Print this page
NEX-3562 filename normalization doesn't work for removes (sync with upstream)
NEX-8972 Async-delete side-effect that may cause unmount EBUSY
Reviewed by: Alek Pinchuk <alek@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
NEX-8466 Failed to do 'pkg update' because of beadm failed to create BE, because of EBUSY in umount
Reviewed by: Evan Layton <evan.layton@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
NEX-5108 make deletes interruptible on pool export
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Reviewed by: Sanjay Nadkarni <sanjay.nadkarni@nexenta.com>
NEX-5036 case-insensitive matching is not working on casesensitivity=mixed, normalization=none FS
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-3562 filename normalization doesn't work for removes
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-4290 'insensitive' case sensitivity is broken
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
NEX-3762 Appliance crashes with a NULL pointer dereference during a zpool export when a zfs_vn_rele_taskq thread attempts to check a bogus rwlock from rw_write_held
Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com>
6328 Fix cstyle errors in zfs codebase (fix studio)
6328 Fix cstyle errors in zfs codebase
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Alex Reece <alex@delphix.com>
Reviewed by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed by: Jorgen Lundman <lundman@lundman.net>
Approved by: Robert Mustacchi <rm@joyent.com>
NEX-3329 libnsl: set_up_connection() over TCP does not adhere the specified timeout
Reviewed by: Dan Fields <dan.fields@nexenta.com>
NEX-3521 CLONE - Port NEX-3209 normalization=formD and casesensitivity=mixed behaves improperly, squashing case
Reviewed by: Jean McCormack <jean.mccormack@nexenta.com>
Reviewed by: Saso Kiselkov <saso.kiselkov@nexenta.com>
Reviewed by: Dan Fields <dan.fields@nexenta.com>
NEX-2965 4.0.3-FP2: deferred deletes causing RSF import failure during fail-over of service
Reviewed by: Josef Sipek <josef.sipek@nexenta.com>
Reviewed by: Alek Pinchuk <alek.pinchuk@nexenta.com>
SUP-780 zfs with casesensitivity=mixed and Unicode normalization doesn't normalize properly/fully
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/zfs_dir.c
+++ new/usr/src/uts/common/fs/zfs/zfs_dir.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
25 25 * Copyright 2017 Nexenta Systems, Inc.
26 26 * Copyright (c) 2015, Joyent, Inc.
27 27 */
28 28
29 29 #include <sys/types.h>
30 30 #include <sys/param.h>
31 31 #include <sys/time.h>
32 32 #include <sys/systm.h>
33 33 #include <sys/sysmacros.h>
34 34 #include <sys/resource.h>
35 35 #include <sys/vfs.h>
36 36 #include <sys/vnode.h>
37 37 #include <sys/file.h>
38 38 #include <sys/mode.h>
39 39 #include <sys/kmem.h>
40 40 #include <sys/uio.h>
41 41 #include <sys/pathname.h>
42 42 #include <sys/cmn_err.h>
43 43 #include <sys/errno.h>
44 44 #include <sys/stat.h>
|
↓ open down ↓ |
44 lines elided |
↑ open up ↑ |
45 45 #include <sys/unistd.h>
46 46 #include <sys/sunddi.h>
47 47 #include <sys/random.h>
48 48 #include <sys/policy.h>
49 49 #include <sys/zfs_dir.h>
50 50 #include <sys/zfs_acl.h>
51 51 #include <sys/fs/zfs.h>
52 52 #include "fs/fs_subr.h"
53 53 #include <sys/zap.h>
54 54 #include <sys/dmu.h>
55 +#include <sys/dmu_objset.h>
55 56 #include <sys/atomic.h>
56 57 #include <sys/zfs_ctldir.h>
57 58 #include <sys/zfs_fuid.h>
58 59 #include <sys/sa.h>
59 60 #include <sys/zfs_sa.h>
60 61 #include <sys/dnlc.h>
61 62 #include <sys/extdirent.h>
62 63
63 64 /*
64 65 * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
65 66 * of names after deciding which is the appropriate lookup interface.
66 67 */
67 68 static int
68 69 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, matchtype_t mt,
69 70 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
70 71 {
71 72 int error;
72 73
73 74 if (zfsvfs->z_norm) {
74 75 boolean_t conflict = B_FALSE;
75 76 size_t bufsz = 0;
76 77 char *buf = NULL;
77 78
78 79 if (rpnp) {
79 80 buf = rpnp->pn_buf;
80 81 bufsz = rpnp->pn_bufsize;
81 82 }
82 83
83 84 /*
84 85 * In the non-mixed case we only expect there would ever
85 86 * be one match, but we need to use the normalizing lookup.
86 87 */
87 88 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
88 89 zoid, mt, buf, bufsz, &conflict);
89 90 if (!error && deflags)
90 91 *deflags = conflict ? ED_CASE_CONFLICT : 0;
91 92 } else {
92 93 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
93 94 }
94 95 *zoid = ZFS_DIRENT_OBJ(*zoid);
95 96
96 97 if (error == ENOENT && update)
97 98 dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
98 99
99 100 return (error);
100 101 }
101 102
102 103 /*
103 104 * Lock a directory entry. A dirlock on <dzp, name> protects that name
104 105 * in dzp's directory zap object. As long as you hold a dirlock, you can
105 106 * assume two things: (1) dzp cannot be reaped, and (2) no other thread
106 107 * can change the zap entry for (i.e. link or unlink) this name.
107 108 *
108 109 * Input arguments:
109 110 * dzp - znode for directory
110 111 * name - name of entry to lock
111 112 * flag - ZNEW: if the entry already exists, fail with EEXIST.
