERM);
573 * get user's copy
575 if (copyin(uftp, &ftp, sizeof (ftp)))
576 return (EFAULT);
579 * some minimal sanity checks
581 if ((ftp.maxcontig <= 0) ||
582 (ftp.rotdelay != 0) ||
583 (ftp.maxbpg <= 0) ||
584 (ftp.minfree < 0) ||
585 (ftp.minfree > 99) ||
586 ((ftp.optim != FS_OPTTIME) && (ftp.optim != FS_OPTSPACE)))
587 return (EINVAL);
590 * update superblock but don't write it! If it gets out, fine.
592 fs = VTOI(vp)->i_fs;
594 fs->fs_maxcontig = ftp.maxcontig;
595 fs->fs_rotdelay = ftp.rotdelay;
596 fs->fs_maxbpg = ftp.maxbpg;
597 fs->fs_minfree = ftp.minfree;
598 fs->fs_optim = ftp.optim;
601 * Adjust cluster based on the new maxcontig. The cluster size
602 * can be any positive value. The check for this is done above.
604 ufsvfsp = VTOI(vp)->i_ufsvfs;
605 ufsvfsp->vfs_ioclustsz = fs->fs_bsize * fs->fs_maxcontig;
608 * Adjust minfrags from minfree
610 ufsvfsp->vfs_minfrags = (int)((int64_t)fs->fs_dsize *
611 fs->fs_minfree / 100);
614 * Write the superblock
616 if (fs->fs_ronly == 0) {
617 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SBUPDATE_UPDATE,
618 TOP_SBWRITE_SIZE);
619 TRANS_SBWRITE(ufsvfsp, TOP_SBUPDATE_UPDATE);
620 TRANS_END_ASYNC(ufsvfsp, TOP_SBUPDATE_UPDATE, TOP_SBWRITE_SIZE);
623 return (0);
627 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
628 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
631 ufs_fio_holey(vnode_t *vp, int cmd, offset_t *off)
633 inode_t *ip = VTOI(vp);
634 uoff_t noff = (uoff_t)*off; /* new offset */
635 uoff_t isz;
636 int error;
637 boolean_t hole;
639 rw_enter(&ip->i_contents, RW_READER);
640 isz = ip->i_size;
641 if (noff >= isz) {
642 rw_exit(&ip->i_contents);
643 return (ENXIO);
647 * Check for the usual case where a file has no holes.
648 * If so we can optimise to set the end of the file as the first
649 * (virtual) hole. This avoids bmap_find() searching through
650 * every block in the file for a (non-existent) hole.
652 if (!bmap_has_holes(ip)) {
653 rw_exit(&ip->i_contents);
654 if (cmd == _FIO_SEEK_HOLE) {
655 *off = isz;
656 return (0);
658 /* *off must already point to valid data (non hole) */
659 return (0);
663 * Calling bmap_read() one block at a time on a 1TB file takes forever,
664 * so we use a special function to search for holes or blocks.
666 if (cmd == _FIO_SEEK_HOLE)
667 hole = B_TRUE;
668 else
669 hole = B_FALSE;
670 error = bmap_find(ip, hole, &noff);
671 rw_exit(&ip->i_contents);
673 /* end of file? */
674 if (error == ENXIO) {
676 * Handle the virtual hole at the end of file.
678 if (cmd == _FIO_SEEK_HOLE) {
679 *off = isz;
680 return (0);
682 return (ENXIO);
684 if (noff < *off)
685 return (error);
686 *off = noff;
687 return (error);
691 ufs_mark_compressed(struct vnode *vp)
693 struct inode *ip = VTOI(vp);
694 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
696 if (vp->v_type != VREG)
697 return (EINVAL);
699 rw_enter(&ip->i_contents, RW_WRITER);
700 ip->i_cflags |= ICOMPRESS;
701 TRANS_INODE(ufsvfsp, ip);
702 ip->i_flag |= (ICHG|ISEQ);
703 ip->i_seq++;
704 if (!TRANS_ISTRANS(ufsvfsp))
705 ufs_iupdat(ip, I_ASYNC);
706 rw_exit(&ip->i_contents);
708 return (0);