xfs_trans_buf.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785
  1. /*
  2. * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_inum.h"
  24. #include "xfs_trans.h"
  25. #include "xfs_sb.h"
  26. #include "xfs_ag.h"
  27. #include "xfs_mount.h"
  28. #include "xfs_bmap_btree.h"
  29. #include "xfs_alloc_btree.h"
  30. #include "xfs_ialloc_btree.h"
  31. #include "xfs_dinode.h"
  32. #include "xfs_inode.h"
  33. #include "xfs_buf_item.h"
  34. #include "xfs_trans_priv.h"
  35. #include "xfs_error.h"
  36. #include "xfs_rw.h"
  37. #include "xfs_trace.h"
  38. /*
  39. * Check to see if a buffer matching the given parameters is already
  40. * a part of the given transaction.
  41. */
  42. STATIC struct xfs_buf *
  43. xfs_trans_buf_item_match(
  44. struct xfs_trans *tp,
  45. struct xfs_buftarg *target,
  46. xfs_daddr_t blkno,
  47. int len)
  48. {
  49. struct xfs_log_item_desc *lidp;
  50. struct xfs_buf_log_item *blip;
  51. len = BBTOB(len);
  52. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  53. blip = (struct xfs_buf_log_item *)lidp->lid_item;
  54. if (blip->bli_item.li_type == XFS_LI_BUF &&
  55. blip->bli_buf->b_target == target &&
  56. XFS_BUF_ADDR(blip->bli_buf) == blkno &&
  57. BBTOB(blip->bli_buf->b_length) == len)
  58. return blip->bli_buf;
  59. }
  60. return NULL;
  61. }
  62. /*
  63. * Add the locked buffer to the transaction.
  64. *
  65. * The buffer must be locked, and it cannot be associated with any
  66. * transaction.
  67. *
  68. * If the buffer does not yet have a buf log item associated with it,
  69. * then allocate one for it. Then add the buf item to the transaction.
  70. */
  71. STATIC void
  72. _xfs_trans_bjoin(
  73. struct xfs_trans *tp,
  74. struct xfs_buf *bp,
  75. int reset_recur)
  76. {
  77. struct xfs_buf_log_item *bip;
  78. ASSERT(bp->b_transp == NULL);
  79. /*
  80. * The xfs_buf_log_item pointer is stored in b_fsprivate. If
  81. * it doesn't have one yet, then allocate one and initialize it.
  82. * The checks to see if one is there are in xfs_buf_item_init().
  83. */
  84. xfs_buf_item_init(bp, tp->t_mountp);
  85. bip = bp->b_fspriv;
  86. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  87. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  88. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  89. if (reset_recur)
  90. bip->bli_recur = 0;
  91. /*
  92. * Take a reference for this transaction on the buf item.
  93. */
  94. atomic_inc(&bip->bli_refcount);
  95. /*
  96. * Get a log_item_desc to point at the new item.
  97. */
  98. xfs_trans_add_item(tp, &bip->bli_item);
  99. /*
  100. * Initialize b_fsprivate2 so we can find it with incore_match()
  101. * in xfs_trans_get_buf() and friends above.
  102. */
  103. bp->b_transp = tp;
  104. }
  105. void
  106. xfs_trans_bjoin(
  107. struct xfs_trans *tp,
  108. struct xfs_buf *bp)
  109. {
  110. _xfs_trans_bjoin(tp, bp, 0);
  111. trace_xfs_trans_bjoin(bp->b_fspriv);
  112. }
  113. /*
  114. * Get and lock the buffer for the caller if it is not already
  115. * locked within the given transaction. If it is already locked
  116. * within the transaction, just increment its lock recursion count
  117. * and return a pointer to it.
  118. *
  119. * If the transaction pointer is NULL, make this just a normal
  120. * get_buf() call.
  121. */
  122. xfs_buf_t *
  123. xfs_trans_get_buf(xfs_trans_t *tp,
  124. xfs_buftarg_t *target_dev,
  125. xfs_daddr_t blkno,
  126. int len,
  127. uint flags)
  128. {
  129. xfs_buf_t *bp;
  130. xfs_buf_log_item_t *bip;
  131. if (flags == 0)
  132. flags = XBF_MAPPED;
  133. /*
  134. * Default to a normal get_buf() call if the tp is NULL.