112 113 * ZEXISTS: if the entry does not exist, fail with ENOENT.
113 114 * ZSHARED: allow concurrent access with other ZSHARED callers.
114 115 * ZXATTR: we want dzp's xattr directory
115 116 * ZCILOOK: On a mixed sensitivity file system,
116 117 * this lookup should be case-insensitive.
117 118 * ZCIEXACT: On a purely case-insensitive file system,
118 119 * this lookup should be case-sensitive.
119 120 * ZRENAMING: we are locking for renaming, force narrow locks
120 121 * ZHAVELOCK: Don't grab the z_name_lock for this call. The
121 122 * current thread already holds it.
122 123 *
123 124 * Output arguments:
124 125 * zpp - pointer to the znode for the entry (NULL if there isn't one)
125 126 * dlpp - pointer to the dirlock for this entry (NULL on error)
126 127 * direntflags - (case-insensitive lookup only)
127 128 * flags if multiple case-sensitive matches exist in directory
128 129 * realpnp - (case-insensitive lookup only)
129 130 * actual name matched within the directory
130 131 *
131 132 * Return value: 0 on success or errno on failure.
132 133 *
133 134 * NOTE: Always checks for, and rejects, '.' and '..'.
134 135 * NOTE: For case-insensitive file systems we take wide locks (see below),
135 136 * but return znode pointers to a single match.
136 137 */
137 138 int
138 139 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
139 140 int flag, int *direntflags, pathname_t *realpnp)
140 141 {
141 142 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
142 143 zfs_dirlock_t *dl;
143 144 boolean_t update;
144 145 matchtype_t mt = 0;
145 146 uint64_t zoid;
146 147 vnode_t *vp = NULL;
147 148 int error = 0;
148 149 int cmpflags;
149 150
150 151 *zpp = NULL;
151 152 *dlpp = NULL;
152 153
153 154 /*
154 155 * Verify that we are not trying to lock '.', '..', or '.zfs'
155 156 */
156 157 if (name[0] == '.' &&
157 158 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
158 159 zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
159 160 return (SET_ERROR(EEXIST));
160 161
161 162 /*
162 163 * Case sensitivity and normalization preferences are set when
163 164 * the file system is created. These are stored in the
164 165 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices
165 166 * affect what vnodes can be cached in the DNLC, how we
166 167 * perform zap lookups, and the "width" of our dirlocks.
167 168 *
168 169 * A normal dirlock locks a single name. Note that with
169 170 * normalization a name can be composed multiple ways, but
170 171 * when normalized, these names all compare equal. A wide
171 172 * dirlock locks multiple names. We need these when the file
172 173 * system is supporting mixed-mode access. It is sometimes
173 174 * necessary to lock all case permutations of file name at
174 175 * once so that simultaneous case-insensitive/case-sensitive
175 176 * behaves as rationally as possible.
176 177 */
177 178
178 179 /*
179 180 * When matching we may need to normalize & change case according to
180 181 * FS settings.
181 182 *
182 183 * Note that a normalized match is necessary for a case insensitive
183 184 * filesystem when the lookup request is not exact because normalization
184 185 * can fold case independent of normalizing code point sequences.
185 186 *
186 187 * See the table above zfs_dropname().
187 188 */
188 189 if (zfsvfs->z_norm != 0) {
189 190 mt = MT_NORMALIZE;
190 191
191 192 /*
192 193 * Determine if the match needs to honor the case specified in
193 194 * lookup, and if so keep track of that so that during
194 195 * normalization we don't fold case.
195 196 */
196 197 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
197 198 (flag & ZCIEXACT)) ||
198 199 (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) {
199 200 mt |= MT_MATCH_CASE;
200 201 }
201 202 }
202 203
203 204 /*
204 205 * Only look in or update the DNLC if we are looking for the
205 206 * name on a file system that does not require normalization
206 207 * or case folding. We can also look there if we happen to be
207 208 * on a non-normalizing, mixed sensitivity file system IF we
208 209 * are looking for the exact name.
209 210 *
210 211 * Maybe can add TO-UPPERed version of name to dnlc in ci-only
211 212 * case for performance improvement?
212 213 */
213 214 update = !zfsvfs->z_norm ||
214 215 (zfsvfs->z_case == ZFS_CASE_MIXED &&
215 216 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
216 217
217 218 /*
218 219 * ZRENAMING indicates we are in a situation where we should
219 220 * take narrow locks regardless of the file system's
220 221 * preferences for normalizing and case folding. This will
221 222 * prevent us deadlocking trying to grab the same wide lock
222 223 * twice if the two names happen to be case-insensitive
223 224 * matches.
224 225 */
225 226 if (flag & ZRENAMING)
226 227 cmpflags = 0;
227 228 else
228 229 cmpflags = zfsvfs->z_norm;
229 230
230 231 /*
231 232 * Wait until there are no locks on this name.
232 233 *
233 234 * Don't grab the the lock if it is already held. However, cannot
234 235 * have both ZSHARED and ZHAVELOCK together.
235 236 */
236 237 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
237 238 if (!(flag & ZHAVELOCK))
238 239 rw_enter(&dzp->z_name_lock, RW_READER);
239 240
240 241 mutex_enter(&dzp->z_lock);
241 242 for (;;) {
242 243 if (dzp->z_unlinked) {
243 244 mutex_exit(&dzp->z_lock);
244 245 if (!(flag & ZHAVELOCK))
245 246 rw_exit(&dzp->z_name_lock);
246 247 return (SET_ERROR(ENOENT));
247 248 }
248 249 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
249 250 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
250 251 U8_UNICODE_LATEST, &error) == 0) || error != 0)
251 252 break;
252 253 }
253 254 if (error != 0) {
254 255 mutex_exit(&dzp->z_lock);
255 256 if (!(flag & ZHAVELOCK))
256 257 rw_exit(&dzp->z_name_lock);
257 258 return (SET_ERROR(ENOENT));
258 259 }
259 260 if (dl == NULL) {
260 261 /*
261 262 * Allocate a new dirlock and add it to the list.