  135. */
  136. if (tp == NULL)
  137. return xfs_buf_get(target_dev, blkno, len, flags);
  138. /*
  139. * If we find the buffer in the cache with this transaction
  140. * pointer in its b_fsprivate2 field, then we know we already
  141. * have it locked. In this case we just increment the lock
  142. * recursion count and return the buffer to the caller.
  143. */
  144. bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
  145. if (bp != NULL) {
  146. ASSERT(xfs_buf_islocked(bp));
  147. if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
  148. xfs_buf_stale(bp);
  149. XFS_BUF_DONE(bp);
  150. }
  151. ASSERT(bp->b_transp == tp);
  152. bip = bp->b_fspriv;
  153. ASSERT(bip != NULL);
  154. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  155. bip->bli_recur++;
  156. trace_xfs_trans_get_buf_recur(bip);
  157. return (bp);
  158. }
  159. bp = xfs_buf_get(target_dev, blkno, len, flags);
  160. if (bp == NULL) {
  161. return NULL;
  162. }
  163. ASSERT(!bp->b_error);
  164. _xfs_trans_bjoin(tp, bp, 1);
  165. trace_xfs_trans_get_buf(bp->b_fspriv);
  166. return (bp);
  167. }
  168. /*
  169. * Get and lock the superblock buffer of this file system for the
  170. * given transaction.
  171. *
  172. * We don't need to use incore_match() here, because the superblock
  173. * buffer is a private buffer which we keep a pointer to in the
  174. * mount structure.
  175. */
  176. xfs_buf_t *
  177. xfs_trans_getsb(xfs_trans_t *tp,
  178. struct xfs_mount *mp,
  179. int flags)
  180. {
  181. xfs_buf_t *bp;
  182. xfs_buf_log_item_t *bip;
  183. /*
  184. * Default to just trying to lock the superblock buffer
  185. * if tp is NULL.
  186. */
  187. if (tp == NULL) {
  188. return (xfs_getsb(mp, flags));
  189. }
  190. /*
  191. * If the superblock buffer already has this transaction
  192. * pointer in its b_fsprivate2 field, then we know we already
  193. * have it locked. In this case we just increment the lock
  194. * recursion count and return the buffer to the caller.
  195. */
  196. bp = mp->m_sb_bp;
  197. if (bp->b_transp == tp) {
  198. bip = bp->b_fspriv;
  199. ASSERT(bip != NULL);
  200. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  201. bip->bli_recur++;
  202. trace_xfs_trans_getsb_recur(bip);
  203. return (bp);
  204. }
  205. bp = xfs_getsb(mp, flags);
  206. if (bp == NULL)
  207. return NULL;
  208. _xfs_trans_bjoin(tp, bp, 1);
  209. trace_xfs_trans_getsb(bp->b_fspriv);
  210. return (bp);
  211. }
  212. #ifdef DEBUG
  213. xfs_buftarg_t *xfs_error_target;
  214. int xfs_do_error;
  215. int xfs_req_num;
  216. int xfs_error_mod = 33;
  217. #endif
  218. /*
  219. * Get and lock the buffer for the caller if it is not already
  220. * locked within the given transaction. If it has not yet been
  221. * read in, read it from disk. If it is already locked
  222. * within the transaction and already read in, just increment its
  223. * lock recursion count and return a pointer to it.
  224. *
  225. * If the transaction pointer is NULL, make this just a normal
  226. * read_buf() call.
  227. */
  228. int
  229. xfs_trans_read_buf(
  230. xfs_mount_t *mp,
  231. xfs_trans_t *tp,
  232. xfs_buftarg_t *target,
  233. xfs_daddr_t blkno,
  234. int len,
  235. uint flags,
  236. xfs_buf_t **bpp)
  237. {
  238. xfs_buf_t *bp;
  239. xfs_buf_log_item_t *bip;
  240. int error;
  241. *bpp = NULL;
  242. if (flags == 0)
  243. flags = XBF_MAPPED;
  244. /*
  245. * Default to a normal get_buf() call if the tp is NULL.
  246. */
  247. if (tp == NULL) {
  248. bp = xfs_buf_read(target, blkno, len, flags);
  249. if (!bp)
  250. return (flags & XBF_TRYLOCK) ?
  251. EAGAIN : XFS_ERROR(ENOMEM);
  252. if (bp->b_error) {
  253. error = bp->b_error;
  254. xfs_buf_ioerror_alert(bp, __func__);
  255. XFS_BUF_UNDONE(bp);
  256. xfs_buf_stale(bp);
  257. xfs_buf_relse(bp);
  258. return error;
  259. }
  260. #ifdef DEBUG
  261. if (xfs_do_error) {
  262. if (xfs_error_target == target) {
  263. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  264. xfs_buf_relse(bp);
  265. xfs_debug(mp, "Returning error!");
  266. return XFS_ERROR(EIO);
  267. }
  268. }
  269. }
  270. #endif
  271. if (XFS_FORCED_SHUTDOWN(mp))
  272. goto shutdown_abort;
  273. *bpp = bp;
  274. return 0;
  275. }
  276. /*
  277. * If we find the buffer in the cache with this transaction
  278. * pointer in its b_fsprivate2 field, then we know we already
  279. * have it locked. If it is already read in we just increment
  280. * the lock recursion count and return the buffer to the caller.
  281. * If the buffer is not yet read in, then we read it in, increment
  282. * the lock recursion count, and return it to the caller.
  283. */
  284. bp = xfs_trans_buf_item_match(tp, target, blkno, len);
  285. if (bp != NULL) {
  286. ASSERT(xfs_buf_islocked(bp));
  287. ASSERT(bp->b_transp == tp);
  288. ASSERT(bp->b_fspriv != NULL);
  289. ASSERT(!bp->b_error);
  290. if (!(XFS_BUF_ISDONE(bp))) {
  291. trace_xfs_trans_read_buf_io(bp, _RET_IP_);
  292. ASSERT(!XFS_BUF_ISASYNC(bp));
  293. XFS_BUF_READ(bp);
  294. xfsbdstrat(tp->t_mountp, bp);
  295. error = xfs_buf_iowait(bp);
  296. if (error) {
  297. xfs_buf_ioerror_alert(bp, __func__);
  298. xfs_buf_relse(bp);
  299. /*
  300. * We can gracefully recover from most read
  301. * errors. Ones we can't are those that happen
  302. * after the transaction's already dirty.
  303. */
  304. if (tp->t_flags & XFS_TRANS_DIRTY)
  305. xfs_force_shutdown(tp->t_mountp,
  306. SHUTDOWN_META_IO_ERROR);
  307. return error;
  308. }
  309. }
  310. /*
  311. * We never locked this buf ourselves, so we shouldn't
  312. * brelse it either. Just get out.
  313. */
  314. if (XFS_FORCED_SHUTDOWN(mp)) {
  315. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  316. *bpp = NULL;
  317. return XFS_ERROR(EIO);
  318. }
  319. bip = bp->b_fspriv;
  320. bip->bli_recur++;
  321. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  322. trace_xfs_trans_read_buf_recur(bip);
  323. *bpp = bp;
  324. return 0;
  325. }
  326. bp = xfs_buf_read(target, blkno, len, flags);
  327. if (bp == NULL) {
  328. *bpp = NULL;
  329. return (flags & XBF_TRYLOCK) ?
  330. 0 : XFS_ERROR(ENOMEM);
  331. }
  332. if (bp->b_error) {
  333. error = bp->b_error;
  334. xfs_buf_stale(bp);
  335. XFS_BUF_DONE(bp);
  336. xfs_buf_ioerror_alert(bp, __func__);
  337. if (tp->t_flags & XFS_TRANS_DIRTY)
  338. xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
  339. xfs_buf_relse(bp);
  340. return error;
  341. }
  342. #ifdef DEBUG
  343. if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
  344. if (xfs_error_target == target) {
  345. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  346. xfs_force_shutdown(tp->t_mountp,
  347. SHUTDOWN_META_IO_ERROR);
  348. xfs_buf_relse(bp);
  349. xfs_debug(mp, "Returning trans error!");
  350. return XFS_ERROR(EIO);
  351. }
  352. }
  353. }
  354. #endif
  355. if (XFS_FORCED_SHUTDOWN(mp))
  356. goto shutdown_abort;
  357. _xfs_trans_bjoin(tp, bp, 1);
  358. trace_xfs_trans_read_buf(bp->b_fspriv);
  359. *bpp = bp;
  360. return 0;
  361. shutdown_abort:
  362. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  363. xfs_buf_relse(bp);
  364. *bpp = NULL;
  365. return XFS_ERROR(EIO);
  366. }
  367. /*
  368. * Release the buffer bp which was previously acquired with one of the
  369. * xfs_trans_... buffer allocation routines if the buffer has not
  370. * been modified within this transaction. If the buffer is modified
  371. * within this transaction, do decrement the recursion count but do
  372. * not release the buffer even if the count goes to 0. If the buffer is not
  373. * modified within the transaction, decrement the recursion count and
  374. * release the buffer if the recursion count goes to 0.