262 263 */
263 264 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
264 265 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
265 266 dl->dl_name = name;
266 267 dl->dl_sharecnt = 0;
267 268 dl->dl_namelock = 0;
268 269 dl->dl_namesize = 0;
269 270 dl->dl_dzp = dzp;
270 271 dl->dl_next = dzp->z_dirlocks;
271 272 dzp->z_dirlocks = dl;
272 273 break;
273 274 }
274 275 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
275 276 break;
276 277 cv_wait(&dl->dl_cv, &dzp->z_lock);
277 278 }
278 279
279 280 /*
280 281 * If the z_name_lock was NOT held for this dirlock record it.
281 282 */
282 283 if (flag & ZHAVELOCK)
283 284 dl->dl_namelock = 1;
284 285
285 286 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
286 287 /*
287 288 * We're the second shared reference to dl. Make a copy of
288 289 * dl_name in case the first thread goes away before we do.
289 290 * Note that we initialize the new name before storing its
290 291 * pointer into dl_name, because the first thread may load
291 292 * dl->dl_name at any time. It'll either see the old value,
292 293 * which belongs to it, or the new shared copy; either is OK.
293 294 */
294 295 dl->dl_namesize = strlen(dl->dl_name) + 1;
295 296 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
296 297 bcopy(dl->dl_name, name, dl->dl_namesize);
297 298 dl->dl_name = name;
298 299 }
299 300
300 301 mutex_exit(&dzp->z_lock);
301 302
302 303 /*
303 304 * We have a dirlock on the name. (Note that it is the dirlock,
304 305 * not the dzp's z_lock, that protects the name in the zap object.)
305 306 * See if there's an object by this name; if so, put a hold on it.
306 307 */
307 308 if (flag & ZXATTR) {
308 309 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
309 310 sizeof (zoid));
310 311 if (error == 0)
311 312 error = (zoid == 0 ? ENOENT : 0);
312 313 } else {
313 314 if (update)
314 315 vp = dnlc_lookup(ZTOV(dzp), name);
315 316 if (vp == DNLC_NO_VNODE) {
316 317 VN_RELE(vp);
317 318 error = SET_ERROR(ENOENT);
318 319 } else if (vp) {
319 320 if (flag & ZNEW) {
320 321 zfs_dirent_unlock(dl);
321 322 VN_RELE(vp);
322 323 return (SET_ERROR(EEXIST));
323 324 }
324 325 *dlpp = dl;
325 326 *zpp = VTOZ(vp);
326 327 return (0);
327 328 } else {
328 329 error = zfs_match_find(zfsvfs, dzp, name, mt,
329 330 update, direntflags, realpnp, &zoid);
330 331 }
331 332 }
332 333 if (error) {
333 334 if (error != ENOENT || (flag & ZEXISTS)) {
334 335 zfs_dirent_unlock(dl);
335 336 return (error);
336 337 }
337 338 } else {
338 339 if (flag & ZNEW) {
339 340 zfs_dirent_unlock(dl);
340 341 return (SET_ERROR(EEXIST));
341 342 }
342 343 error = zfs_zget(zfsvfs, zoid, zpp);
343 344 if (error) {
344 345 zfs_dirent_unlock(dl);
345 346 return (error);
346 347 }
347 348 if (!(flag & ZXATTR) && update)
348 349 dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
349 350 }
350 351
351 352 *dlpp = dl;
352 353
353 354 return (0);
354 355 }
355 356
356 357 /*
357 358 * Unlock this directory entry and wake anyone who was waiting for it.
358 359 */
359 360 void
360 361 zfs_dirent_unlock(zfs_dirlock_t *dl)
361 362 {
362 363 znode_t *dzp = dl->dl_dzp;
363 364 zfs_dirlock_t **prev_dl, *cur_dl;
364 365
365 366 mutex_enter(&dzp->z_lock);
366 367
367 368 if (!dl->dl_namelock)
368 369 rw_exit(&dzp->z_name_lock);
369 370
370 371 if (dl->dl_sharecnt > 1) {
371 372 dl->dl_sharecnt--;
372 373 mutex_exit(&dzp->z_lock);
373 374 return;
374 375 }
375 376 prev_dl = &dzp->z_dirlocks;
376 377 while ((cur_dl = *prev_dl) != dl)
377 378 prev_dl = &cur_dl->dl_next;
378 379 *prev_dl = dl->dl_next;
379 380 cv_broadcast(&dl->dl_cv);
380 381 mutex_exit(&dzp->z_lock);
381 382
382 383 if (dl->dl_namesize != 0)
383 384 kmem_free(dl->dl_name, dl->dl_namesize);
384 385 cv_destroy(&dl->dl_cv);
385 386 kmem_free(dl, sizeof (*dl));
386 387 }
387 388
388 389 /*
389 390 * Look up an entry in a directory.
390 391 *
391 392 * NOTE: '.' and '..' are handled as special cases because
392 393 * no directory entries are actually stored for them. If this is
393 394 * the root of a filesystem, then '.zfs' is also treated as a
394 395 * special pseudo-directory.