  375. *
  376. * If the buffer is to be released and it was not modified before
  377. * this transaction began, then free the buf_log_item associated with it.
  378. *
  379. * If the transaction pointer is NULL, make this just a normal
  380. * brelse() call.
  381. */
  382. void
  383. xfs_trans_brelse(xfs_trans_t *tp,
  384. xfs_buf_t *bp)
  385. {
  386. xfs_buf_log_item_t *bip;
  387. /*
  388. * Default to a normal brelse() call if the tp is NULL.
  389. */
  390. if (tp == NULL) {
  391. ASSERT(bp->b_transp == NULL);
  392. xfs_buf_relse(bp);
  393. return;
  394. }
  395. ASSERT(bp->b_transp == tp);
  396. bip = bp->b_fspriv;
  397. ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
  398. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  399. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  400. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  401. trace_xfs_trans_brelse(bip);
  402. /*
  403. * If the release is just for a recursive lock,
  404. * then decrement the count and return.
  405. */
  406. if (bip->bli_recur > 0) {
  407. bip->bli_recur--;
  408. return;
  409. }
  410. /*
  411. * If the buffer is dirty within this transaction, we can't
  412. * release it until we commit.
  413. */
  414. if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
  415. return;
  416. /*
  417. * If the buffer has been invalidated, then we can't release
  418. * it until the transaction commits to disk unless it is re-dirtied
  419. * as part of this transaction. This prevents us from pulling
  420. * the item from the AIL before we should.
  421. */
  422. if (bip->bli_flags & XFS_BLI_STALE)
  423. return;
  424. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  425. /*
  426. * Free up the log item descriptor tracking the released item.
  427. */
  428. xfs_trans_del_item(&bip->bli_item);
  429. /*
  430. * Clear the hold flag in the buf log item if it is set.
  431. * We wouldn't want the next user of the buffer to
  432. * get confused.
  433. */
  434. if (bip->bli_flags & XFS_BLI_HOLD) {
  435. bip->bli_flags &= ~XFS_BLI_HOLD;
  436. }
  437. /*
  438. * Drop our reference to the buf log item.
  439. */
  440. atomic_dec(&bip->bli_refcount);
  441. /*
  442. * If the buf item is not tracking data in the log, then
  443. * we must free it before releasing the buffer back to the
  444. * free pool. Before releasing the buffer to the free pool,
  445. * clear the transaction pointer in b_fsprivate2 to dissolve
  446. * its relation to this transaction.
  447. */
  448. if (!xfs_buf_item_dirty(bip)) {
  449. /***
  450. ASSERT(bp->b_pincount == 0);
  451. ***/
  452. ASSERT(atomic_read(&bip->bli_refcount) == 0);
  453. ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
  454. ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
  455. xfs_buf_item_relse(bp);
  456. }
  457. bp->b_transp = NULL;
  458. xfs_buf_relse(bp);
  459. }
  460. /*
  461. * Mark the buffer as not needing to be unlocked when the buf item's
  462. * IOP_UNLOCK() routine is called. The buffer must already be locked
  463. * and associated with the given transaction.
  464. */
  465. /* ARGSUSED */
  466. void
  467. xfs_trans_bhold(xfs_trans_t *tp,
  468. xfs_buf_t *bp)
  469. {
  470. xfs_buf_log_item_t *bip = bp->b_fspriv;
  471. ASSERT(bp->b_transp == tp);
  472. ASSERT(bip != NULL);
  473. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  474. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  475. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  476. bip->bli_flags |= XFS_BLI_HOLD;
  477. trace_xfs_trans_bhold(bip);
  478. }
  479. /*
  480. * Cancel the previous buffer hold request made on this buffer
  481. * for this transaction.
  482. */
  483. void
  484. xfs_trans_bhold_release(xfs_trans_t *tp,
  485. xfs_buf_t *bp)
  486. {
  487. xfs_buf_log_item_t *bip = bp->b_fspriv;
  488. ASSERT(bp->b_transp == tp);
  489. ASSERT(bip != NULL);
  490. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  491. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  492. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  493. ASSERT(bip->bli_flags & XFS_BLI_HOLD);
  494. bip->bli_flags &= ~XFS_BLI_HOLD;
  495. trace_xfs_trans_bhold_release(bip);
  496. }
  497. /*
  498. * This is called to mark bytes first through last inclusive of the given
  499. * buffer as needing to be logged when the transaction is committed.
  500. * The buffer must already be associated with the given transaction.
  501. *
  502. * First and last are numbers relative to the beginning of this buffer,
  503. * so the first byte in the buffer is numbered 0 regardless of the
  504. * value of b_blkno.
  505. */
  506. void
  507. xfs_trans_log_buf(xfs_trans_t *tp,
  508. xfs_buf_t *bp,
  509. uint first,
  510. uint last)
  511. {
  512. xfs_buf_log_item_t *bip = bp->b_fspriv;
  513. ASSERT(bp->b_transp == tp);
  514. ASSERT(bip != NULL);
  515. ASSERT(first <= last && last < BBTOB(bp->b_length));
  516. ASSERT(bp->b_iodone == NULL ||
  517. bp->b_iodone == xfs_buf_iodone_callbacks);
  518. /*
  519. * Mark the buffer as needing to be written out eventually,
  520. * and set its iodone function to remove the buffer's buf log
  521. * item from the AIL and free it when the buffer is flushed
  522. * to disk. See xfs_buf_attach_iodone() for more details
  523. * on li_cb and xfs_buf_iodone_callbacks().
  524. * If we end up aborting this transaction, we trap this buffer
  525. * inside the b_bdstrat callback so that this won't get written to
  526. * disk.
  527. */
  528. XFS_BUF_DONE(bp);
  529. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  530. bp->b_iodone = xfs_buf_iodone_callbacks;
  531. bip->bli_item.li_cb = xfs_buf_iodone;
  532. trace_xfs_trans_log_buf(bip);
  533. /*
  534. * If we invalidated the buffer within this transaction, then
  535. * cancel the invalidation now that we're dirtying the buffer
  536. * again. There are no races with the code in xfs_buf_item_unpin(),
  537. * because we have a reference to the buffer this entire time.
  538. */
  539. if (bip->bli_flags & XFS_BLI_STALE) {
  540. bip->bli_flags &= ~XFS_BLI_STALE;
  541. ASSERT(XFS_BUF_ISSTALE(bp));
  542. XFS_BUF_UNSTALE(bp);
  543. bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
  544. }
  545. tp->t_flags |= XFS_TRANS_DIRTY;
  546. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  547. bip->bli_flags |= XFS_BLI_LOGGED;
  548. xfs_buf_item_log(bip, first, last);
  549. }
  550. /*
  551. * Invalidate a buffer that is being used within a transaction.
  552. *
  553. * Typically this is because the blocks in the buffer are being freed, so we
  554. * need to prevent it from being written out when we're done. Allowing it
  555. * to be written again might overwrite data in the free blocks if they are
  556. * reallocated to a file.
  557. *
  558. * We prevent the buffer from being written out by marking it stale. We can't
  559. * get rid of the buf log item at this point because the buffer may still be
  560. * pinned by another transaction. If that is the case, then we'll wait until
  561. * the buffer is committed to disk for the last time (we can tell by the ref
  562. * count) and free it in xfs_buf_item_unpin(). Until that happens we will
  563. * keep the buffer locked so that the buffer and buf log item are not reused.
  564. *
  565. * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
  566. * the buf item. This will be used at recovery time to determine that copies
  567. * of the buffer in the log before this should not be replayed.
  568. *
  569. * We mark the item descriptor and the transaction dirty so that we'll hold
  570. * the buffer until after the commit.
  571. *
  572. * Since we're invalidating the buffer, we also clear the state about which
  573. * parts of the buffer have been logged. We also clear the flag indicating
  574. * that this is an inode buffer since the data in the buffer will no longer
  575. * be valid.
  576. *
  577. * We set the stale bit in the buffer as well since we're getting rid of it.
  578. */
  579. void
  580. xfs_trans_binval(
  581. xfs_trans_t *tp,
  582. xfs_buf_t *bp)
  583. {
  584. xfs_buf_log_item_t *bip = bp->b_fspriv;
  585. ASSERT(bp->b_transp == tp);
  586. ASSERT(bip != NULL);
  587. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  588. trace_xfs_trans_binval(bip);
  589. if (bip->bli_flags & XFS_BLI_STALE) {
  590. /*
  591. * If the buffer is already invalidated, then
  592. * just return.