395 396 */
396 397 int
397 398 zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
398 399 int *deflg, pathname_t *rpnp)
399 400 {
400 401 zfs_dirlock_t *dl;
401 402 znode_t *zp;
402 403 int error = 0;
403 404 uint64_t parent;
404 405
405 406 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
406 407 *vpp = ZTOV(dzp);
407 408 VN_HOLD(*vpp);
408 409 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
409 410 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
410 411
411 412 /*
412 413 * If we are a snapshot mounted under .zfs, return
413 414 * the vp for the snapshot directory.
414 415 */
415 416 if ((error = sa_lookup(dzp->z_sa_hdl,
416 417 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
417 418 return (error);
418 419 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
419 420 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
420 421 "snapshot", vpp, NULL, 0, NULL, kcred,
421 422 NULL, NULL, NULL);
422 423 return (error);
423 424 }
424 425 rw_enter(&dzp->z_parent_lock, RW_READER);
425 426 error = zfs_zget(zfsvfs, parent, &zp);
426 427 if (error == 0)
427 428 *vpp = ZTOV(zp);
428 429 rw_exit(&dzp->z_parent_lock);
429 430 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
430 431 *vpp = zfsctl_root(dzp);
431 432 } else {
432 433 int zf;
433 434
434 435 zf = ZEXISTS | ZSHARED;
435 436 if (flags & FIGNORECASE)
436 437 zf |= ZCILOOK;
437 438
438 439 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
439 440 if (error == 0) {
440 441 *vpp = ZTOV(zp);
441 442 zfs_dirent_unlock(dl);
442 443 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
443 444 }
444 445 rpnp = NULL;
445 446 }
446 447
447 448 if ((flags & FIGNORECASE) && rpnp && !error)
448 449 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
449 450
450 451 return (error);
451 452 }
452 453
453 454 /*
454 455 * unlinked Set (formerly known as the "delete queue") Error Handling
455 456 *
456 457 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
457 458 * don't specify the name of the entry that we will be manipulating. We
458 459 * also fib and say that we won't be adding any new entries to the
459 460 * unlinked set, even though we might (this is to lower the minimum file
460 461 * size that can be deleted in a full filesystem). So on the small
461 462 * chance that the nlink list is using a fat zap (ie. has more than
462 463 * 2000 entries), we *may* not pre-read a block that's needed.
463 464 * Therefore it is remotely possible for some of the assertions
464 465 * regarding the unlinked set below to fail due to i/o error. On a
465 466 * nondebug system, this will result in the space being leaked.
466 467 */
467 468 void
468 469 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
469 470 {
470 471 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
471 472
472 473 ASSERT(zp->z_unlinked);
|
↓ open down ↓ |
408 lines elided |
↑ open up ↑ |
473 474 ASSERT(zp->z_links == 0);
474 475
475 476 VERIFY3U(0, ==,
476 477 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
477 478 }
478 479
479 480 /*
480 481 * Clean up any znodes that had no links when we either crashed or
481 482 * (force) umounted the file system.
482 483 */
483 -void
484 -zfs_unlinked_drain(zfsvfs_t *zfsvfs)
484 +static void
485 +zfs_unlinked_drain_impl(zfsvfs_t *zfsvfs)
485 486 {
486 487 zap_cursor_t zc;
487 488 zap_attribute_t zap;
488 489 dmu_object_info_t doi;
489 490 znode_t *zp;
490 491 int error;
491 492
493 + ASSERT(zfsvfs->z_drain_state != ZFS_DRAIN_SHUTDOWN);
492 494 /*
493 495 * Interate over the contents of the unlinked set.
494 496 */
495 497 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
496 - zap_cursor_retrieve(&zc, &zap) == 0;
498 + zap_cursor_retrieve(&zc, &zap) == 0 &&
499 + /* Only checking for a shutdown request, so no locking reqd. */
500 + zfsvfs->z_drain_state == ZFS_DRAIN_RUNNING;
497 501 zap_cursor_advance(&zc)) {
498 502
499 503 /*
500 504 * See what kind of object we have in list
501 505 */
502 506
503 507 error = dmu_object_info(zfsvfs->z_os,
504 508 zap.za_first_integer, &doi);
505 509 if (error != 0)
506 510 continue;
507 511
508 512 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
509 513 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
510 514 /*
511 515 * We need to re-mark these list entries for deletion,
512 516 * so we pull them back into core and set zp->z_unlinked.
513 517 */
514 518 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
515 519
516 520 /*
|
↓ open down ↓ |
10 lines elided |
↑ open up ↑ |
517 521 * We may pick up znodes that are already marked for deletion.
518 522 * This could happen during the purge of an extended attribute
519 523 * directory. All we need to do is skip over them, since they
520 524 * are already in the system marked z_unlinked.
521 525 */
522 526 if (error != 0)
523 527 continue;
524 528
525 529 zp->z_unlinked = B_TRUE;
526 530 VN_RELE(ZTOV(zp));
531 +
532 + ASSERT(!zfsvfs->z_unmounted);
527 533 }
528 534 zap_cursor_fini(&zc);
535 +
536 + mutex_enter(&zfsvfs->z_drain_lock);
537 + zfsvfs->z_drain_state = ZFS_DRAIN_SHUTDOWN;
538 + cv_broadcast(&zfsvfs->z_drain_cv);
539 + mutex_exit(&zfsvfs->z_drain_lock);
529 540 }
530 541
542 +/*
543 + * Setup required hold. After that tries to dispatch
544 + * async unlinked drain logic. Otherwise executes
545 + * the logic synchronously.
546 + */
547 +void
548 +zfs_unlinked_drain(zfsvfs_t *zfsvfs)
549 +{
550 + ASSERT(!zfsvfs->z_unmounted);
551 +
552 + mutex_enter(&zfsvfs->z_drain_lock);
553 + ASSERT(zfsvfs->z_drain_state == ZFS_DRAIN_SHUTDOWN);
554 + zfsvfs->z_drain_state = ZFS_DRAIN_RUNNING;
555 + mutex_exit(&zfsvfs->z_drain_lock);
556 +
557 + if (taskq_dispatch(dsl_pool_vnrele_taskq(
558 + spa_get_dsl(zfsvfs->z_os->os_spa)),
559 + (void (*)(void *))zfs_unlinked_drain_impl, zfsvfs,
560 + TQ_NOSLEEP) == NULL) {
561 + cmn_err(CE_WARN, "async zfs_unlinked_drain dispatch failed");
562 + zfs_unlinked_drain_impl(zfsvfs);
563 + }
564 +}
565 +
566 +/*
567 + * Stops an asynchronous zfs_unlinked_drain. This must be called prior to
568 + * destroying the zfsvfs_t, as the drain doesn't hold the z_teardown_lock.
569 + */
570 +void
571 +zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs)
572 +{
573 + ASSERT(!zfsvfs->z_unmounted);
574 +
575 + mutex_enter(&zfsvfs->z_drain_lock);
576 + while (zfsvfs->z_drain_state != ZFS_DRAIN_SHUTDOWN) {
577 + zfsvfs->z_drain_state = ZFS_DRAIN_SHUTDOWN_REQ;
578 + cv_wait(&zfsvfs->z_drain_cv, &zfsvfs->z_drain_lock);
579 + }
580 + mutex_exit(&zfsvfs->z_drain_lock);
581 +}
582 +
531 583 /*
532 584 * Delete the entire contents of a directory. Return a count
533 585 * of the number of entries that could not be deleted. If we encounter
534 586 * an error, return a count of at least one so that the directory stays
535 587 * in the unlinked set.
536 588 *
537 589 * NOTE: this function assumes that the directory is inactive,
538 590 * so there is no need to lock its entries before deletion.
539 591 * Also, it assumes the directory contents is *only* regular
540 592 * files.
541 593 */
542 594 static int
543 595 zfs_purgedir(znode_t *dzp)
544 596 {
545 597 zap_cursor_t zc;
546 598 zap_attribute_t zap;
547 599 znode_t *xzp;
548 600 dmu_tx_t *tx;
549 601 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
550 602 zfs_dirlock_t dl;
551 603 int skipped = 0;
552 604 int error;
553 605
554 606 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
555 607 (error = zap_cursor_retrieve(&zc, &zap)) == 0;
556 608 zap_cursor_advance(&zc)) {
557 609 error = zfs_zget(zfsvfs,
558 610 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
559 611 if (error) {
560 612 skipped += 1;
561 613 continue;
562 614 }
563 615
564 616 ASSERT((ZTOV(xzp)->v_type == VREG) ||
565 617 (ZTOV(xzp)->v_type == VLNK));
566 618
567 619 tx = dmu_tx_create(zfsvfs->z_os);
568 620 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
569 621 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
570 622 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
571 623 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
572 624 /* Is this really needed ? */
573 625 zfs_sa_upgrade_txholds(tx, xzp);
574 626 dmu_tx_mark_netfree(tx);
575 627 error = dmu_tx_assign(tx, TXG_WAIT);
576 628 if (error) {
577 629 dmu_tx_abort(tx);
578 630 VN_RELE(ZTOV(xzp));
579 631 skipped += 1;
580 632 continue;
581 633 }
582 634 bzero(&dl, sizeof (dl));
583 635 dl.dl_dzp = dzp;
584 636 dl.dl_name = zap.za_name;
585 637
586 638 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
587 639 if (error)
588 640 skipped += 1;
589 641 dmu_tx_commit(tx);
590 642
591 643 VN_RELE(ZTOV(xzp));
592 644 }
593 645 zap_cursor_fini(&zc);
594 646 if (error != ENOENT)
595 647 skipped += 1;
596 648 return (skipped);
597 649 }
598 650
599 651 void
600 652 zfs_rmnode(znode_t *zp)
601 653 {
602 654 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
603 655 objset_t *os = zfsvfs->z_os;
604 656 znode_t *xzp = NULL;
605 657 dmu_tx_t *tx;
606 658 uint64_t acl_obj;
607 659 uint64_t xattr_obj;
608 660 int error;
609 661
610 662 ASSERT(zp->z_links == 0);
611 663 ASSERT(ZTOV(zp)->v_count == 0);
612 664
613 665 /*
614 666 * If this is an attribute directory, purge its contents.
615 667 */
616 668 if (ZTOV(zp)->v_type == VDIR && (zp->z_pflags & ZFS_XATTR)) {
617 669 if (zfs_purgedir(zp) != 0) {
618 670 /*
619 671 * Not enough space to delete some xattrs.
620 672 * Leave it in the unlinked set.
621 673 */
622 674 zfs_znode_dmu_fini(zp);
623 675 zfs_znode_free(zp);
624 676 return;
625 677 }
626 678 } else {
627 679 /*
628 680 * Free up all the data in the file. We don't do this for
629 681 * XATTR directories because we need truncate and remove to be
630 682 * in the same tx, like in zfs_znode_delete(). Otherwise, if
631 683 * we crash here we'll end up with an inconsistent truncated
632 684 * zap object in the delete queue. Note a truncated file is
633 685 * harmless since it only contains user data.
634 686 */
635 687 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
636 688 if (error) {
637 689 /*
638 690 * Not enough space or we were interrupted by unmount.
639 691 * Leave the file in the unlinked set.
640 692 */
641 693 zfs_znode_dmu_fini(zp);
642 694 zfs_znode_free(zp);
643 695 return;
644 696 }
645 697 }
646 698
647 699 /*
648 700 * If the file has extended attributes, we're going to unlink
649 701 * the xattr dir.