  593. */
  594. ASSERT(XFS_BUF_ISSTALE(bp));
  595. ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
  596. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
  597. ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
  598. ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
  599. ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
  600. return;
  601. }
  602. xfs_buf_stale(bp);
  603. bip->bli_flags |= XFS_BLI_STALE;
  604. bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
  605. bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
  606. bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
  607. memset((char *)(bip->bli_format.blf_data_map), 0,
  608. (bip->bli_format.blf_map_size * sizeof(uint)));
  609. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  610. tp->t_flags |= XFS_TRANS_DIRTY;
  611. }
  612. /*
  613. * This call is used to indicate that the buffer contains on-disk inodes which
  614. * must be handled specially during recovery. They require special handling
  615. * because only the di_next_unlinked from the inodes in the buffer should be
  616. * recovered. The rest of the data in the buffer is logged via the inodes
  617. * themselves.
  618. *
  619. * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
  620. * transferred to the buffer's log format structure so that we'll know what to
  621. * do at recovery time.
  622. */
  623. void
  624. xfs_trans_inode_buf(
  625. xfs_trans_t *tp,
  626. xfs_buf_t *bp)
  627. {
  628. xfs_buf_log_item_t *bip = bp->b_fspriv;
  629. ASSERT(bp->b_transp == tp);
  630. ASSERT(bip != NULL);
  631. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  632. bip->bli_flags |= XFS_BLI_INODE_BUF;
  633. }
  634. /*
  635. * This call is used to indicate that the buffer is going to
  636. * be staled and was an inode buffer. This means it gets
  637. * special processing during unpin - where any inodes
  638. * associated with the buffer should be removed from ail.
  639. * There is also special processing during recovery,
  640. * any replay of the inodes in the buffer needs to be
  641. * prevented as the buffer may have been reused.
  642. */
  643. void
  644. xfs_trans_stale_inode_buf(
  645. xfs_trans_t *tp,
  646. xfs_buf_t *bp)
  647. {
  648. xfs_buf_log_item_t *bip = bp->b_fspriv;
  649. ASSERT(bp->b_transp == tp);
  650. ASSERT(bip != NULL);
  651. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  652. bip->bli_flags |= XFS_BLI_STALE_INODE;
  653. bip->bli_item.li_cb = xfs_buf_iodone;
  654. }
  655. /*
  656. * Mark the buffer as being one which contains newly allocated
  657. * inodes. We need to make sure that even if this buffer is
  658. * relogged as an 'inode buf' we still recover all of the inode
  659. * images in the face of a crash. This works in coordination with
  660. * xfs_buf_item_committed() to ensure that the buffer remains in the
  661. * AIL at its original location even after it has been relogged.
  662. */
  663. /* ARGSUSED */
  664. void
  665. xfs_trans_inode_alloc_buf(
  666. xfs_trans_t *tp,
  667. xfs_buf_t *bp)
  668. {
  669. xfs_buf_log_item_t *bip = bp->b_fspriv;
  670. ASSERT(bp->b_transp == tp);
  671. ASSERT(bip != NULL);
  672. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  673. bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
  674. }
  675. /*
  676. * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
  677. * dquots. However, unlike in inode buffer recovery, dquot buffers get
  678. * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
  679. * The only thing that makes dquot buffers different from regular
  680. * buffers is that we must not replay dquot bufs when recovering
  681. * if a _corresponding_ quotaoff has happened. We also have to distinguish
  682. * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
  683. * can be turned off independently.
  684. */
  685. /* ARGSUSED */
  686. void
  687. xfs_trans_dquot_buf(
  688. xfs_trans_t *tp,
  689. xfs_buf_t *bp,
  690. uint type)
  691. {
  692. xfs_buf_log_item_t *bip = bp->b_fspriv;
  693. ASSERT(bp->b_transp == tp);
  694. ASSERT(bip != NULL);
  695. ASSERT(type == XFS_BLF_UDQUOT_BUF ||
  696. type == XFS_BLF_PDQUOT_BUF ||
  697. type == XFS_BLF_GDQUOT_BUF);
  698. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  699. bip->bli_format.blf_flags |= type;
  700. }