650 702 */
651 703 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
652 704 &xattr_obj, sizeof (xattr_obj));
653 705 if (error == 0 && xattr_obj) {
654 706 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
655 707 ASSERT(error == 0);
656 708 }
657 709
658 710 acl_obj = zfs_external_acl(zp);
659 711
660 712 /*
661 713 * Set up the final transaction.
662 714 */
663 715 tx = dmu_tx_create(os);
664 716 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
665 717 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
666 718 if (xzp) {
667 719 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
668 720 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
669 721 }
670 722 if (acl_obj)
671 723 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
672 724
673 725 zfs_sa_upgrade_txholds(tx, zp);
674 726 error = dmu_tx_assign(tx, TXG_WAIT);
675 727 if (error) {
676 728 /*
677 729 * Not enough space to delete the file. Leave it in the
678 730 * unlinked set, leaking it until the fs is remounted (at
679 731 * which point we'll call zfs_unlinked_drain() to process it).
680 732 */
681 733 dmu_tx_abort(tx);
682 734 zfs_znode_dmu_fini(zp);
683 735 zfs_znode_free(zp);
684 736 goto out;
685 737 }
686 738
687 739 if (xzp) {
688 740 ASSERT(error == 0);
689 741 mutex_enter(&xzp->z_lock);
690 742 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
691 743 xzp->z_links = 0; /* no more links to it */
692 744 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
693 745 &xzp->z_links, sizeof (xzp->z_links), tx));
694 746 mutex_exit(&xzp->z_lock);
695 747 zfs_unlinked_add(xzp, tx);
696 748 }
697 749
698 750 /* Remove this znode from the unlinked set */
699 751 VERIFY3U(0, ==,
700 752 zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
701 753
702 754 zfs_znode_delete(zp, tx);
703 755
704 756 dmu_tx_commit(tx);
705 757 out:
706 758 if (xzp)
707 759 VN_RELE(ZTOV(xzp));
708 760 }
709 761
710 762 static uint64_t
711 763 zfs_dirent(znode_t *zp, uint64_t mode)
712 764 {
713 765 uint64_t de = zp->z_id;
714 766
715 767 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
716 768 de |= IFTODT(mode) << 60;
717 769 return (de);
718 770 }
719 771
720 772 /*
721 773 * Link zp into dl. Can only fail if zp has been unlinked.
722 774 */
723 775 int
724 776 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
725 777 {
726 778 znode_t *dzp = dl->dl_dzp;
727 779 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
728 780 vnode_t *vp = ZTOV(zp);
729 781 uint64_t value;
730 782 int zp_is_dir = (vp->v_type == VDIR);
731 783 sa_bulk_attr_t bulk[5];
732 784 uint64_t mtime[2], ctime[2];
733 785 int count = 0;
734 786 int error;
735 787
736 788 mutex_enter(&zp->z_lock);
737 789
738 790 if (!(flag & ZRENAMING)) {
739 791 if (zp->z_unlinked) { /* no new links to unlinked zp */
740 792 ASSERT(!(flag & (ZNEW | ZEXISTS)));
741 793 mutex_exit(&zp->z_lock);
742 794 return (SET_ERROR(ENOENT));
743 795 }
744 796 zp->z_links++;
745 797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
746 798 &zp->z_links, sizeof (zp->z_links));
747 799
748 800 }
749 801 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
750 802 &dzp->z_id, sizeof (dzp->z_id));
751 803 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
752 804 &zp->z_pflags, sizeof (zp->z_pflags));
753 805
754 806 if (!(flag & ZNEW)) {
755 807 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
756 808 ctime, sizeof (ctime));
757 809 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
758 810 ctime, B_TRUE);
759 811 }
760 812 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
761 813 ASSERT(error == 0);
762 814
763 815 mutex_exit(&zp->z_lock);
764 816
765 817 mutex_enter(&dzp->z_lock);
766 818 dzp->z_size++;
767 819 dzp->z_links += zp_is_dir;
768 820 count = 0;
769 821 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
770 822 &dzp->z_size, sizeof (dzp->z_size));
771 823 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
772 824 &dzp->z_links, sizeof (dzp->z_links));
773 825 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
774 826 mtime, sizeof (mtime));
775 827 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
776 828 ctime, sizeof (ctime));
777 829 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
778 830 &dzp->z_pflags, sizeof (dzp->z_pflags));
779 831 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
780 832 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
781 833 ASSERT(error == 0);
782 834 mutex_exit(&dzp->z_lock);
783 835
784 836 value = zfs_dirent(zp, zp->z_mode);
785 837 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
786 838 8, 1, &value, tx);
787 839 ASSERT(error == 0);
788 840
789 841 dnlc_update(ZTOV(dzp), dl->dl_name, vp);
790 842
791 843 return (0);
792 844 }
793 845
794 846 /*
795 847 * The match type in the code for this function should conform to:
796 848 *
797 849 * ------------------------------------------------------------------------
798 850 * fs type | z_norm | lookup type | match type
799 851 * ---------|-------------|-------------|----------------------------------
800 852 * CS !norm | 0 | 0 | 0 (exact)
801 853 * CS norm | formX | 0 | MT_NORMALIZE
802 854 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE
803 855 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
804 856 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE
805 857 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
806 858 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
807 859 * CM !norm | upper | ZCILOOK | MT_NORMALIZE
808 860 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
809 861 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE
810 862 *
811 863 * Abbreviations:
812 864 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
813 865 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
814 866 * formX = unicode normalization form set on fs creation
815 867 */
816 868 static int
817 869 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
818 870 int flag)
819 871 {
820 872 int error;
821 873
822 874 if (zp->z_zfsvfs->z_norm) {
823 875 matchtype_t mt = MT_NORMALIZE;
824 876
825 877 if ((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
826 878 (flag & ZCIEXACT)) ||
827 879 (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED &&
828 880 !(flag & ZCILOOK))) {
829 881 mt |= MT_MATCH_CASE;
830 882 }
831 883
832 884 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id,
833 885 dl->dl_name, mt, tx);
834 886 } else {
835 887 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
836 888 tx);
837 889 }
838 890
839 891 return (error);
840 892 }
841 893
842 894 /*
843 895 * Unlink zp from dl, and mark zp for deletion if this was the last link.
844 896 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
845 897 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
846 898 * If it's non-NULL, we use it to indicate whether the znode needs deletion,
847 899 * and it's the caller's job to do it.
848 900 */
849 901 int
850 902 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
851 903 boolean_t *unlinkedp)
852 904 {
853 905 znode_t *dzp = dl->dl_dzp;
854 906 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
855 907 vnode_t *vp = ZTOV(zp);
856 908 int zp_is_dir = (vp->v_type == VDIR);
857 909 boolean_t unlinked = B_FALSE;
858 910 sa_bulk_attr_t bulk[5];
859 911 uint64_t mtime[2], ctime[2];
860 912 int count = 0;
861 913 int error;
862 914
863 915 dnlc_remove(ZTOV(dzp), dl->dl_name);
864 916
865 917 if (!(flag & ZRENAMING)) {
866 918 if (vn_vfswlock(vp)) /* prevent new mounts on zp */
867 919 return (SET_ERROR(EBUSY));
868 920
869 921 if (vn_ismntpt(vp)) { /* don't remove mount point */
870 922 vn_vfsunlock(vp);
871 923 return (SET_ERROR(EBUSY));
872 924 }
873 925
874 926 mutex_enter(&zp->z_lock);
875 927
876 928 if (zp_is_dir && !zfs_dirempty(zp)) {
877 929 mutex_exit(&zp->z_lock);
878 930 vn_vfsunlock(vp);
879 931 return (SET_ERROR(EEXIST));
880 932 }
881 933
882 934 /*
883 935 * If we get here, we are going to try to remove the object.
884 936 * First try removing the name from the directory; if that
885 937 * fails, return the error.
886 938 */
887 939 error = zfs_dropname(dl, zp, dzp, tx, flag);
888 940 if (error != 0) {
889 941 mutex_exit(&zp->z_lock);
890 942 vn_vfsunlock(vp);
891 943 return (error);
892 944 }
893 945
894 946 if (zp->z_links <= zp_is_dir) {
895 947 zfs_panic_recover("zfs: link count on %s is %u, "
896 948 "should be at least %u",
897 949 zp->z_vnode->v_path != vn_vpath_empty ?
898 950 zp->z_vnode->v_path : "<unknown>",
899 951 (int)zp->z_links, zp_is_dir + 1);
900 952 zp->z_links = zp_is_dir + 1;
901 953 }
902 954 if (--zp->z_links == zp_is_dir) {
903 955 zp->z_unlinked = B_TRUE;
904 956 zp->z_links = 0;
905 957 unlinked = B_TRUE;
906 958 } else {
907 959 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
908 960 NULL, &ctime, sizeof (ctime));
909 961 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
910 962 NULL, &zp->z_pflags, sizeof (zp->z_pflags));
911 963 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
912 964 B_TRUE);
913 965 }
914 966 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
915 967 NULL, &zp->z_links, sizeof (zp->z_links));
916 968 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
917 969 count = 0;
918 970 ASSERT(error == 0);
919 971 mutex_exit(&zp->z_lock);
920 972 vn_vfsunlock(vp);
921 973 } else {
922 974 error = zfs_dropname(dl, zp, dzp, tx, flag);
923 975 if (error != 0)
924 976 return (error);
925 977 }
926 978
927 979 mutex_enter(&dzp->z_lock);
928 980 dzp->z_size--; /* one dirent removed */
929 981 dzp->z_links -= zp_is_dir; /* ".." link from zp */
930 982 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
931 983 NULL, &dzp->z_links, sizeof (dzp->z_links));
932 984 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
933 985 NULL, &dzp->z_size, sizeof (dzp->z_size));
934 986 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
935 987 NULL, ctime, sizeof (ctime));
936 988 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
937 989 NULL, mtime, sizeof (mtime));
938 990 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
939 991 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
940 992 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
941 993 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
942 994 ASSERT(error == 0);
943 995 mutex_exit(&dzp->z_lock);
944 996
945 997 if (unlinkedp != NULL)
946 998 *unlinkedp = unlinked;
947 999 else if (unlinked)
948 1000 zfs_unlinked_add(zp, tx);
949 1001
950 1002 return (0);
951 1003 }
952 1004
953 1005 /*
954 1006 * Indicate whether the directory is empty. Works with or without z_lock
955 1007 * held, but can only be consider a hint in the latter case. Returns true
956 1008 * if only "." and ".." remain and there's no work in progress.
957 1009 */
958 1010 boolean_t
959 1011 zfs_dirempty(znode_t *dzp)
960 1012 {
961 1013 return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
962 1014 }
963 1015
964 1016 int
965 1017 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
966 1018 {
967 1019 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
968 1020 znode_t *xzp;
969 1021 dmu_tx_t *tx;
970 1022 int error;
971 1023 zfs_acl_ids_t acl_ids;
972 1024 boolean_t fuid_dirtied;
973 1025 uint64_t parent;
974 1026
975 1027 *xvpp = NULL;
976 1028
977 1029 if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
978 1030 return (error);
979 1031
980 1032 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
981 1033 &acl_ids)) != 0)
982 1034 return (error);
983 1035 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
984 1036 zfs_acl_ids_free(&acl_ids);
985 1037 return (SET_ERROR(EDQUOT));
986 1038 }
987 1039
988 1040 tx = dmu_tx_create(zfsvfs->z_os);
989 1041 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
990 1042 ZFS_SA_BASE_ATTR_SIZE);
991 1043 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
992 1044 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
993 1045 fuid_dirtied = zfsvfs->z_fuid_dirty;
994 1046 if (fuid_dirtied)
995 1047 zfs_fuid_txhold(zfsvfs, tx);
996 1048 error = dmu_tx_assign(tx, TXG_WAIT);
997 1049 if (error) {
998 1050 zfs_acl_ids_free(&acl_ids);
999 1051 dmu_tx_abort(tx);
1000 1052 return (error);
1001 1053 }
1002 1054 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1003 1055
1004 1056 if (fuid_dirtied)
1005 1057 zfs_fuid_sync(zfsvfs, tx);
1006 1058
1007 1059 #ifdef DEBUG
1008 1060 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1009 1061 &parent, sizeof (parent));
1010 1062 ASSERT(error == 0 && parent == zp->z_id);
1011 1063 #endif
1012 1064
1013 1065 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
1014 1066 sizeof (xzp->z_id), tx));
1015 1067
1016 1068 (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
1017 1069 xzp, "", NULL, acl_ids.z_fuidp, vap);
1018 1070
1019 1071 zfs_acl_ids_free(&acl_ids);
1020 1072 dmu_tx_commit(tx);
1021 1073
1022 1074 *xvpp = ZTOV(xzp);
1023 1075
1024 1076 return (0);
1025 1077 }
1026 1078
1027 1079 /*
1028 1080 * Return a znode for the extended attribute directory for zp.
1029 1081 * ** If the directory does not already exist, it is created **
1030 1082 *
1031 1083 * IN: zp - znode to obtain attribute directory from
1032 1084 * cr - credentials of caller
1033 1085 * flags - flags from the VOP_LOOKUP call
1034 1086 *
1035 1087 * OUT: xzpp - pointer to extended attribute znode
1036 1088 *
1037 1089 * RETURN: 0 on success
1038 1090 * error number on failure
1039 1091 */
1040 1092 int
1041 1093 zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
1042 1094 {
1043 1095 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1044 1096 znode_t *xzp;
1045 1097 zfs_dirlock_t *dl;
1046 1098 vattr_t va;
1047 1099 int error;
1048 1100 top:
1049 1101 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1050 1102 if (error)
1051 1103 return (error);
1052 1104
1053 1105 if (xzp != NULL) {
1054 1106 *xvpp = ZTOV(xzp);
1055 1107 zfs_dirent_unlock(dl);
1056 1108 return (0);
1057 1109 }
1058 1110
1059 1111
1060 1112 if (!(flags & CREATE_XATTR_DIR)) {
1061 1113 zfs_dirent_unlock(dl);
1062 1114 return (SET_ERROR(ENOENT));
1063 1115 }
1064 1116
1065 1117 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1066 1118 zfs_dirent_unlock(dl);
1067 1119 return (SET_ERROR(EROFS));
1068 1120 }
1069 1121
1070 1122 /*
1071 1123 * The ability to 'create' files in an attribute
1072 1124 * directory comes from the write_xattr permission on the base file.
1073 1125 *
1074 1126 * The ability to 'search' an attribute directory requires
1075 1127 * read_xattr permission on the base file.
1076 1128 *
1077 1129 * Once in a directory the ability to read/write attributes
1078 1130 * is controlled by the permissions on the attribute file.
1079 1131 */
1080 1132 va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1081 1133 va.va_type = VDIR;
1082 1134 va.va_mode = S_IFDIR | S_ISVTX | 0777;
1083 1135 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1084 1136
1085 1137 error = zfs_make_xattrdir(zp, &va, xvpp, cr);
1086 1138 zfs_dirent_unlock(dl);
1087 1139
1088 1140 if (error == ERESTART) {
1089 1141 /* NB: we already did dmu_tx_wait() if necessary */
1090 1142 goto top;
1091 1143 }
1092 1144
1093 1145 return (error);
1094 1146 }
1095 1147
1096 1148 /*
1097 1149 * Decide whether it is okay to remove within a sticky directory.
1098 1150 *
1099 1151 * In sticky directories, write access is not sufficient;
1100 1152 * you can remove entries from a directory only if:
1101 1153 *
1102 1154 * you own the directory,
1103 1155 * you own the entry,
1104 1156 * the entry is a plain file and you have write access,
1105 1157 * or you are privileged (checked in secpolicy...).
1106 1158 *
1107 1159 * The function returns 0 if remove access is granted.
1108 1160 */
1109 1161 int
1110 1162 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1111 1163 {
1112 1164 uid_t uid;
1113 1165 uid_t downer;
1114 1166 uid_t fowner;
1115 1167 zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1116 1168
1117 1169 if (zdp->z_zfsvfs->z_replay)
1118 1170 return (0);
1119 1171
1120 1172 if ((zdp->z_mode & S_ISVTX) == 0)
1121 1173 return (0);
1122 1174
1123 1175 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
1124 1176 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
1125 1177
1126 1178 if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1127 1179 (ZTOV(zp)->v_type == VREG &&
1128 1180 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1129 1181 return (0);
1130 1182 else
1131 1183 return (secpolicy_vnode_remove(cr));
1132 1184 }
|
↓ open down ↓ |
592 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX